Fix the msgpack 2^32-1 element recovery test on 32-bit targets

On a 32-bit target (clang-cl Win32) 0xFFFFFFFF elements exceed the
array's max_size(), so the array is rejected before it is started and
the recovered value is null rather than an empty array. Expect that.

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

.github/CONTRIBUTING.md

@@ -111,6 +111,8 @@ context which existing file needs to be extended, and only very few cases requir
 When fixing a bug, edit `unit-regression3.cpp` and add a section referencing the fixed issue.
 `unit-regression2.cpp` holds the older tests; the two files exist because a single one grew large enough for the
 MinGW linker to fail relocating it, so please keep adding to the smaller file rather than growing the larger one.
+Tests that call `sax_parse` go into `unit-sax_parse.cpp` rather than into a large test file: every call instantiates
+the parser and the binary reader that recover from errors, which grows a test file considerably.
 
 #### Exceptions
 

.github/workflows/cancel_closed_pr_runs.yml

@@ -28,7 +28,7 @@ jobs:
 
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/check_amalgamation.yml

@@ -22,7 +22,7 @@ jobs:
     runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 
@@ -46,7 +46,7 @@ jobs:
 
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/check_docs_links.yml

@@ -20,7 +20,7 @@ jobs:
     timeout-minutes: 30
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/cifuzz.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-22.04
     steps:
     - name: Harden Runner
-      uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+      uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
       with:
         egress-policy: audit
 

.github/workflows/codeql-analysis.yml

@@ -27,7 +27,7 @@ jobs:
 
     steps:
     - name: Harden Runner
-      uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+      uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
       with:
         egress-policy: audit
 

.github/workflows/comment_check_amalgamation.yml

@@ -20,7 +20,7 @@ jobs:
       pull-requests: write
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/dependency-review.yml

@@ -21,7 +21,7 @@ jobs:
     runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/flawfinder.yml

@@ -31,7 +31,7 @@ jobs:
       security-events: write
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/json_view_benchmarks.yml

@@ -54,7 +54,7 @@ jobs:
     timeout-minutes: 60
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/labeler.yml

@@ -23,7 +23,7 @@ jobs:
 
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/publish_documentation.yml

@@ -37,7 +37,7 @@ jobs:
     runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/scorecards.yml

@@ -40,7 +40,7 @@ jobs:
 
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/semgrep.yml

@@ -36,7 +36,7 @@ jobs:
     runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/stale.yml

@@ -16,7 +16,7 @@ jobs:
 
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 

.github/workflows/ubuntu.yml

@@ -25,7 +25,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -38,7 +38,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Check that Meson and pkg-config offer the CMake options
         run: make check_build_options
       - name: Install Meson
@@ -80,7 +80,7 @@ jobs:
     runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 
@@ -92,7 +92,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -112,7 +112,7 @@ jobs:
         target: [ci_test_single_header, ci_cppcheck, ci_cpplint, ci_reproducible_tests, ci_non_git_tests, ci_offline_testdata, ci_reuse_compliance, ci_test_valgrind]
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 
@@ -123,7 +123,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -142,7 +142,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -168,7 +168,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -178,7 +178,7 @@ jobs:
     runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 
@@ -244,7 +244,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: CXX=g++-${{ matrix.compiler }} cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -262,7 +262,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -279,7 +279,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Disable filesystem support (std::filesystem is unusable) for clang 7 / 8 / 9 / 10
         run: echo "JSON_FORCED_GLOBAL_COMPILE_OPTIONS=-DJSON_HAS_FILESYSTEM=0;-DJSON_HAS_EXPERIMENTAL_FILESYSTEM=0" >> "$GITHUB_ENV"
         if: ${{ matrix.compiler == '7' || matrix.compiler == '8' || matrix.compiler == '9' || matrix.compiler == '10' }}
@@ -299,7 +299,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -319,7 +319,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build with libc++
@@ -346,7 +346,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -366,7 +366,7 @@ jobs:
       # import-std support. Its opt-in token is CMake-version-specific, so pin
       # CMake to the version whose token is set in tests/module_cpp20/CMakeLists.txt.
       - name: Get pinned CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
         with:
           cmakeVersion: 4.3.4
       # Clang: the std library module is provided by libc++ (the image's libstdc++
@@ -392,7 +392,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -409,7 +409,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -423,7 +423,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DJSON_CI=On
       - name: Build
@@ -433,7 +433,7 @@ jobs:
     runs-on: ubuntu-26.04
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 
@@ -443,7 +443,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Run CMake
         run: cmake -S . -B build -DCMAKE_TOOLCHAIN_FILE=$EMSDK/upstream/emscripten/cmake/Modules/Platform/Emscripten.cmake -GNinja
       - name: Build
@@ -456,7 +456,7 @@ jobs:
         target: [ci_test_examples, ci_test_build_documentation, ci_test_documentation_mermaid]
     steps:
       - name: Harden Runner
-        uses: step-security/harden-runner@dc2769abcc0fcfbadfc1df54081c1608579ef6b2 # v2.22.2
+        uses: step-security/harden-runner@46b06cea07f9e64621fdf066ba77ac6915163d43 # v2.22.3
         with:
           egress-policy: audit
 
@@ -484,7 +484,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Check that the job runs natively on arm64
         run: uname -m && test "$(uname -m)" = aarch64
       - name: Run CMake

.github/workflows/windows.yml

@@ -90,7 +90,7 @@ jobs:
         with:
           persist-credentials: false
       - name: Get latest CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
       - name: Set extra CXX_FLAGS for latest std_version
         # /wd5285 silences C5285 emitted by the bundled third-party doctest.h, which
         # specializes std::tuple (newly diagnosed by the VS2026 v145 toolset)
@@ -216,7 +216,7 @@ jobs:
       # import-std support. Its opt-in token is CMake-version-specific, so pin
       # CMake to the version whose token is set in tests/module_cpp20/CMakeLists.txt.
       - name: Get pinned CMake and ninja
-        uses: lukka/get-cmake@a89e43e22a96abafcd7a62db652a65ecd33052c9 # v4.4.4
+        uses: lukka/get-cmake@f86dd60acae0d132056913e248bdabf0e0580b7a # v4.4.5
         with:
           cmakeVersion: 4.3.4
       # CMAKE_CXX_FLAGS replaces CMake's MSVC defaults (which include /EHsc), and

README.md

@@ -495,7 +495,7 @@ bool key(string_t& val);
 bool parse_error(std::size_t position, const std::string& last_token, const detail::exception& ex);
 ```
 
-The return value of each function determines whether parsing should proceed.
+The return value of each function determines whether parsing should proceed. For `parse_error`, returning `true` [recovers from the error](https://json.nlohmann.me/features/parsing/error_recovery/): the parser repairs the input and continues.
 
 To implement your own SAX handler, proceed as follows:
 
@@ -503,7 +503,7 @@ To implement your own SAX handler, proceed as follows:
 2. Create an object of your SAX interface class, e.g. `my_sax`.
 3. Call `bool json::sax_parse(input, &my_sax)`; where the first parameter can be any input like a string or an input stream and the second parameter is a pointer to your SAX interface.
 
-Note the `sax_parse` function only returns a `bool` indicating the result of the last executed SAX event. It does not return a  `json` value - it is up to you to decide what to do with the SAX events. Furthermore, no exceptions are thrown in case of a parse error -- it is up to you what to do with the exception object passed to your `parse_error` implementation. Internally, the SAX interface is used for the DOM parser (class `json_sax_dom_parser`) as well as the acceptor (`json_sax_acceptor`), see file [`json_sax.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/json_sax.hpp).
+Note the `sax_parse` function only returns a `bool` indicating whether the input was parsed without errors and no SAX event returned `false`. It does not return a  `json` value - it is up to you to decide what to do with the SAX events. Furthermore, no exceptions are thrown in case of a parse error -- it is up to you what to do with the exception object passed to your `parse_error` implementation. Internally, the SAX interface is used for the DOM parser (class `json_sax_dom_parser`) as well as the acceptor (`json_sax_acceptor`), see file [`json_sax.hpp`](https://github.com/nlohmann/json/blob/develop/include/nlohmann/detail/input/json_sax.hpp).
 
 ### STL-like access
 

docs/mkdocs/docs/api/basic_json/patch_inplace.md

@@ -15,7 +15,9 @@ the patch. This function applies a JSON patch in place and returns void.
 
 ## Exception safety
 
-No guarantees, value may be corrupted by an unsuccessful patch operation.
+No guarantees: if an operation fails, the operations before it stay applied, and a failing `move` operation may already
+have removed the value at its "from" location. Use [`patch`](patch.md) if the value must be unchanged after a failed
+patch.
 
 ## Exceptions
 
@@ -62,8 +64,19 @@ the JSON value, the complexity can usually be neglected.
 ## Notes
 
 Unlike [`patch`](patch.md), `patch_inplace` applies the operation "in place" and no copy of the JSON value is created.
-That makes it faster for large documents by avoiding the copy. However, the JSON value might be corrupted if the
+That makes it faster for large documents by avoiding the copy.
-function throws an exception.
+
+!!! warning "Not atomic"
+
+    `patch_inplace` applies the operations one after the other and does not undo them when one fails. After an
+    exception, the value holds the result of the operations before the failing one, and a failing `move` operation
+    (a "remove" followed by an "add") may already have removed the value at its "from" location; see the second example
+    below. [RFC 6902](https://tools.ietf.org/html/rfc6902#section-5) requires that a patch with a failing operation is
+    not deemed successful, but leaves open what happens to the document.
+
+    If the value must stay unchanged when the patch fails, use [`patch`](patch.md), which applies the patch to a copy:
+    `#!cpp j = j.patch(json_patch);`. Validating the patch beforehand is not enough, as "test" operations and missing
+    paths are only detected while the patch is applied.
 
 Duplicate member names inside an operation object are resolved by the parser before the patch is applied (the last one
 is kept, see [`object_t`](object_t.md#behavior)), so `#!json {"op": "add", "op": "remove", "path": "/a"}` is a

docs/mkdocs/docs/api/basic_json/sax_parse.md

@@ -104,7 +104,9 @@ The SAX event lister must follow the interface of [`json_sax`](../json_sax/index
 
 ## Return value
 
-return value of the last processed SAX event
+`#!cpp true` if the input was parsed without errors and no SAX event returned `#!cpp false`; `#!cpp false` otherwise.
+In particular, the result is `#!cpp false` for input with errors, even if the SAX parser recovered from all of them
+(see [error recovery](../../features/parsing/error_recovery.md)).
 
 ## Exception safety
 
@@ -163,6 +165,7 @@ use the same file with stdio functions, but it must not wait for another thread
 - Added `ignore_trailing_commas` in version 3.13.0.
 - Added `tag_handler` in version 3.13.0.
 - Extended container support (1) to include types with lvalue-only ADL `begin`/`end` (matching `std::begin`/`std::end` semantics) in version 3.13.0.
+- Recovering from parse errors (see [`parse_error`](../json_sax/parse_error.md)) added in version 3.13.0.
 - Extended overload (2) to accept heterogeneous iterator+sentinel pairs (C++20 ranges support) in version 3.13.0.
 - `JSON_PRECISE_STREAM_POSITION` added in version 3.13.0 to optionally leave a `#!cpp std::istream` positioned right
   after the parsed value when `strict` is `#!cpp false`.

docs/mkdocs/docs/api/json_sax/index.md

@@ -7,7 +7,8 @@ struct json_sax;
 
 This class describes the SAX interface used by [sax_parse](../basic_json/sax_parse.md). Each function is called in
 different situations while the input is parsed. The boolean return value informs the parser whether to continue
-processing the input.
+processing the input; for [`parse_error`](parse_error.md), it decides whether to
+[recover from the error](../../features/parsing/error_recovery.md).
 
 For instance, parsing the JSON text `{"a": [1, true]}` triggers the following callbacks, in order:
 

docs/mkdocs/docs/api/json_sax/parse_error.md

@@ -21,11 +21,18 @@ A parse error occurred.
 
 ## Return value
 
-Whether parsing should proceed (**must return `#!cpp false`**).
+Whether to recover from the error:
+
+- `#!cpp false` stops parsing.
+- `#!cpp true` recovers from the error: the error is repaired and parsing continues. If that is not possible, which
+  happens in the binary formats when the end of the item with the error is unknown, the value read so far is completed
+  and parsing stops. See [error recovery](../../features/parsing/error_recovery.md) for how errors are repaired.
+
+Either way, [`sax_parse`](../basic_json/sax_parse.md) returns `#!cpp false`.
 
 ## Examples
 
-??? example
+??? example "Example: (1) the SAX interface"
 
     The example below shows how the SAX interface is used.
 
@@ -39,13 +46,30 @@ Whether parsing should proceed (**must return `#!cpp false`**).
     --8<-- "examples/sax_parse.output"
     ```
 
+??? example "Example: (2) recovering from errors"
+
+    The example below shows how a SAX parser recovers from errors.
+
+    ```cpp
+    --8<-- "examples/sax_parse__error_recovery.cpp"
+    ```
+    
+    Output:
+    
+    ```
+    --8<-- "examples/sax_parse__error_recovery.output"
+    ```
+
 ## See also
 
 - [`sax_parse`](../basic_json/sax_parse.md) - SAX parser
 - [Parsing and Exceptions](../../features/parsing/parse_exceptions.md) - the article on handling parse errors without
   exceptions
+- [Error Recovery](../../features/parsing/error_recovery.md) - the article on recovering from parse errors
 - [SAX Interface](../../features/parsing/sax_interface.md) - guide to event-based parsing
 
 ## Version history
 
 - Added in version 3.2.0.
+- Returning `#!cpp true` recovers from the error since version 3.13.0; before, parsing stopped, but the result of
+  [`sax_parse`](../basic_json/sax_parse.md) could be wrong.

docs/mkdocs/docs/api/macros/json_disable_unlocked_file_io.md

@@ -5,7 +5,9 @@
 ```
 
 When defined to `1`, parsing from a `#!cpp FILE*` reads every character with `std::fgetc`, as in version 3.12.0 and
-earlier, instead of locking the file once and reading it with the unlocked stdio functions.
+earlier, instead of locking the file once and reading it with the unlocked stdio functions. This also turns off the
+scan of the `FILE`'s buffer in bulk (see below), which [`JSON_DISABLE_STREAM_BULK_SCAN`](json_disable_stream_bulk_scan.md)
+does not affect.
 
 It does not change any result: values, exceptions, and the position of the file after parsing are the same either way.
 Only the speed and the locking of the file differ (see the notes on locking below).
@@ -37,9 +39,17 @@ The default value is `0`: the file is locked once and read with the unlocked std
     | Microsoft (MSVC, clang-cl)                                 | `_lock_file`, `_fgetc_nolock`, `_unlock_file`       |
     | others (e.g., musl, MinGW, newlib)                         | `std::fgetc`, as before                             |
 
-    This makes parsing from a `FILE*` 3 to 4 times as fast with the Apple C library and 1.1 to 1.3 times as fast with
+    With glibc and the Apple C library, the parser also scans the characters that the `FILE` already holds in its
-    glibc, whose locking is cheaper. The binary formats ([`from_cbor`](../basic_json/from_cbor.md),
+    buffer in bulk, as it scans the get area of a stream (see
-    [`from_msgpack`](../basic_json/from_msgpack.md), ...) read from a `FILE*` the same way.
+    [`JSON_DISABLE_STREAM_BULK_SCAN`](json_disable_stream_bulk_scan.md)): it reads the same fields of the `FILE` that
+    the C library's own `getc_unlocked` reads, and it consumes exactly the characters that `getc_unlocked` would have
+    returned, so the file is left where it was left before.
+
+    Locking the file once makes parsing from a `FILE*` 3 to 4 times as fast with the Apple C library and 1.1 to 1.3
+    times as fast with glibc, whose locking is cheaper. Scanning the buffer in bulk makes `parse()` another 1.7 to 2.6
+    times as fast (and `accept()` 2.9 to 4.4 times), about as fast as parsing from a `std::ifstream`. The binary
+    formats ([`from_cbor`](../basic_json/from_cbor.md), [`from_msgpack`](../basic_json/from_msgpack.md), ...) read
+    from a `FILE*` the same way.
 
 !!! note "Locking"
 

docs/mkdocs/docs/examples/sax_parse__error_recovery.cpp

@@ -0,0 +1,43 @@
+#include <iostream>
+#include <iomanip>
+#include <nlohmann/json.hpp>
+
+using json = nlohmann::json;
+
+// a SAX parser that creates a JSON value like json::parse does, but that
+// recovers from parse errors instead of stopping at the first one
+class recovering_parser : public nlohmann::detail::json_sax_dom_parser<json>
+{
+  public:
+    explicit recovering_parser(json& result)
+        : nlohmann::detail::json_sax_dom_parser<json>(result, false)
+    {}
+
+    bool parse_error(std::size_t position,
+                     const std::string& /*last_token*/,
+                     const json::exception& ex)
+    {
+        std::cout << "byte " << position << ": " << ex.what() << '\n';
+
+        // repair the input and continue
+        return true;
+    }
+};
+
+int main()
+{
+    // JSON text with several mistakes that ends too early
+    const std::string text = R"({
+    "name": "Hello World",
+    "tags": ["a" "b",],
+    "valid": tru,
+    "size": 1.,
+    "nested": {"x": 1)";
+
+    json result;
+    recovering_parser sax(result);
+    const bool valid = json::sax_parse(text, &sax);
+
+    std::cout << "\nvalid JSON: " << std::boolalpha << valid << '\n'
+              << std::setw(4) << result << std::endl;
+}

docs/mkdocs/docs/examples/sax_parse__error_recovery.output

@@ -0,0 +1,19 @@
+byte 49: [json.exception.parse_error.101] parse error at line 3, column 20: syntax error while parsing array - unexpected string literal; expected ']'
+byte 51: [json.exception.parse_error.101] parse error at line 3, column 22: syntax error while parsing value - unexpected ']'; expected '[', '{', or a literal
+byte 70: [json.exception.parse_error.101] parse error at line 4, column 17: syntax error while parsing value - invalid literal; last read: '"valid": tru,'
+byte 86: [json.exception.parse_error.101] parse error at line 5, column 15: syntax error while parsing value - invalid number; expected digit after '.'; last read: '1.,'
+byte 109: [json.exception.parse_error.101] parse error at line 6, column 22: syntax error while parsing object - unexpected end of input; expected '}'
+
+valid JSON: false
+{
+    "name": "Hello World",
+    "nested": {
+        "x": 1
+    },
+    "size": 1,
+    "tags": [
+        "a",
+        "b"
+    ],
+    "valid": null
+}

docs/mkdocs/docs/features/files.md

@@ -0,0 +1,130 @@
+# Reading and writing files
+
+The library has no functions that take a file name. It reads from and writes to streams, so files are handled with
+`#!cpp std::ifstream` and `#!cpp std::ofstream` (or a `#!cpp FILE*` when reading). This page collects the recipes for
+the common cases and the mistakes that show up again and again.
+
+## Reading a JSON file
+
+Pass the stream to [`parse`](../api/basic_json/parse.md) directly, and check that the file was opened:
+
+```cpp
+#include <fstream>
+#include <stdexcept>
+#include <string>
+#include <nlohmann/json.hpp>
+
+using json = nlohmann::json;
+
+json read_json_file(const std::string& path)
+{
+    std::ifstream f(path);
+    if (!f.is_open())
+    {
+        throw std::runtime_error("cannot open " + path);
+    }
+    return json::parse(f);
+}
+```
+
+- **Check that the file is open.** A stream that failed to open looks like an empty input, so `parse` reports
+  [`parse_error.101`](../home/exceptions.md#jsonexceptionparse_error101) ("unexpected end of input") at line 1,
+  column 1 instead of saying that the file is missing. A relative path is resolved against the working directory of the
+  running program, which is often not the directory of the source file (see the
+  [FAQ](../home/faq.md#parsing-a-file-fails-with-unexpected-end-of-input)).
+- **Don't read the file into a string first.** Parsing from a `#!cpp std::ifstream` is only about 10 to 20 % slower
+  than reading the whole file into a string with one `read()` and parsing that, and it needs no memory for the copy (see
+  [input sources and performance](parsing/index.md#input-sources-and-performance)). The common way to read a file with
+  `#!cpp std::istreambuf_iterator<char>` is so slow that it takes about twice as long as parsing from the stream. Read
+  the file into a string only if you need the text itself, for example for an error message. This advice is for
+  `#!cpp json::parse`. [`json_document`](json_view.md) reads a stream into a buffer that it owns anyway, so there is no
+  copy to save: reading the file into a `#!cpp std::string` yourself and passing that as an lvalue, which the document
+  borrows, was about 6 to 12 % faster than parsing from the stream, with the same single buffer.
+- `parse` requires the file to hold exactly one JSON value. Files with one value per line are read with
+  [JSON Lines](parsing/json_lines.md); files too large for memory are covered in
+  [Large, streamed, and partial input](parsing/large_and_partial_input.md).
+
+## Writing a JSON file
+
+[`dump`](../api/basic_json/dump.md) the value into a string first, then write the string:
+
+```cpp
+void write_json_file(const std::string& path, const json& j)
+{
+    const std::string text = j.dump(4) + '\n';  // may throw: nothing is written yet
+
+    std::ofstream f(path, std::ios::binary);
+    if (!f.is_open())
+    {
+        throw std::runtime_error("cannot open " + path);
+    }
+    f << text;
+    f.close();
+    if (f.fail())
+    {
+        throw std::runtime_error("cannot write " + path);
+    }
+}
+```
+
+- **Serialize before opening the file.** Opening a `#!cpp std::ofstream` truncates the file. `dump` and
+  [`operator<<`](../api/operator_ltlt.md) throw [`type_error.316`](../home/exceptions.md#jsonexceptiontype_error316)
+  when a string is not valid UTF-8 (see [handling invalid UTF-8](serialization.md#handling-invalid-utf-8)), and
+  `#!cpp f << std::setw(4) << j` may already have written part of the value when it throws. Serializing into a string
+  first leaves the old file untouched in that case.
+- **Check the stream after closing it.** A full disk or a missing permission sets the stream's error state; it does not
+  throw unless you enabled exceptions on the stream with `exceptions()`.
+- `std::ios::binary` writes the text as it is. Without it, Windows writes `\r\n` for every newline of a pretty-printed
+  value.
+
+## Replacing a file without corrupting it
+
+The library never writes a file itself, so this section is not about the library. It is general C++ advice for a
+problem users of the library keep running into.
+
+A program that crashes or is killed while it writes leaves a truncated file behind. To make sure readers see either the
+old or the new contents, write to a temporary file in the same directory and rename it over the original
+(C++17, `#!cpp <filesystem>`):
+
+```cpp
+#include <filesystem>
+
+void replace_json_file(const std::filesystem::path& path, const json& j)
+{
+    std::filesystem::path tmp = path;
+    tmp += ".tmp";  // use a unique name if several processes may write the file
+
+    write_json_file(tmp.string(), j);
+    std::filesystem::rename(tmp, path);  // replaces path
+}
+```
+
+The temporary file must be in the same directory (more precisely, on the same file system) as the target: a rename
+across file systems does not fall back to a copy, it fails and throws `#!cpp std::filesystem::filesystem_error`. On
+POSIX systems, the rename atomically replaces the file. Standard C++ has no way to flush a file to the disk (`fsync`),
+so this protects against a crashed program, but not necessarily against a power loss.
+
+## Binary formats
+
+The [binary formats](binary_formats/index.md) work the same way, with the streams opened in binary mode:
+
+```cpp
+// read
+std::ifstream in("data.cbor", std::ios::binary);
+json j = json::from_cbor(in);
+
+// write
+const std::vector<std::uint8_t> bytes = json::to_cbor(j);
+std::ofstream out("data.cbor", std::ios::binary);
+out.write(reinterpret_cast<const char*>(bytes.data()), static_cast<std::streamsize>(bytes.size()));
+```
+
+Without `std::ios::binary`, Windows turns the bytes `0x0D 0x0A` into `0x0A` and may stop at a byte `0x1A` while reading,
+and turns `0x0A` into `0x0D 0x0A` while writing, which corrupts the data.
+
+## See also
+
+- [Parsing](parsing/index.md) - the inputs `parse` accepts
+- [Serialization](serialization.md) - the options of `dump`
+- [`parse`](../api/basic_json/parse.md), [`dump`](../api/basic_json/dump.md)
+- [`operator>>`](../api/operator_gtgt.md), [`operator<<`](../api/operator_ltlt.md)

docs/mkdocs/docs/features/index.md

@@ -46,6 +46,8 @@ C++ types, and finally serialize it again.
   [BJData](binary_formats/bjdata.md), [BON8](binary_formats/bon8.md), [BSON](binary_formats/bson.md),
   [CBOR](binary_formats/cbor.md), [MessagePack](binary_formats/messagepack.md), or [UBJSON](binary_formats/ubjson.md).
 - [Binary values](binary_values.md) — store and exchange raw byte sequences.
+- [Reading and writing files](files.md) — read and write files safely, including replacing a file without
+  corrupting it.
 
 ## How values are stored and configured
 

docs/mkdocs/docs/features/json_patch.md

@@ -29,7 +29,8 @@ other three continue on its result.
 
     [`patch`](../api/basic_json/patch.md) leaves the original value unchanged and returns the patched result as a copy.
     If the document is large and the original value is no longer needed,
-    [`patch_inplace`](../api/basic_json/patch_inplace.md) applies the same operations in place instead.
+    [`patch_inplace`](../api/basic_json/patch_inplace.md) applies the same operations in place instead. Unlike
+    `patch`, it is not atomic: if an operation fails, the operations before it stay applied.
 
 !!! note "Duplicate member names"
 

docs/mkdocs/docs/features/json_view.md

@@ -160,7 +160,9 @@ document: `#!cpp auto v = json_document::parse(text).root();` does not compile.
   wide. The limit is 16 bytes below 2^32, because the parser reads in 16-byte steps, and a position one step past the
   end must still fit.
 - **A stream is always read to its end.** There is no partial/streaming read of an `#!cpp std::istream`: the stream
-  is read to its end into a buffer the document owns, and parsing starts after that.
+  is read to its end into a buffer the document owns, and parsing starts after that. For a file, reading it into a
+  `#!cpp std::string` yourself and parsing that (an lvalue is borrowed) is therefore not more expensive in memory, and
+  was about 6 to 12 % faster than parsing from an `#!cpp std::ifstream`.
 - **Fewer source positions after edits.** With
   [`JSON_DIAGNOSTIC_POSITIONS`](../api/macros/json_diagnostic_positions.md) enabled,
   [`materialize()`](../api/basic_json_view/materialize.md) sets the same positions as `#!cpp basic_json::parse()`.

docs/mkdocs/docs/features/parsing/error_recovery.md

@@ -0,0 +1,122 @@
+# Error Recovery
+
+By default, parsing stops at the first error. With the [SAX interface](sax_interface.md), you can instead ask the
+parser to *recover*: to repair the error and continue, so that you get as much as possible out of malformed input, for
+instance a file that was cut off, JSON edited by hand, or the output of a language model.
+
+## Recovering from errors
+
+The SAX parser's [`parse_error`](../../api/json_sax/parse_error.md) function is called for every error. Its return value
+decides what happens next:
+
+- `#!cpp false` stops parsing. This is what the SAX parsers of the library do, so [`parse`](../../api/basic_json/parse.md)
+  and [`accept`](../../api/basic_json/accept.md) never recover.
+- `#!cpp true` repairs the error and continues parsing.
+
+When recovering, the SAX parser still receives well-formed events: every `start_object` or `start_array` is followed by
+the matching `end_object` or `end_array`, and every `key` is followed by exactly one value. A SAX parser that creates a
+JSON value, such as the one in the example below, therefore gets a complete value. Parsing always ends, and
+[`sax_parse`](../../api/basic_json/sax_parse.md) returns `#!cpp false` for input that is not valid JSON, even if every
+error was repaired. Each token is reported at most once, and the SAX parser can stop at any error by returning
+`#!cpp false`.
+
+!!! example
+
+    The example below derives a SAX parser from the library's parser for `json` values (`json_sax_dom_parser`),
+    and recovers from all errors.
+
+    ```cpp
+    --8<-- "examples/sax_parse__error_recovery.cpp"
+    ```
+    
+    Output:
+    
+    ```
+    --8<-- "examples/sax_parse__error_recovery.output"
+    ```
+
+## How errors are repaired
+
+Each error is repaired with the smallest local edit: a missing separator is inserted, a stray token is removed, what can
+be read of a broken string or number is kept, and a value that cannot be read at all becomes `#!json null`.
+
+| Mistake                   | Repair                                                                         | Example                                  | Result                     |
+|---------------------------|--------------------------------------------------------------------------------|------------------------------------------|----------------------------|
+| missing `,` or `:`        | inserted                                                                       | `#!json [1 2]`, `#!json {"a" 1}`         | `[1,2]`, `{"a":1}`         |
+| missing value             | `#!json null` for an object key or between commas in an array                  | `#!json {"a":}`, `#!json [1,,2]`         | `{"a":null}`, `[1,null,2]` |
+| trailing comma            | removed                                                                        | `#!json [1,2,]`                          | `[1,2]`                    |
+| broken string             | invalid escapes and bytes are replaced (see below); a line break ends the string | `#!json ["a\qb"]`                        | `["aqb"]`                  |
+| broken number             | the longest valid beginning is kept                                            | `#!json [1., 2e+]`                       | `[1,2]`                    |
+| unreadable value          | `#!json null`                                                                  | `#!json [1, NaN, tru]`                   | `[1,null,null]`            |
+| number too large          | passed as infinity, together with its text                                     | `#!json [1e999]`                         | infinity (see below)       |
+| stray `:`                 | removed                                                                        | `#!json ["a":1]`                         | `["a",1]`                  |
+| member without a key      | skipped up to the next `,` or `}`                                              | `#!json {1:2, "b":3}`                    | `{"b":3}`                  |
+| wrong closing bracket     | closes the innermost array or object                                           | `#!json {"a":[1,2}, "b":3}`              | `{"a":[1,2],"b":3}`        |
+| input ends too early      | all open arrays and objects are closed                                         | `#!json {"a":[1,2`                       | `{"a":[1,2]}`              |
+| text before the value     | skipped                                                                        | `#!json )]}'{"a":1}`                     | `{"a":1}`                  |
+
+In a string, an unknown escape like `\q` stands for the escaped character (`q`), as in JavaScript. An invalid `\u`
+escape, a lone surrogate, and ill-formed UTF-8 are each replaced by U+FFFD (REPLACEMENT CHARACTER), and control
+characters are kept. A string without its closing quote ends at the next line break or at the end of the input.
+
+The input after the top-level value is not repaired: as without recovery, it is reported as an error, and parsing stops.
+
+## Binary formats
+
+The binary formats ([BJData](../binary_formats/bjdata.md), [BON8](../binary_formats/bon8.md),
+[BSON](../binary_formats/bson.md), [CBOR](../binary_formats/cbor.md), [MessagePack](../binary_formats/messagepack.md),
+and [UBJSON](../binary_formats/ubjson.md)) have no delimiters to find the next value by. So what can be repaired depends
+on whether the end of the item with the error is known, a distinction that
+[RFC 8949, Section 5.3](https://www.rfc-editor.org/rfc/rfc8949.html#section-5.3) makes for CBOR, too.
+
+If the item is complete, but cannot be passed on as it is, it is replaced, and parsing continues after it:
+
+| Mistake                                                             | Formats                 | Repair                                                                  |
+|---------------------------------------------------------------------|-------------------------|-------------------------------------------------------------------------|
+| tag, if `tag_handler` is `cbor_tag_handler_t::error` (the default)  | CBOR                    | ignored                                                                 |
+| simple value other than `false`, `true`, and `null`, like undefined | CBOR                    | `#!json null`                                                           |
+| number too large for a custom `number_float_t`, like `float`        | all                     | infinity                                                                |
+| character (`C`) that is not ASCII                                   | BJData, UBJSON          | U+FFFD                                                                  |
+| invalid high-precision number (`H`)                                 | BJData, UBJSON          | the longest valid beginning is kept, as for JSON text, or `#!json null` |
+| high-precision number too large                                     | BJData, UBJSON          | passed as infinity, together with its text                              |
+| object key that is not a string                                     | BON8, CBOR, MessagePack | the member is skipped                                                   |
+| element of a type the library does not read, like ObjectId or date  | BSON                    | `#!json null`                                                           |
+| string without its terminator                                       | BSON                    | kept                                                                    |
+| document whose size does not match its content                      | BSON                    | kept                                                                    |
+
+CBOR tags and simple values are repaired as [RFC 8949, Section 6.1](https://www.rfc-editor.org/rfc/rfc8949.html#section-6.1)
+suggests for converting CBOR to JSON. By default, [`sax_parse`](../../api/basic_json/sax_parse.md) reports every tag as
+an error; when recovering, tags are then ignored like with
+[`cbor_tag_handler_t::ignore`](../../api/basic_json/cbor_tag_handler_t.md). Strings that are not valid UTF-8 are no
+error: like [`from_cbor`](../../api/basic_json/from_cbor.md) and the other functions by default, `sax_parse` passes
+them on as they are.
+
+After any other error, the end of the item is unknown: the input ended, a byte is not a valid type marker, or a size
+cannot be right. Parsing then stops, and the value read so far is completed: a key that waits for its value gets
+`#!json null`, and all open arrays and objects are closed. This keeps everything before the error of an input that was
+cut off. The exception is BSON, which stores the size of every document: an element whose end is unknown gets
+`#!json null`, the rest of its document is skipped, and parsing continues after the document.
+
+## Limitations
+
+- A repair is a guess. For example, `#!json {"a" "b": 1}` could be meant as `#!json {"a": "b"}` or as
+  `#!json {"a": null, "b": 1}`; it is repaired to the former. Treat recovered values as a best effort, and check the
+  reported errors.
+- A closing bracket always closes the innermost array or object. If a bracket is missing rather than wrong, the
+  repair differs from the intention: `#!json {"a": {"b": [1, 2}, "c": 3}` is repaired to
+  `#!json {"a": {"b": [1, 2], "c": 3}}`, although `#!json {"a": {"b": [1, 2]}, "c": 3}` may have been meant.
+- Keys without quotes, and strings in single quotes, are not supported; such members are skipped.
+- In the binary formats, a member that is skipped because its key is not a string is lost, and so are the elements of a
+  BSON document after one whose end is unknown.
+- A number that is too large for `number_float_t` is passed as positive or negative infinity. The SAX parser's
+  `number_float` also gets the number's text, but a JSON value cannot store it, and
+  [`dump`](../../api/basic_json/dump.md) serializes infinity as `#!json null`.
+- When parsing is not strict (see [`sax_parse`](../../api/basic_json/sax_parse.md)), a repair may read parts of the
+  input after the value, for instance of the next value in a stream of concatenated values.
+
+## See also
+
+- [SAX interface](sax_interface.md) - implement a custom SAX handler
+- [`parse_error`](../../api/json_sax/parse_error.md) - the SAX event for parse errors
+- [`sax_parse`](../../api/basic_json/sax_parse.md) - generate SAX events
+- [parsing and exceptions](parse_exceptions.md) - control error handling

docs/mkdocs/docs/features/parsing/index.md

@@ -35,6 +35,8 @@ std::ifstream f("example.json");
 json data = json::parse(f);
 ```
 
+See [Reading and writing files](../files.md) for checking that the file was opened and for writing files safely.
+
 The input must be encoded in UTF-8; other encodings are not supported. The exception is wide-character input, which
 is read as UTF-16 or UTF-32 — depending on the character type — and converted to UTF-8; see the
 [`parse`](../../api/basic_json/parse.md) page for an example. A single input may contain only one JSON value.
@@ -75,7 +77,7 @@ them faster than others:
 |--------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------|
 | strings, string views, string literals, contiguous containers, pointer or iterator ranges over contiguous memory | the whole input is scanned in bulk: runs of string characters, numbers, and indentation are recognized several bytes at a time |
 | `#!cpp std::istream` (`#!cpp std::ifstream`, `#!cpp std::istringstream`, ...)            | the characters the stream buffer holds are scanned in bulk like a string; the stream buffer is asked for more when they are used up |
-| `#!cpp FILE*`                                                                              | one character at a time, with the file locked once for the whole parse              |
+| `#!cpp FILE*`                                                                              | with glibc and the Apple C library, the characters the `FILE` holds in its buffer are scanned in bulk like a stream's; elsewhere one character at a time; the file is locked once for the whole parse |
 | other iterator ranges (e.g., `#!cpp std::list<char>`, `#!cpp std::istreambuf_iterator<char>`), wide strings | one character at a time                                                             |
 
 Parsing a file from a `#!cpp std::ifstream` is therefore only about 10 to 20 % slower than reading the whole file into a
@@ -84,6 +86,9 @@ ahead: after parsing, it is positioned exactly as if every character had been re
 [`operator>>`](../../api/operator_gtgt.md) can still read several values from one stream (see also
 [JSON Lines](json_lines.md)).
 
+This does not hold for [`json_document`](../json_view.md): it reads a stream into a buffer it owns, so the stream saves
+no memory, and reading the file into a `#!cpp std::string` yourself and borrowing it is about 6 to 12 % faster.
+
 Two macros switch these optimizations off without changing any result:
 [`JSON_DISABLE_STREAM_BULK_SCAN`](../../api/macros/json_disable_stream_bulk_scan.md) for stream buffers that do not
 follow the usual `std::streambuf` contract, and
@@ -141,7 +146,7 @@ See [Zero-copy JSON views](../json_view.md) for the concept and the trade-offs.
 When the input is not valid JSON, the `parse` function throws an exception by default. If exceptions are undesired or
 unavailable, the parser can instead return a discarded value, or [`accept`](../../api/basic_json/accept.md) can be used
 to only check whether an input is valid JSON. See [parsing and exceptions](parse_exceptions.md) for the available
-options.
+options. To get as much as possible out of malformed input, a SAX parser can [recover from errors](error_recovery.md).
 
 ## See also
 
@@ -155,6 +160,7 @@ options.
 - [SAX interface](sax_interface.md) - implement a custom SAX handler
 - [Zero-copy JSON views](../json_view.md) - parse into a read-only document and read it through views
 - [parsing and exceptions](parse_exceptions.md) - control error handling
+- [error recovery](error_recovery.md) - get as much as possible out of malformed input
 - [parsing untrusted input](untrusted_input.md) - what to consider when parsing input from untrusted sources
 - [`operator>>`](../../api/operator_gtgt.md) - deserialize from stream
 - [`json_sax`](../../api/json_sax/index.md) - documentation of the SAX interface

docs/mkdocs/docs/features/parsing/parse_exceptions.md

@@ -68,7 +68,8 @@ bool parse_error(std::size_t position,
                  const json::exception& ex);
 ```
 
-The return value indicates whether the parsing should continue, so the function should usually return `#!cpp false`.
+The return value decides whether to stop parsing (`#!cpp false`) or to repair the error and continue
+(`#!cpp true`); see [error recovery](error_recovery.md) for the latter.
 
 ??? example "Example: report parse errors without exceptions"
 

docs/mkdocs/docs/features/parsing/sax_interface.md

@@ -62,7 +62,8 @@ bool key(string_t& val);
 bool parse_error(std::size_t position, const std::string& last_token, const json::exception& ex);
 ```
 
-The return value of each function determines whether parsing should proceed.
+The return value of each function determines whether parsing should proceed. For `parse_error`, returning
+`#!cpp true` [recovers from the error](error_recovery.md).
 
 Object keys are passed to `key` decoded and validated: escape sequences are already resolved, so different spellings of
 the same name yield the same string. See [`key`](../../api/json_sax/key.md#notes) for the exact guarantees.
@@ -101,7 +102,7 @@ To implement your own SAX handler, proceed as follows:
 2. Create an object of your SAX interface class, e.g. `my_sax`.
 3. Call `#!cpp bool json::sax_parse(input, &my_sax);` where the first parameter can be any input like a string or an input stream and the second parameter is a pointer to your SAX interface.
 
-Note the `sax_parse` function only returns a `#!cpp bool` indicating the result of the last executed SAX event. It does not return `json` value - it is up to you to decide what to do with the SAX events. Furthermore, no exceptions are thrown in case of a parse error - it is up to you what to do with the exception object passed to your `parse_error` implementation. Internally, the SAX interface is used for the DOM parser (class `json_sax_dom_parser`) as well as the acceptor (`json_sax_acceptor`), see file `json_sax.hpp`.
+Note the `sax_parse` function only returns a `#!cpp bool` indicating whether the input was parsed without errors and no SAX event returned `#!cpp false`. It does not return `json` value - it is up to you to decide what to do with the SAX events. Furthermore, no exceptions are thrown in case of a parse error - it is up to you what to do with the exception object passed to your `parse_error` implementation. Internally, the SAX interface is used for the DOM parser (class `json_sax_dom_parser`) as well as the acceptor (`json_sax_acceptor`), see file `json_sax.hpp`.
 
 ## See also
 

docs/mkdocs/docs/features/performance.md

@@ -27,8 +27,8 @@ A `#!cpp std::istream` (including `#!cpp std::ifstream`) gets the same bulk scan
 characters its stream buffer already holds (the *get area*) and asks the stream buffer for more only when they are used
 up. Nothing is read ahead, so the stream is left exactly where it would be after reading one character at a time, and
 [`operator>>`](../api/operator_gtgt.md) can still read several values from one stream. A stream buffer without a get
-area is read one character at a time. A `#!cpp FILE*` is still read one character at a time, but `file_input_adapter`
+area is read one character at a time. A `#!cpp FILE*` is locked once per parse instead of once per character, and with
-locks the file once per parse instead of once per character. See
+glibc and the Apple C library, the characters in its buffer are scanned in bulk as a stream's get area is. See
 [input sources and performance](parsing/index.md#input-sources-and-performance) for the full table. According to the
 description of [#5828](https://github.com/nlohmann/json/pull/5828), parsing from a stream is within about 10 to 20 % of
 parsing an equivalent `#!cpp std::string`.
@@ -39,7 +39,9 @@ iterator pair over non-contiguous but random-access storage (e.g. `#!cpp std::de
 optimization but not the second, since the byte-scanning fast path requires contiguous storage.
 
 Practically: a file can be passed directly to `parse`, which also avoids the memory for a copy of the file. Reading it
-into a string first is no longer necessary for speed:
+into a string first is no longer necessary for speed (this is for `#!cpp json::parse`; a
+[`json_document`](json_view.md) reads a stream into an owned buffer anyway, so for it reading the file into a string
+and borrowing that is the faster way):
 
 ```cpp
 // fast: the stream buffer's get area is scanned in bulk

docs/mkdocs/docs/features/serialization.md

@@ -17,6 +17,9 @@ To write a value directly to a stream (for example, a file or `#!cpp std::cout`)
 std::cout << j << std::endl;
 ```
 
+To write a file without leaving a truncated file behind when serialization fails, see
+[Reading and writing files](files.md#writing-a-json-file).
+
 !!! note "String, not raw value"
 
     `dump` always returns a **JSON text**. Serializing a JSON string therefore includes the surrounding quotes and

docs/mkdocs/docs/home/faq.md

@@ -110,7 +110,7 @@ in the source files that need it, instantiate `basic_json` once, and consider pr
 The parser did not see any character: the file was most likely not opened. A relative path is resolved against the
 working directory of the running program, which is often the build directory of your IDE, not the source directory.
 Check `is_open()` and try an absolute path. Other causes are an empty file and a stream that was already read. See
-[`parse_error.101`](exceptions.md#jsonexceptionparse_error101).
+[`parse_error.101`](exceptions.md#jsonexceptionparse_error101) and [Reading and writing files](../features/files.md).
 
 ### Input that arrives in chunks
 

docs/mkdocs/docs/images/compile_times.svg

@@ -1,21 +1,21 @@
-<svg class="nj-fig nj-charts nj-charts--compile" viewBox="0 0 470 516" role="img" aria-labelledby="nj-charts-ct nj-charts-ct-d" xmlns="http://www.w3.org/2000/svg">
+<svg class="nj-fig nj-charts nj-charts--compile" viewBox="0 0 470 620" role="img" aria-labelledby="nj-charts-ct nj-charts-ct-d" xmlns="http://www.w3.org/2000/svg">
 <title id="nj-charts-ct">Change in compile time against the default configuration at -O0, Apple clang and G&#8203;CC 16</title>
-<desc id="nj-charts-ct-d">Grouped horizontal bars, negative is faster. json_fwd.hpp instead of json.hpp on a header-only translation unit: Apple clang -62 %, G&#8203;CC 16 -82 %; extern template on a parsing translation unit: Apple clang -7 %, G&#8203;CC 16 -21 %; JSON_NO_AUTOMATIC_UDLS on a model translation unit: Apple clang -11 %, G&#8203;CC 16 -11 %; JSON_NO_AUTOMATIC_UDLS on a parsing translation unit: Apple clang -1 %, G&#8203;CC 16 -2 %; JSON_HAS_FILESYSTEM=0 + JSON_NO_IO on a model translation unit: Apple clang -8 %, G&#8203;CC 16 -11 %; JSON_HAS_FILESYSTEM=0 on a model translation unit: Apple clang -5 %, G&#8203;CC 16 -9 %; JSON_NO_IO on a model translation unit: Apple clang -2 %, G&#8203;CC 16 -2 %; JSON_USE_GLOBAL_UDLS=0 on a model translation unit: Apple clang 0 %, G&#8203;CC 16 -1 %.</desc>
+<desc id="nj-charts-ct-d">Grouped horizontal bars, negative is faster. json_fwd.hpp instead of json.hpp on a header-only translation unit: Apple clang -62 %, G&#8203;CC 16 -82 %; extern template on a parsing translation unit: Apple clang -7 %, G&#8203;CC 16 -21 %; JSON_NO_AUTOMATIC_UDLS on a model translation unit: Apple clang -11 %, G&#8203;CC 16 -11 %; JSON_NO_AUTOMATIC_UDLS on a parsing translation unit: Apple clang -1 %, G&#8203;CC 16 -2 %; JSON_HAS_FILESYSTEM=0 + JSON_NO_IO on a model translation unit: Apple clang -8 %, G&#8203;CC 16 -11 %; JSON_HAS_FILESYSTEM=0 on a model translation unit: Apple clang -5 %, G&#8203;CC 16 -9 %; JSON_HAS_STD_FORMAT=0 on a model translation unit: Apple clang -1 %, G&#8203;CC 16 -4 %; JSON_HAS_RANGES=0 on a model translation unit: Apple clang -4 %, G&#8203;CC 16 -1 %; JSON_NO_IO on a model translation unit: Apple clang -2 %, G&#8203;CC 16 -2 %; JSON_USE_GLOBAL_UDLS=0 on a model translation unit: Apple clang 0 %, G&#8203;CC 16 -1 %.</desc>
 <rect class="nj-charts-clang" x="8" y="10" width="12" height="12" rx="3"/>
 <text class="nj-charts-leg" x="26" y="20.5">Apple clang</text>
 <rect class="nj-charts-gcc" x="126" y="10" width="12" height="12" rx="3"/>
 <text class="nj-charts-leg" x="144" y="20.5">G&#8203;CC 16</text>
-<line class="nj-charts-grid" x1="47.4" y1="58" x2="47.4" y2="472"/>
+<line class="nj-charts-grid" x1="47.4" y1="58" x2="47.4" y2="576"/>
-<text class="nj-charts-tick" x="47.4" y="488">&#8722;80 %</text>
+<text class="nj-charts-tick" x="47.4" y="592">&#8722;80 %</text>
-<line class="nj-charts-grid" x1="126.2" y1="58" x2="126.2" y2="472"/>
+<line class="nj-charts-grid" x1="126.2" y1="58" x2="126.2" y2="576"/>
-<text class="nj-charts-tick" x="126.2" y="488">&#8722;60 %</text>
+<text class="nj-charts-tick" x="126.2" y="592">&#8722;60 %</text>
-<line class="nj-charts-grid" x1="205" y1="58" x2="205" y2="472"/>
+<line class="nj-charts-grid" x1="205" y1="58" x2="205" y2="576"/>
-<text class="nj-charts-tick" x="205" y="488">&#8722;40 %</text>
+<text class="nj-charts-tick" x="205" y="592">&#8722;40 %</text>
-<line class="nj-charts-grid" x1="283.8" y1="58" x2="283.8" y2="472"/>
+<line class="nj-charts-grid" x1="283.8" y1="58" x2="283.8" y2="576"/>
-<text class="nj-charts-tick" x="283.8" y="488">&#8722;20 %</text>
+<text class="nj-charts-tick" x="283.8" y="592">&#8722;20 %</text>
-<text class="nj-charts-tick" x="362.6" y="488">0 %</text>
+<text class="nj-charts-tick" x="362.6" y="592">0 %</text>
-<text class="nj-charts-tick nj-charts-tick--l" x="8" y="506">&#8592; faster</text>
+<text class="nj-charts-tick nj-charts-tick--l" x="8" y="610">&#8592; faster</text>
-<text class="nj-charts-tick nj-charts-tick--r" x="462" y="506">slower &#8594;</text>
+<text class="nj-charts-tick nj-charts-tick--r" x="462" y="610">slower &#8594;</text>
 <text class="nj-charts-rowlabel" x="8" y="62"><tspan class="nj-charts-code">json_fwd.hpp instead of json.hpp</tspan><tspan class="nj-charts-sub"> &#183; header-only TU</tspan></text>
 <path class="nj-charts-clang" d="M362.6 70 H122.32 Q118.32 70 118.32 74 V78 Q118.32 82 122.32 82 H362.6 Z"><title>json_fwd.hpp instead of json.hpp, header-only TU (Apple clang): &#8722;62 %</title></path>
 <text class="nj-charts-val" x="367.6" y="80.5">&#8722;62 %</text>
@@ -46,15 +46,25 @@
 <text class="nj-charts-val" x="367.6" y="340.5">&#8722;5 %</text>
 <path class="nj-charts-gcc" d="M362.6 344 H331.14 Q327.14 344 327.14 348 V352 Q327.14 356 331.14 356 H362.6 Z"><title>JSON_HAS_FILESYSTEM=0, model TU (G&#8203;CC 16): &#8722;9 %</title></path>
 <text class="nj-charts-val" x="367.6" y="354.5">&#8722;9 %</text>
-<text class="nj-charts-rowlabel" x="8" y="374"><tspan class="nj-charts-code">JSON_NO_IO</tspan><tspan class="nj-charts-sub"> &#183; model TU*</tspan></text>
+<text class="nj-charts-rowlabel" x="8" y="374"><tspan class="nj-charts-code">JSON_HAS_STD_FORMAT=0</tspan><tspan class="nj-charts-sub"> &#183; model TU</tspan></text>
-<path class="nj-charts-clang" d="M362.6 382 H358.72 Q354.72 382 354.72 386 V390 Q354.72 394 358.72 394 H362.6 Z"><title>JSON_NO_IO, model TU (Apple clang): &#8722;2 %</title></path>
+<path class="nj-charts-clang" d="M362.6 382 H362.6 Q358.66 382 358.66 385.94 V390.06 Q358.66 394 362.6 394 H362.6 Z"><title>JSON_HAS_STD_FORMAT=0, model TU (Apple clang): &#8722;1 %</title></path>
-<text class="nj-charts-val" x="367.6" y="392.5">&#8722;2 %</text>
+<text class="nj-charts-val" x="367.6" y="392.5">&#8722;1 %</text>
-<path class="nj-charts-gcc" d="M362.6 396 H358.72 Q354.72 396 354.72 400 V404 Q354.72 408 358.72 408 H362.6 Z"><title>JSON_NO_IO, model TU (G&#8203;CC 16): &#8722;2 %</title></path>
+<path class="nj-charts-gcc" d="M362.6 396 H350.84 Q346.84 396 346.84 400 V404 Q346.84 408 350.84 408 H362.6 Z"><title>JSON_HAS_STD_FORMAT=0, model TU (G&#8203;CC 16): &#8722;4 %</title></path>
-<text class="nj-charts-val" x="367.6" y="406.5">&#8722;2 %</text>
+<text class="nj-charts-val" x="367.6" y="406.5">&#8722;4 %</text>
-<text class="nj-charts-rowlabel" x="8" y="426"><tspan class="nj-charts-code">JSON_USE_GLOBAL_UDLS=0</tspan><tspan class="nj-charts-sub"> &#183; model TU*</tspan></text>
+<text class="nj-charts-rowlabel" x="8" y="426"><tspan class="nj-charts-code">JSON_HAS_RANGES=0</tspan><tspan class="nj-charts-sub"> &#183; model TU</tspan></text>
-<path class="nj-charts-clang" d="M362.6 434 h0.5 v12 h-0.5 Z"><title>JSON_USE_GLOBAL_UDLS=0, model TU (Apple clang): 0 %</title></path>
+<path class="nj-charts-clang" d="M362.6 434 H350.84 Q346.84 434 346.84 438 V442 Q346.84 446 350.84 446 H362.6 Z"><title>JSON_HAS_RANGES=0, model TU (Apple clang): &#8722;4 %</title></path>
-<text class="nj-charts-val" x="367.6" y="444.5">0 %</text>
+<text class="nj-charts-val" x="367.6" y="444.5">&#8722;4 %</text>
-<path class="nj-charts-gcc" d="M362.6 448 H362.6 Q358.66 448 358.66 451.94 V456.06 Q358.66 460 362.6 460 H362.6 Z"><title>JSON_USE_GLOBAL_UDLS=0, model TU (G&#8203;CC 16): &#8722;1 %</title></path>
+<path class="nj-charts-gcc" d="M362.6 448 H362.6 Q358.66 448 358.66 451.94 V456.06 Q358.66 460 362.6 460 H362.6 Z"><title>JSON_HAS_RANGES=0, model TU (G&#8203;CC 16): &#8722;1 %</title></path>
 <text class="nj-charts-val" x="367.6" y="458.5">&#8722;1 %</text>
-<line class="nj-charts-zero" x1="362.6" y1="58" x2="362.6" y2="472"/>
+<text class="nj-charts-rowlabel" x="8" y="478"><tspan class="nj-charts-code">JSON_NO_IO</tspan><tspan class="nj-charts-sub"> &#183; model TU*</tspan></text>
+<path class="nj-charts-clang" d="M362.6 486 H358.72 Q354.72 486 354.72 490 V494 Q354.72 498 358.72 498 H362.6 Z"><title>JSON_NO_IO, model TU (Apple clang): &#8722;2 %</title></path>
+<text class="nj-charts-val" x="367.6" y="496.5">&#8722;2 %</text>
+<path class="nj-charts-gcc" d="M362.6 500 H358.72 Q354.72 500 354.72 504 V508 Q354.72 512 358.72 512 H362.6 Z"><title>JSON_NO_IO, model TU (G&#8203;CC 16): &#8722;2 %</title></path>
+<text class="nj-charts-val" x="367.6" y="510.5">&#8722;2 %</text>
+<text class="nj-charts-rowlabel" x="8" y="530"><tspan class="nj-charts-code">JSON_USE_GLOBAL_UDLS=0</tspan><tspan class="nj-charts-sub"> &#183; model TU*</tspan></text>
+<path class="nj-charts-clang" d="M362.6 538 h0.5 v12 h-0.5 Z"><title>JSON_USE_GLOBAL_UDLS=0, model TU (Apple clang): 0 %</title></path>
+<text class="nj-charts-val" x="367.6" y="548.5">0 %</text>
+<path class="nj-charts-gcc" d="M362.6 552 H362.6 Q358.66 552 358.66 555.94 V560.06 Q358.66 564 362.6 564 H362.6 Z"><title>JSON_USE_GLOBAL_UDLS=0, model TU (G&#8203;CC 16): &#8722;1 %</title></path>
+<text class="nj-charts-val" x="367.6" y="562.5">&#8722;1 %</text>
+<line class="nj-charts-zero" x1="362.6" y1="58" x2="362.6" y2="576"/>
 </svg>

docs/mkdocs/docs/integration/compile_times.md

@@ -11,6 +11,11 @@ that cost, ordered by how much they typically save.
     16.2.0 on macOS (Apple M1 Max). All configurations were run interleaved, in alternating order. They show the order
     of magnitude to expect; measure your own code before and after a change.
 
+    The two sections on `std::format` and ranges apply to C++20 only and were measured with `-std=c++20`, so their
+    baselines are slower than the others. Their differences are mostly within the run-to-run noise of a few percent:
+    two complete runs of the same configuration differed by up to 10% in the median. The tables show the later run (one
+    outlier cell was repeated).
+
     The translation units are listed below. They replace the ones used for earlier versions of this page, whose
     sources were not published, so the numbers differ from earlier versions.
 
@@ -66,7 +71,8 @@ that cost, ordered by how much they typically save.
 --8<-- "images/compile_times.svg"
 </div>
 <p class="nj-charts-caption" markdown>Change in compile time at `-O0` against the default configuration of the same
-translation unit, taken from the tables below (medians of nine runs, see above). * computed from the milliseconds in
+translation unit, taken from the tables below (medians of nine runs, see above; the `std::format` and ranges rows are
+measured with `-std=c++20`). * computed from the milliseconds in
 [Options without effect on compile times](#options-without-effect-on-compile-times).</p>
 
 ## Include `json_fwd.hpp` in headers
@@ -214,6 +220,88 @@ from and to `std::string`; whether that fallback exists depends on the platform'
 (`std::wstring` on Windows), so check code that converts paths. `JSON_NO_IO` additionally removes all stream
 functions (see [Options without effect](#options-without-effect-on-compile-times)).
 
+## Switch off `std::format` support
+
+With C++20 and a standard library that provides `<format>`, the library includes it to provide a
+[`std::formatter<basic_json>`](../api/basic_json/std_formatter.md) specialization (see
+[`JSON_HAS_STD_FORMAT`](../api/macros/json_has_std_format.md)). If you do not format JSON values with `std::format`,
+switch the detection off for the whole project:
+
+```cmake
+target_compile_definitions(my_target PRIVATE JSON_HAS_STD_FORMAT=0)
+```
+
+| Compiler    | Translation unit | Default (`-O0` / `-O2`) | `std::format` off (`-O0` / `-O2`) | Change     |
+|-------------|------------------|------------------------:|----------------------------------:|-----------:|
+| Apple clang | model            |       1283 ms / 1544 ms |                1269 ms / 1539 ms |  −1% / 0%  |
+| GCC 16      | model            |       2026 ms / 2482 ms |                1944 ms / 2416 ms | −4% / −3%  |
+| Apple clang | parsing          |       1350 ms / 2278 ms |                1347 ms / 2300 ms |  0% / +1%  |
+| GCC 16      | parsing          |       2715 ms / 4351 ms |                2644 ms / 4287 ms | −3% / −1%  |
+
+<p class="nj-table-caption" markdown>Compile time with `JSON_HAS_STD_FORMAT=0` against the default configuration, both with `-std=c++20`, at `-O0` and `-O2`.</p>
+
+The saving is small, and only GCC shows it: libstdc++'s `<format>` is an additional header that nothing else in the
+translation unit includes, so the macro removes it (and the headers it pulls in). With libc++ (Apple clang), other
+standard headers include `<format>` anyway, so the macro only removes the library's own formatter, and the difference is
+within the noise.
+
+The trade-off: the `std::formatter<basic_json>` specialization is removed, so `#!cpp std::format("{}", j)`,
+and `#!cpp std::format_to` with a JSON value no longer compile (in a test with both compilers on
+macOS, the error was that `std::formatter` must be specialized for the type). Use [`dump()`](../api/basic_json/dump.md),
+or the stream operators, instead.
+
+## Switch off ranges support
+
+With C++20 and a standard library that provides `<ranges>`, the library includes it to declare the
+[`items()`](../api/basic_json/items.md) proxy a *borrowed range*, and uses the `std::ranges` concepts and algorithms for
+construction from range views and for iterator input (see [`JSON_HAS_RANGES`](../api/macros/json_has_ranges.md)). If you
+do not need these, switch the detection off for the whole project:
+
+```cmake
+target_compile_definitions(my_target PRIVATE JSON_HAS_RANGES=0)
+```
+
+| Compiler    | Translation unit | Default (`-O0` / `-O2`) | Ranges off (`-O0` / `-O2`) | Change     |
+|-------------|------------------|------------------------:|---------------------------:|-----------:|
+| Apple clang | model            |       1283 ms / 1544 ms |           1237 ms / 1495 ms | −4% / −3%  |
+| GCC 16      | model            |       2026 ms / 2482 ms |           2004 ms / 2464 ms | −1% / −1%  |
+| Apple clang | parsing          |       1350 ms / 2278 ms |           1311 ms / 2284 ms |  −3% / 0%  |
+| GCC 16      | parsing          |       2715 ms / 4351 ms |           2705 ms / 4427 ms |  0% / +2%  |
+
+<p class="nj-table-caption" markdown>Compile time with `JSON_HAS_RANGES=0` against the default configuration, both with `-std=c++20`, at `-O0` and `-O2`.</p>
+
+The saving is small as well, and with GCC it is not distinguishable from the noise: its `<format>` includes `<span>`,
+so with `std::format` support left on, part of the ranges machinery is still parsed.
+Both macros together save more than either alone:
+
+```cmake
+target_compile_definitions(my_target PRIVATE JSON_HAS_STD_FORMAT=0 JSON_HAS_RANGES=0)
+```
+
+| Compiler    | Translation unit | Default (`-O0` / `-O2`) | Both off (`-O0` / `-O2`) | Change     |
+|-------------|------------------|------------------------:|-------------------------:|-----------:|
+| Apple clang | model            |       1283 ms / 1544 ms |         1223 ms / 1504 ms | −5% / −3%  |
+| GCC 16      | model            |       2026 ms / 2482 ms |         1914 ms / 2378 ms | −6% / −4%  |
+| Apple clang | parsing          |       1350 ms / 2278 ms |         1311 ms / 2271 ms |  −3% / 0%  |
+| GCC 16      | parsing          |       2715 ms / 4351 ms |         2635 ms / 4260 ms | −3% / −2%  |
+
+<p class="nj-table-caption" markdown>Compile time with `JSON_HAS_STD_FORMAT=0` and `JSON_HAS_RANGES=0` against the default configuration, both with `-std=c++20`, at `-O0` and `-O2`.</p>
+
+The trade-offs of `JSON_HAS_RANGES=0` are larger than those of the other options on this page:
+
+- `items()` is no longer a borrowed range. On a temporary, `#!cpp std::ranges::begin(make_json().items())` no longer
+  compiles, and `std::ranges` algorithms such as `#!cpp std::ranges::find_if(make_json().items(), pred)` return
+  `std::ranges::dangling` instead of an iterator, so code that uses the result no longer compiles. Calling them on
+  `#!cpp j.items()` for a named `j` is not affected.
+- [`JSON_HAS_RANGE_VIEW_CONVERSION`](../api/macros/json_has_range_view_conversion.md) becomes `0`, so a JSON array can no
+  longer be constructed from a `std::ranges::view` such as the result of `std::views::filter`. The assignment
+  `#!cpp json j = v | std::views::filter(pred);` no longer compiles.
+- Parsing from an iterator pair loses the C++20 fast path for contiguous iterators that are not pointers (for
+  example those of `std::vector`) and falls back to byte-at-a-time scanning. The result is the same; parsing may be
+  slower (not measured here).
+
+The first two were checked with both compilers on macOS; the third is read from the library's source.
+
 ## Use C++20 modules
 
 With a toolchain that supports named modules, `import nlohmann.json;` compiles the library once into a module and
@@ -343,6 +431,8 @@ and −2% / 0% (GCC). To get a saving, combine it with
 
 - [`JSON_HAS_FILESYSTEM`](../api/macros/json_has_filesystem.md) - control the conversions from and to
   `std::filesystem::path`
+- [`JSON_HAS_STD_FORMAT`](../api/macros/json_has_std_format.md) - control the `std::formatter` specialization
+- [`JSON_HAS_RANGES`](../api/macros/json_has_ranges.md) - control the use of `std::ranges`
 - [`JSON_NO_AUTOMATIC_UDLS`](../api/macros/json_no_automatic_udls.md) - do not include the user-defined string
   literals automatically
 - [Modules](../features/modules.md) - C++20 module support

docs/mkdocs/mkdocs.yml

@@ -130,6 +130,7 @@ nav:
           - features/element_access/unchecked_access.md
           - features/element_access/checked_access.md
           - features/element_access/default_value.md
+      - features/files.md
       - features/iterators.md
       - features/merge_patch.md
       - features/json_patch.md
@@ -142,6 +143,7 @@ nav:
       - features/performance.md
       - Parsing:
           - features/parsing/index.md
+          - features/parsing/error_recovery.md
           - features/parsing/json_lines.md
           - features/parsing/large_and_partial_input.md
           - features/parsing/parse_exceptions.md

docs/mkdocs/scripts/figures/charts.py

@@ -73,6 +73,8 @@ COMPILE = [
     ("JSON_NO_AUTOMATIC_UDLS", "parsing TU", -1, -2, False),
     ("JSON_HAS_FILESYSTEM=0 + JSON_NO_IO", "model TU", -8, -11, False),
     ("JSON_HAS_FILESYSTEM=0", "model TU", -5, -9, False),
+    ("JSON_HAS_STD_FORMAT=0", "model TU", -1, -4, False),
+    ("JSON_HAS_RANGES=0", "model TU", -4, -1, False),
     # computed from the ms in "Options without effect on compile times": 1034/1051, 1756/1801, 1055/1051, 1783/1801
     ("JSON_NO_IO", "model TU", round((1034 / 1051 - 1) * 100), round((1756 / 1801 - 1) * 100), True),
     ("JSON_USE_GLOBAL_UDLS=0", "model TU", round((1055 / 1051 - 1) * 100), round((1783 / 1801 - 1) * 100), True),

include/nlohmann/detail/input/input_adapters.hpp

@@ -58,6 +58,19 @@
         #define JSON_UNLOCKED_FILE_IO 0
     #endif
     #undef JSON_UNLOCKED_FILE_IO_MSAN
+
+    // Whether file_input_adapter scans the buffer of the FILE directly, as
+    // getc_unlocked() reads it: the fields it uses are those that the C
+    // library's own getc_unlocked() macro uses (glibc: _IO_read_ptr and
+    // _IO_read_end; Apple: _p and _r). It needs the file locked for the whole
+    // parse, as JSON_UNLOCKED_FILE_IO == 1 has it (so JSON_DISABLE_UNLOCKED_FILE_IO
+    // and MemorySanitizer builds turn it off).
+    // (uClibc defines __GLIBC__ as well, but its FILE has other fields)
+    #if JSON_UNLOCKED_FILE_IO == 1 && ((defined(__GLIBC__) && !defined(__UCLIBC__)) || defined(__APPLE__))
+        #define JSON_FILE_WINDOW 1
+    #else
+        #define JSON_FILE_WINDOW 0
+    #endif
 #endif  // JSON_NO_IO
 
 NLOHMANN_JSON_NAMESPACE_BEGIN
@@ -179,8 +192,12 @@ class token_carry_buffer
     char* m_capacity_end = nullptr;
 };
 /*!
-Input adapter for stdio file access. This adapter read only 1 byte and do not use any
+Input adapter for stdio file access. The file is locked for the lifetime of the
- buffer. This adapter is a very low level adapter.
+adapter and read with the unlocked stdio functions (see
+JSON_DISABLE_UNLOCKED_FILE_IO). With glibc and the Apple C library, the
+characters in the buffer of the FILE are also scanned in bulk, as the get area
+of a stream is (see input_stream_adapter): exactly the characters getc_unlocked()
+would return are consumed, so the file is left where it was left before.
 */
 class file_input_adapter
 {
@@ -199,6 +216,12 @@ class file_input_adapter
     file_input_adapter(const file_input_adapter&) = delete;
     file_input_adapter(file_input_adapter&& rhs) noexcept
         : m_file(rhs.m_file)
+#if JSON_FILE_WINDOW
+        , token_carry(std::move(rhs.token_carry))
+        , token_segment(rhs.token_segment)
+        , current_in_buffer(rhs.current_in_buffer)
+        , tracking_token(rhs.tracking_token)
+#endif
     {
         // the lock is owned by the thread, not the adapter: hand it over
         rhs.m_file = nullptr;
@@ -210,6 +233,20 @@ class file_input_adapter
         unlock();
     }
 
+#if JSON_FILE_WINDOW
+    std::char_traits<char>::int_type get_character()
+    {
+        // what getc_unlocked() does, while the buffer has characters
+        const char* const next = buffer_ptr();
+        if (JSON_HEDLEY_LIKELY(next != buffer_end()))
+        {
+            current_in_buffer = true;
+            buffer_bump(1);
+            return std::char_traits<char>::to_int_type(*next);
+        }
+        return get_character_slow();
+    }
+#else
     std::char_traits<char>::int_type get_character() noexcept
     {
         // the file is locked for the lifetime of the adapter, so the
@@ -222,6 +259,7 @@ class file_input_adapter
         return std::fgetc(m_file);
 #endif
     }
+#endif
 
     // returns the number of characters successfully read
     template<class T>
@@ -230,7 +268,134 @@ class file_input_adapter
         return fread(dest, 1, sizeof(T) * count, m_file);
     }
 
+#if JSON_FILE_WINDOW
+    // The lexer may scan the buffer of the FILE directly, as the window of
+    // input_stream_adapter: the buffer ends wherever the C library's buffer
+    // ends, and is refilled by the next get_character().
+    static constexpr bool supports_bulk_scan = true;
+    static constexpr bool bulk_window = true;
+
+    // Pointer to the next unread character; only valid when bulk_remaining() > 0.
+    const char* bulk_data() const
+    {
+        return buffer_ptr();
+    }
+
+    // Number of unread characters in the buffer.
+    std::size_t bulk_remaining() const
+    {
+        if (!current_in_buffer)
+        {
+            return 0;
+        }
+        return static_cast<std::size_t>(buffer_end() - buffer_ptr());
+    }
+
+    // Consume @a n characters previously inspected via bulk_data().
+    void bulk_skip(std::size_t n)
+    {
+        JSON_ASSERT(n <= bulk_remaining());
+        buffer_bump(n);
+    }
+
+    // The adapter keeps the current token for the lexer's error messages, as
+    // input_stream_adapter does.
+    static constexpr bool supports_token_tracking = true;
+
+    void begin_token() noexcept
+    {
+        tracking_token = true;
+        token_carry.clear();
+        token_segment = buffer_ptr();
+    }
+
+    void mark_token_start(std::char_traits<char>::int_type c)
+    {
+        tracking_token = true;
+        token_carry.clear();
+        if (current_in_buffer)
+        {
+            token_segment = buffer_ptr() - 1;
+        }
+        else
+        {
+            token_segment = nullptr;
+            token_carry.push_back(std::char_traits<char>::to_char_type(c));
+        }
+    }
+
+    void copy_token(std::vector<char>& out, bool drop_last) const
+    {
+        token_carry.copy_to(out);
+        if (token_segment != nullptr)
+        {
+            out.insert(out.end(), token_segment, buffer_ptr());
+        }
+        if (drop_last && !out.empty())
+        {
+            out.pop_back();
+        }
+    }
+#endif
+
   private:
+#if JSON_FILE_WINDOW
+    // the buffer of the FILE: [buffer_ptr(), buffer_end()) are the characters
+    // getc_unlocked() would return next
+#if defined(__GLIBC__)
+    const char* buffer_ptr() const noexcept
+    {
+        return m_file->_IO_read_ptr;
+    }
+    const char* buffer_end() const noexcept
+    {
+        return m_file->_IO_read_end;
+    }
+    void buffer_bump(std::size_t n) noexcept
+    {
+        m_file->_IO_read_ptr += n;
+    }
+#else
+    const char* buffer_ptr() const noexcept
+    {
+        return reinterpret_cast<const char*>(m_file->_p);
+    }
+    const char* buffer_end() const noexcept
+    {
+        return m_file->_r > 0 ? reinterpret_cast<const char*>(m_file->_p) + m_file->_r : reinterpret_cast<const char*>(m_file->_p);
+    }
+    void buffer_bump(std::size_t n) noexcept
+    {
+        m_file->_p += n;
+        m_file->_r -= static_cast<int>(n);
+    }
+#endif
+
+    // the buffer is exhausted: getc_unlocked() refills it
+    JSON_HEDLEY_NEVER_INLINE std::char_traits<char>::int_type get_character_slow()
+    {
+        if (token_segment != nullptr)
+        {
+            token_carry.append(token_segment, buffer_end());
+            token_segment = nullptr;
+        }
+        const auto res = getc_unlocked(m_file);
+        if (res == EOF)
+        {
+            current_in_buffer = false;
+            return std::char_traits<char>::eof();
+        }
+        // the character was taken from the refilled buffer, at buffer_ptr() - 1
+        current_in_buffer = true;
+        if (tracking_token)
+        {
+            token_segment = buffer_ptr() - 1;
+        }
+        return res;
+    }
+
+#endif
+
     // Lock the file once for the whole parse instead of once per character.
     // The lock is recursive, so fread() and the user's own stdio calls on the
     // same thread keep working.
@@ -259,6 +424,16 @@ class file_input_adapter
 
     /// the file pointer to read from
     std::FILE* m_file;
+#if JSON_FILE_WINDOW
+    /// the part of the current token that was in earlier buffers
+    token_carry_buffer token_carry{}; // NOLINT(readability-redundant-member-init)
+    /// start of the current token's part in the buffer, or nullptr
+    const char* token_segment = nullptr;
+    /// whether the character last returned by get_character() is in the buffer
+    bool current_in_buffer = false;
+    /// whether a token is tracked at all; the binary readers never start one
+    bool tracking_token = false;
+#endif
 };
 
 /*!
@@ -789,17 +964,22 @@ class iterator_input_adapter
         return get_consumed_count() + eof_reads;
     }
 
-    // number of newlines in the consumed range [0, last_index); @a line_start
+    // number of newlines in the consumed range [first_index, last_index); if
-    // is set to the index after the last of them (0 if there is none); only
+    // there is one, @a line_start is set to the index after the last of them,
-    // valid when supports_seek
+    // otherwise it is left unchanged; only valid when supports_seek
-    std::size_t count_consumed_lines(std::size_t last_index, std::size_t& line_start) const
+    std::size_t count_consumed_lines(std::size_t first_index, std::size_t last_index, std::size_t& line_start) const
     {
-        const auto to = std::next(begin, static_cast<typename std::iterator_traits<IteratorType>::difference_type>(last_index));
+        using difference_type = typename std::iterator_traits<IteratorType>::difference_type;
+        const auto from = std::next(begin, static_cast<difference_type>(first_index));
+        const auto to = std::next(from, static_cast<difference_type>(last_index - first_index));
         const auto newline = char_traits<char_type>::to_char_type('\n');
         using reverse_it = std::reverse_iterator<IteratorType>;
-        const auto last_newline = std::find(reverse_it(to), reverse_it(begin), newline);
+        const auto last_newline = std::find(reverse_it(to), reverse_it(from), newline);
+        if (last_newline.base() != from)
+        {
             line_start = static_cast<std::size_t>(std::distance(begin, last_newline.base()));
-        return static_cast<std::size_t>(std::count(begin, last_newline.base(), newline));
+        }
+        return static_cast<std::size_t>(std::count(from, last_newline.base(), newline));
     }
 
     // append the already-consumed characters in the half-open range

include/nlohmann/detail/input/json_sax.hpp

@@ -138,7 +138,9 @@ struct json_sax
     @param[in] position    the position in the input where the error occurs
     @param[in] last_token  the last read token
     @param[in] ex          an exception object describing the error
-    @return whether parsing should proceed (must return false)
+    @return whether to recover from the error: false stops parsing; true
+            repairs the error and continues, or, if that is not possible,
+            stops after completing the value read so far
     */
     virtual bool parse_error(std::size_t position,
                              const std::string& last_token,
@@ -485,6 +487,9 @@ stored in place as it arrives, so after a parse error the value holds what was
 read up to it.
 
 @tparam BasicJsonType  the JSON type
+@tparam InputAdapterType  the input adapter of the lexer that can be passed to
+                          the constructor to record diagnostic positions; it
+                          does not matter if no lexer is passed
 @tparam Collect  whether to collect the values of containers of unknown size on
                  a stack (only for a caller that does not use the value after a
                  parse error)
@@ -493,7 +498,7 @@ read up to it.
                    it, so any type with its three functions works (as
                    materialize() of json_view uses it)
 */
-template<typename BasicJsonType, typename InputAdapterType, bool Collect = false, typename LexerType = lexer<BasicJsonType, InputAdapterType>>
+template<typename BasicJsonType, typename InputAdapterType = string_input_adapter_type, bool Collect = false, typename LexerType = lexer<BasicJsonType, InputAdapterType>>
 class json_sax_dom_parser : public json_sax_dom_error_state
 {
   public:
@@ -799,6 +804,8 @@ class json_sax_dom_parser : public json_sax_dom_error_state
         JSON_HEDLEY_NEVER_INLINE
         void grow()
         {
+            // the stack starts with local_capacity frames and only grows
+            JSON_ASSERT(capacity != 0);
             const std::size_t new_capacity = capacity * 2;
             std::unique_ptr<frame[]> new_heap(new frame[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
             for (std::size_t i = 0; i < count; ++i)
@@ -904,7 +911,8 @@ class json_sax_dom_parser : public json_sax_dom_error_state
             {
                 hashes.push_back(hasher(element.first));
             }
-            ordered_data.key_indexes.emplace_back(new object_key_index(std::move(hashes)));
+            std::unique_ptr<object_key_index> index(new object_key_index(std::move(hashes))); // NOLINT(cppcoreguidelines-owning-memory,modernize-make-unique): std::make_unique is C++14
+            ordered_data.key_indexes.push_back(std::move(index));
             f.ordered.index = ordered_data.key_indexes.back().get();
         }
 
@@ -1634,7 +1642,7 @@ class json_sax_dom_binary_parser : public json_sax_dom_error_state
     lexer_t* m_lexer_ref = nullptr;
 };
 
-template<typename BasicJsonType, typename InputAdapterType>
+template<typename BasicJsonType, typename InputAdapterType = string_input_adapter_type>
 class json_sax_dom_callback_parser : public json_sax_dom_error_state
 {
   public:

include/nlohmann/detail/input/lexer.hpp

@@ -10,7 +10,7 @@
 
 #include <array> // array
 #include <cstddef> // size_t
-#include <cstdint> // uint32_t
+#include <cstdint> // uint8_t, uint32_t
 #include <cstdio> // snprintf
 #include <cstring> // memcmp
 #include <initializer_list> // initializer_list
@@ -387,6 +387,20 @@ class lexer : public lexer_base<BasicJsonType>
         return codepoint;
     }
 
+    /// the code point of the surrogate pair @a high (U+D800..U+DBFF) and @a low (U+DC00..U+DFFF)
+    static constexpr int combine_surrogates(const int high, const int low) noexcept
+    {
+        return static_cast<int>(
+                   // high surrogate occupies the most significant 22 bits
+                   (static_cast<unsigned int>(high) << 10u)
+                   // low surrogate occupies the least significant 15 bits
+                   + static_cast<unsigned int>(low)
+                   // there is still the 0xD800, 0xDC00, and 0x10000 noise
+                   // in the result, so we have to subtract with:
+                   // (0xD800 << 10) + DC00 - 0x10000 = 0x35FDC00
+                   - 0x35FDC00u);
+    }
+
     /*!
     @brief check if the next byte(s) are inside a given range
 
@@ -420,6 +434,7 @@ class lexer : public lexer_base<BasicJsonType>
             else
             {
                 error_message = "invalid string: ill-formed UTF-8 byte";
+                error_cause = error_kind::string_ill_formed_utf8;
                 return false;
             }
         }
@@ -542,6 +557,7 @@ class lexer : public lexer_base<BasicJsonType>
                 case char_traits<char_type>::eof():
                 {
                     error_message = "invalid string: missing closing quote";
+                    error_cause = error_kind::string;
                     return token_type::parse_error;
                 }
 
@@ -598,6 +614,7 @@ class lexer : public lexer_base<BasicJsonType>
                             if (JSON_HEDLEY_UNLIKELY(codepoint1 == -1))
                             {
                                 error_message = "invalid string: '\\u' must be followed by 4 hex digits";
+                                error_cause = error_kind::string_invalid_escape;
                                 return token_type::parse_error;
                             }
 
@@ -605,13 +622,23 @@ class lexer : public lexer_base<BasicJsonType>
                             if (0xD800 <= codepoint1 && codepoint1 <= 0xDBFF)
                             {
                                 // expect next \uxxxx entry
-                                if (JSON_HEDLEY_LIKELY(get() == '\\' && get() == 'u'))
+                                if (JSON_HEDLEY_LIKELY(get() == '\\'))
                                 {
+                                    if (JSON_HEDLEY_UNLIKELY(get() != 'u'))
+                                    {
+                                        // current is the character escaped by the backslash
+                                        error_message = "invalid string: surrogate U+D800..U+DBFF must be followed by U+DC00..U+DFFF";
+                                        error_cause = error_kind::string_invalid_escape;
+                                        string_error_resume = resume_kind::escaped_character;
+                                        return token_type::parse_error;
+                                    }
+
                                     const int codepoint2 = get_codepoint();
 
                                     if (JSON_HEDLEY_UNLIKELY(codepoint2 == -1))
                                     {
                                         error_message = "invalid string: '\\u' must be followed by 4 hex digits";
+                                        error_cause = error_kind::string_invalid_escape;
                                         return token_type::parse_error;
                                     }
 
@@ -619,25 +646,23 @@ class lexer : public lexer_base<BasicJsonType>
                                     if (JSON_HEDLEY_LIKELY(0xDC00 <= codepoint2 && codepoint2 <= 0xDFFF))
                                     {
                                         // overwrite codepoint
-                                        codepoint = static_cast<int>(
+                                        codepoint = combine_surrogates(codepoint1, codepoint2);
-                                                        // high surrogate occupies the most significant 22 bits
-                                                        (static_cast<unsigned int>(codepoint1) << 10u)
-                                                        // low surrogate occupies the least significant 15 bits
-                                                        + static_cast<unsigned int>(codepoint2)
-                                                        // there is still the 0xD800, 0xDC00, and 0x10000 noise
-                                                        // in the result, so we have to subtract with:
-                                                        // (0xD800 << 10) + DC00 - 0x10000 = 0x35FDC00
-                                                        - 0x35FDC00u);
                                     }
                                     else
                                     {
+                                        // the second escape was read completely and is a
+                                        // code point of its own
                                         error_message = "invalid string: surrogate U+D800..U+DBFF must be followed by U+DC00..U+DFFF";
+                                        error_cause = error_kind::string_invalid_escape;
+                                        string_error_resume = resume_kind::after_escape;
+                                        string_error_codepoint = codepoint2;
                                         return token_type::parse_error;
                                     }
                                 }
                                 else
                                 {
                                     error_message = "invalid string: surrogate U+D800..U+DBFF must be followed by U+DC00..U+DFFF";
+                                    error_cause = error_kind::string_invalid_escape;
                                     return token_type::parse_error;
                                 }
                             }
@@ -645,7 +670,10 @@ class lexer : public lexer_base<BasicJsonType>
                             {
                                 if (JSON_HEDLEY_UNLIKELY(0xDC00 <= codepoint1 && codepoint1 <= 0xDFFF))
                                 {
+                                    // the escape was read completely
                                     error_message = "invalid string: surrogate U+DC00..U+DFFF must follow U+D800..U+DBFF";
+                                    error_cause = error_kind::string_invalid_escape;
+                                    string_error_resume = resume_kind::after_escape;
                                     return token_type::parse_error;
                                 }
                             }
@@ -662,6 +690,7 @@ class lexer : public lexer_base<BasicJsonType>
                         // other characters after escape
                         default:
                             error_message = "invalid string: forbidden character after backslash";
+                            error_cause = error_kind::string;
                             return token_type::parse_error;
                     }
 
@@ -738,6 +767,7 @@ class lexer : public lexer_base<BasicJsonType>
                         }
                     };
                     error_message = control_character_messages[static_cast<std::size_t>(current)];
+                    error_cause = error_kind::string;
                     return token_type::parse_error;
                 }
 
@@ -960,6 +990,7 @@ class lexer : public lexer_base<BasicJsonType>
                 default:
                 {
                     error_message = "invalid string: ill-formed UTF-8 byte";
+                    error_cause = error_kind::string_ill_formed_utf8;
                     return token_type::parse_error;
                 }
             }
@@ -1015,6 +1046,7 @@ class lexer : public lexer_base<BasicJsonType>
                         case char_traits<char_type>::eof():
                         {
                             error_message = "invalid comment; missing closing '*/'";
+                            error_cause = error_kind::unclosed_comment;
                             return false;
                         }
 
@@ -1045,6 +1077,7 @@ class lexer : public lexer_base<BasicJsonType>
             default:
             {
                 error_message = "invalid comment; expecting '/' or '*' after '/'";
+                error_cause = error_kind::other;
                 return false;
             }
         }
@@ -1170,6 +1203,7 @@ scan_number_minus:
             default:
             {
                 error_message = "invalid number; expected digit after '-'";
+                error_cause = error_kind::number;
                 return token_type::parse_error;
             }
         }
@@ -1256,6 +1290,7 @@ scan_number_decimal1:
             default:
             {
                 error_message = "invalid number; expected digit after '.'";
+                error_cause = error_kind::number;
                 return token_type::parse_error;
             }
         }
@@ -1324,6 +1359,7 @@ scan_number_exponent:
             {
                 error_message =
                     "invalid number; expected '+', '-', or digit after exponent";
+                error_cause = error_kind::number;
                 return token_type::parse_error;
             }
         }
@@ -1350,6 +1386,7 @@ scan_number_sign:
             default:
             {
                 error_message = "invalid number; expected digit after exponent sign";
+                error_cause = error_kind::number;
                 return token_type::parse_error;
             }
         }
@@ -1771,6 +1808,7 @@ scan_number_done:
             if (JSON_HEDLEY_UNLIKELY(char_traits<char_type>::to_char_type(get()) != static_cast<char_type>(rest[i])))
             {
                 error_message = "invalid literal";
+                error_cause = error_kind::other;
                 return token_type::parse_error;
             }
         }
@@ -2192,17 +2230,25 @@ scan_number_done:
         return position.chars_read_total;
     }
 
-    /// seekable adapter: count the lines in the consumed input; O(input read
+    /// seekable adapter: count the lines in the consumed input. Only the
-    /// so far), but only called on errors
+    /// input consumed since the previous call is scanned, so a parse that
+    /// reports many errors (a SAX parser recovering from them) stays linear
+    /// in the input instead of rescanning it for every error.
     position_t get_position_impl(std::true_type /*lazy*/) const
     {
         position_t result;
         result.chars_read_total = get_chars_read_total();
         // reads at EOF count as characters but are not part of the input
         const std::size_t consumed = (std::min)(result.chars_read_total, ia.get_consumed_count());
-        std::size_t line_start = 0;
+        if (JSON_HEDLEY_UNLIKELY(consumed < position_cache.consumed))
-        result.lines_read = ia.count_consumed_lines(consumed, line_start);
+        {
-        result.chars_read_current_line = result.chars_read_total - line_start;
+            // the consumed input never shrinks; recount from the start if it did
+            position_cache = {};
+        }
+        position_cache.lines += ia.count_consumed_lines(position_cache.consumed, consumed, position_cache.line_start);
+        position_cache.consumed = consumed;
+        result.lines_read = position_cache.lines;
+        result.chars_read_current_line = result.chars_read_total - position_cache.line_start;
         return result;
     }
 
@@ -2386,6 +2432,7 @@ scan_number_done:
         if (get_chars_read_total() == 0 && !skip_bom())
         {
             error_message = "invalid BOM; must be 0xEF 0xBB 0xBF if given";
+            error_cause = error_kind::other;
             return token_type::parse_error;
         }
 
@@ -2495,10 +2542,583 @@ scan_number_done:
             // error
             default:
                 error_message = "invalid literal";
+                error_cause = error_kind::other;
                 return token_type::parse_error;
         }
     }
 
+  public:
+    /////////////////////
+    // error recovery
+    /////////////////////
+
+    /*!
+    @brief make the best of the token that scan() rejected
+
+    Called by the parser after scan() returned token_type::parse_error and the
+    SAX parser asked to recover from the error (see #3989). Keeps what can be
+    read of the token and skips the rest:
+
+    - A string keeps its characters. An unknown escape stands for the escaped
+      character itself (as in JavaScript), an invalid Unicode escape and ill-formed
+      UTF-8 become U+FFFD, and a control character is kept. A line break or the
+      end of the input ends a string that lacks its closing quote.
+    - A number keeps its longest valid prefix, e.g. `1` for `1.` or `1e+`.
+    - A block comment that is not closed runs to the end of the input.
+    - Anything else is skipped.
+
+    The rest of an invalid token is skipped up to the next delimiter
+    (whitespace, a structural character, or a quote). A delimiter that the
+    invalid token consumed is returned to the input, so that the next scan()
+    reads it.
+
+    @return token_type::value_string or a number token type if a string or a
+            number could be read, token_type::end_of_input for a block comment
+            that is not closed, token_type::uninitialized otherwise
+    */
+    token_type recover_token()
+    {
+        const resume_kind resume = string_error_resume;
+        const int codepoint = string_error_codepoint;
+        string_error_resume = resume_kind::character;
+        string_error_codepoint = -1;
+
+        const error_kind cause = error_cause;
+        error_cause = error_kind::other;
+
+        switch (cause)
+        {
+            case error_kind::string:
+            case error_kind::string_invalid_escape:
+            case error_kind::string_ill_formed_utf8:
+                return recover_string(cause, resume, codepoint);
+
+            case error_kind::number:
+                return recover_number();
+
+            case error_kind::unclosed_comment:
+                // the comment runs to the end of the input
+                return token_type::end_of_input;
+
+            case error_kind::other:
+            default:
+                skip_to_delimiter();
+                return token_type::uninitialized;
+        }
+    }
+
+    /*!
+    @brief return the token that scan() read last to the input, so that the
+           next scan() reads it again
+
+    Called by the parser when recovering from an error. The token must be a
+    single character (',', ':', '[', ']', '{', or '}') or the end of the
+    input, and scan() must have read it last.
+    */
+    void unget_token()
+    {
+        JSON_ASSERT(!next_unget);
+        unget();
+    }
+
+    /*!
+    @brief let the token string for the next error begin at the current character
+
+    The token string of an error reaches back to the beginning of the last
+    string or number. After an error, the parser calls this function so that
+    the next error does not report (and, with many errors, copy) everything
+    read since then.
+    */
+    void restart_token_string()
+    {
+        restart_token_string_impl(std::integral_constant<bool, lazy_token_string> {});
+    }
+
+  private:
+    /// how recover_string() continues after the error scan_string() reported
+    enum class resume_kind : std::uint8_t
+    {
+        /// current is the next character of the string (or the end of input)
+        character,
+        /// current is the character escaped by the preceding backslash
+        escaped_character,
+        /// current is the last character of a complete escape
+        after_escape
+    };
+
+    /// what kind of token scan() rejected, which decides how recover_token() repairs it
+    enum class error_kind : std::uint8_t
+    {
+        /// anything not listed below; skipped up to the next delimiter
+        other,
+        /// a string
+        string,
+        /// a string with an invalid Unicode escape, which becomes U+FFFD
+        string_invalid_escape,
+        /// a string with ill-formed UTF-8, which becomes U+FFFD
+        string_ill_formed_utf8,
+        /// a number
+        number,
+        /// a block comment that is not closed
+        unclosed_comment
+    };
+
+    /// whether current ends an invalid token (see recover_token())
+    bool current_is_delimiter() const noexcept
+    {
+        switch (current)
+        {
+            case ' ':
+            case '\t':
+            case '\n':
+            case '\r':
+            case '[':
+            case ']':
+            case '{':
+            case '}':
+            case ',':
+            case ':':
+            case '\"':
+#if !JSON_STRICT_NUL_HANDLING
+            case '\0':
+#endif
+            case char_traits<char_type>::eof():
+                return true;
+
+            case '/':
+                return ignore_comments;
+
+            default:
+                return false;
+        }
+    }
+
+    /// skip the rest of an invalid token and return its delimiter to the input
+    void skip_to_delimiter()
+    {
+        while (!current_is_delimiter())
+        {
+            get();
+        }
+
+        if (current != char_traits<char_type>::eof())
+        {
+            unget();
+        }
+    }
+
+    /// append U+FFFD REPLACEMENT CHARACTER to token_buffer
+    void add_replacement_character()
+    {
+        add_codepoint(0xFFFD);
+    }
+
+    /// append the UTF-8 encoding of @a codepoint (not a surrogate) to token_buffer
+    void add_codepoint(const int codepoint)
+    {
+        JSON_ASSERT(0x00 <= codepoint && codepoint <= 0x10FFFF);
+        encode_utf8(static_cast<std::uint32_t>(codepoint), [this](std::uint32_t byte)
+        {
+            add(static_cast<char_int_type>(byte));
+        });
+    }
+
+    /// append a code point read from a Unicode escape; a surrogate becomes U+FFFD
+    void add_escaped_codepoint(const int codepoint)
+    {
+        if (0xD800 <= codepoint && codepoint <= 0xDFFF)
+        {
+            add_replacement_character();
+        }
+        else
+        {
+            add_codepoint(codepoint);
+        }
+    }
+
+    /*!
+    @brief remove an incomplete UTF-8 sequence from the end of token_buffer
+
+    next_byte_in_range() adds the bytes of a sequence as it checks them, so
+    when it rejects a byte, the beginning of the sequence is already in
+    token_buffer, which otherwise holds only complete sequences.
+
+    @return whether an incomplete sequence was removed
+    */
+    bool remove_incomplete_utf8_sequence()
+    {
+        std::size_t lead = token_buffer.size();
+        std::size_t continuation_bytes = 0;
+        while (lead > 0 && continuation_bytes < 3
+                && (static_cast<unsigned char>(token_buffer[lead - 1]) & 0xC0u) == 0x80u)
+        {
+            --lead;
+            ++continuation_bytes;
+        }
+        if (lead == 0)
+        {
+            return false;
+        }
+
+        const auto lead_byte = static_cast<unsigned char>(token_buffer[lead - 1]);
+        std::size_t expected = 0;
+        if (lead_byte >= 0xF0)
+        {
+            expected = 3;
+        }
+        else if (lead_byte >= 0xE0)
+        {
+            expected = 2;
+        }
+        else if (lead_byte >= 0xC0)
+        {
+            expected = 1;
+        }
+        if (continuation_bytes >= expected)
+        {
+            return false;
+        }
+
+        token_buffer.resize(lead - 1);
+        return true;
+    }
+
+    /*!
+    @brief read the UTF-8 sequence that begins with current, which is not ASCII
+    @return whether the next character must be read; false if current still
+            needs to be handled, because it does not belong to the sequence
+    */
+    bool recover_utf8_sequence()
+    {
+        // the number of continuation bytes and the range of the first one;
+        // see the ranges in scan_string()
+        std::size_t count = 0;
+        char_int_type low = 0x80;
+        char_int_type high = 0xBF;
+        if (current >= 0xC2 && current <= 0xDF)
+        {
+            count = 1;
+        }
+        else if (current >= 0xE0 && current <= 0xEF)
+        {
+            count = 2;
+            low = (current == 0xE0) ? 0xA0 : 0x80;
+            high = (current == 0xED) ? 0x9F : 0xBF;
+        }
+        else if (current >= 0xF0 && current <= 0xF4)
+        {
+            count = 3;
+            low = (current == 0xF0) ? 0x90 : 0x80;
+            high = (current == 0xF4) ? 0x8F : 0xBF;
+        }
+        else
+        {
+            // an ill-formed byte
+            add_replacement_character();
+            return true;
+        }
+
+        const std::size_t start = token_buffer.size();
+        add(current);
+        for (std::size_t i = 0; i < count; ++i)
+        {
+            get();
+            if (current < low || current > high)
+            {
+                token_buffer.resize(start);
+                add_replacement_character();
+                return false;
+            }
+            add(current);
+            low = 0x80;
+            high = 0xBF;
+        }
+        return true;
+    }
+
+    /*!
+    @brief read the low surrogate that must follow the high surrogate @a high
+    @return whether the next character must be read; false if current still
+            needs to be handled
+    */
+    bool recover_low_surrogate(int high)
+    {
+        while (true)
+        {
+            if (get() != '\\')
+            {
+                add_replacement_character();
+                return false;
+            }
+            if (get() != 'u')
+            {
+                add_replacement_character();
+                // not 'u', so this does not come back here
+                return recover_escape();
+            }
+
+            const int low = get_codepoint();
+            if (low == -1)
+            {
+                add_replacement_character();
+                return false;
+            }
+            if (0xDC00 <= low && low <= 0xDFFF)
+            {
+                add_codepoint(combine_surrogates(high, low));
+                return true;
+            }
+
+            // high has no low surrogate
+            add_replacement_character();
+            if (low < 0xD800 || low > 0xDBFF)
+            {
+                add_codepoint(low);
+                return true;
+            }
+            // another high surrogate
+            high = low;
+        }
+    }
+
+    /*!
+    @brief read the escape whose backslash was read; current is the escaped character
+    @return whether the next character must be read; false if current still
+            needs to be handled
+    */
+    bool recover_escape()
+    {
+        switch (current)
+        {
+            case '\"':
+                add('\"');
+                return true;
+            case '\\':
+                add('\\');
+                return true;
+            case '/':
+                add('/');
+                return true;
+            case 'b':
+                add('\b');
+                return true;
+            case 'f':
+                add('\f');
+                return true;
+            case 'n':
+                add('\n');
+                return true;
+            case 'r':
+                add('\r');
+                return true;
+            case 't':
+                add('\t');
+                return true;
+
+            case 'u':
+            {
+                const int codepoint = get_codepoint();
+                if (codepoint == -1)
+                {
+                    add_replacement_character();
+                    return false;
+                }
+                if (0xD800 <= codepoint && codepoint <= 0xDBFF)
+                {
+                    return recover_low_surrogate(codepoint);
+                }
+                add_escaped_codepoint(codepoint);
+                return true;
+            }
+
+            // an unknown escape stands for the escaped character
+            default:
+                return false;
+        }
+    }
+
+    /*!
+    @brief read the rest of a string after scan_string() rejected it
+
+    token_buffer holds what scan_string() read before the error. See
+    recover_token() for how errors are repaired.
+
+    @param[in] cause      the kind of error scan_string() reported
+    @param[in] resume     how to continue, see resume_kind
+    @param[in] codepoint  for a high surrogate followed by an escape of another
+                          code point: that code point; -1 otherwise
+    */
+    token_type recover_string(const error_kind cause, const resume_kind resume, const int codepoint)
+    {
+        // whether the next character must be read before it can be handled
+        bool fetch = false;
+
+        if (cause == error_kind::string_invalid_escape
+                || (cause == error_kind::string_ill_formed_utf8 && remove_incomplete_utf8_sequence()))
+        {
+            add_replacement_character();
+        }
+
+        switch (resume)
+        {
+            case resume_kind::escaped_character:
+                fetch = recover_escape();
+                break;
+            case resume_kind::after_escape:
+                if (0xD800 <= codepoint && codepoint <= 0xDBFF)
+                {
+                    fetch = recover_low_surrogate(codepoint);
+                }
+                else
+                {
+                    if (codepoint != -1)
+                    {
+                        add_escaped_codepoint(codepoint);
+                    }
+                    fetch = true;
+                }
+                break;
+            case resume_kind::character:
+            default:
+                break;
+        }
+
+        while (true)
+        {
+            if (fetch)
+            {
+                get();
+            }
+            fetch = true;
+
+            switch (current)
+            {
+                case '\"':
+                // a line break or the end of the input ends a string that
+                // lacks its closing quote
+                case '\n':
+                case '\r':
+                case char_traits<char_type>::eof():
+                    return token_type::value_string;
+
+#if !JSON_STRICT_NUL_HANDLING
+                case '\0':
+                    // the end of the input, see scan()
+                    unget();
+                    return token_type::value_string;
+#endif
+
+                case '\\':
+                    get();
+                    fetch = recover_escape();
+                    break;
+
+                default:
+                    if (current < 0x80)
+                    {
+                        // including control characters
+                        add(current);
+                    }
+                    else
+                    {
+                        fetch = recover_utf8_sequence();
+                    }
+                    break;
+            }
+        }
+    }
+
+    /*!
+    @brief keep the longest valid prefix of a number that scan_number() rejected
+
+    token_buffer holds the characters scan_number() accepted before the error,
+    so the prefix ends at its last digit.
+    */
+    token_type recover_number()
+    {
+        // only size(), operator[], and resize() are used, which every string
+        // type the library supports provides
+        std::size_t length = token_buffer.size();
+        while (length != 0 && (token_buffer[length - 1] < '0' || token_buffer[length - 1] > '9'))
+        {
+            --length;
+        }
+        token_buffer.resize(length);
+
+        if (length == 0)
+        {
+            skip_to_delimiter();
+            return token_type::uninitialized;
+        }
+
+        if (decimal_point_position >= length)
+        {
+            decimal_point_position = std::string::npos;
+        }
+
+        std::size_t exponent = std::string::npos;
+        for (std::size_t i = 0; i < length; ++i)
+        {
+            if (token_buffer[i] == 'e' || token_buffer[i] == 'E')
+            {
+                exponent = i;
+                break;
+            }
+        }
+        const std::size_t mantissa_end = (exponent == std::string::npos) ? length : exponent;
+        token_type number_type = token_type::value_unsigned;
+        if (decimal_point_position != std::string::npos || exponent != std::string::npos)
+        {
+            number_type = token_type::value_float;
+        }
+        else if (token_buffer[0] == '-')
+        {
+            number_type = token_type::value_integer;
+        }
+
+        const token_type result = convert_number(number_type, mantissa_end);
+        skip_to_delimiter();
+        return result;
+    }
+
+    /// seekable or token-tracking adapter: the token string begins at current
+    void restart_token_string_impl(std::true_type /*lazy*/)
+    {
+        restart_token_string_lazy(std::integral_constant<bool, adapter_token> {});
+    }
+
+    /// seekable adapter: the token string begins at current, which was consumed
+    void restart_token_string_lazy(std::false_type /*adapter_token*/) noexcept
+    {
+        const std::size_t consumed = ia.get_consumed_count();
+        token_string_start = (consumed > 0 && current != char_traits<char_type>::eof()) ? consumed - 1 : consumed;
+    }
+
+    /// token-tracking adapter: the token begins with current, or with the next
+    /// character if current was put back or is the end of input
+    void restart_token_string_lazy(std::true_type /*adapter_token*/)
+    {
+        if (!next_unget && current != char_traits<char_type>::eof())
+        {
+            ia.mark_token_start(current);
+        }
+        else
+        {
+            ia.begin_token();
+        }
+    }
+
+    /// streaming adapter: the token string begins at current; a character
+    /// that was put back is copied again when it is read again
+    void restart_token_string_impl(std::false_type /*lazy*/)
+    {
+        token_string.clear();
+        if (!next_unget && current != char_traits<char_type>::eof())
+        {
+            token_string.push_back(char_traits<char_type>::to_char_type(current));
+        }
+    }
+
     /// input adapter
     InputAdapterType ia;
 
@@ -2523,6 +3143,19 @@ scan_number_done:
     /// that unget() can restore the column instead of leaving it at 0
     std::size_t chars_read_before_newline = 0;
 
+    /// seekable adapter: what get_position() counted last, so that the next
+    /// call only scans the input consumed since
+    struct position_cache_t
+    {
+        /// consumed characters scanned so far
+        std::size_t consumed = 0;
+        /// newlines among them
+        std::size_t lines = 0;
+        /// index after the last of these newlines
+        std::size_t line_start = 0;
+    };
+    mutable position_cache_t position_cache {};
+
     /// raw input token string for error messages; only populated for streaming
     /// adapters (seekable adapters reconstruct it lazily via token_string_start)
     std::vector<char_type> token_string {};
@@ -2543,6 +3176,15 @@ scan_number_done:
     /// a description of occurred lexer errors
     const char* error_message = "";
 
+    /// what kind of token scan() rejected; set together with error_message
+    error_kind error_cause = error_kind::other;
+    /// how recover_token() continues a string that scan_string() rejected;
+    /// set only on the error paths that need more than error_message
+    resume_kind string_error_resume = resume_kind::character;
+    /// the code point of the second escape when a high surrogate is followed
+    /// by an escape that is not a low surrogate; -1 otherwise
+    int string_error_codepoint = -1;
+
     // number values
     number_integer_t value_integer = 0;
     number_unsigned_t value_unsigned = 0;

include/nlohmann/detail/input/parser.hpp

@@ -152,25 +152,72 @@ class parser
     bool accept(const bool strict = true)
     {
         json_sax_acceptor<BasicJsonType> sax_acceptor;
-        return sax_parse(&sax_acceptor, strict);
+        return sax_parse_impl<false>(&sax_acceptor, strict);
     }
 
+    /*!
+    @brief public SAX interface
+
+    If the SAX parser's parse_error() returns true, the parser recovers from
+    the error: it repairs the input and continues (see #3989).
+
+    @param[in] sax     the SAX parser
+    @param[in] strict  whether to expect the last token to be EOF
+    @return whether the input was parsed without errors and no SAX event
+            returned false
+    */
     template<typename SAX>
     JSON_HEDLEY_NON_NULL(2)
     bool sax_parse(SAX* sax, const bool strict = true)
     {
-        (void)detail::is_sax_static_asserts<SAX, BasicJsonType> {};
+        return sax_parse_impl<true>(sax, strict);
-        const bool result = sax_parse_internal(sax);
-
-        if (result)
-        {
-            return finish_input(*sax, strict);
-        }
-
-        return false;
     }
 
   private:
+    /// what sax_parse_internal() does after an object key was expected
+    enum class next_step : std::uint8_t
+    {
+        /// stop parsing
+        stop,
+        /// parse a value that begins with last_token
+        parse_value,
+        /// evaluate the state of the innermost container, which reads
+        /// last_token again
+        evaluate_state
+    };
+
+    template<bool AllowRecovery, typename SAX>
+    JSON_HEDLEY_NON_NULL(2)
+    bool sax_parse_impl(SAX* sax, const bool strict)
+    {
+        (void)detail::is_sax_static_asserts<SAX, BasicJsonType> {};
+        const bool result = sax_parse_internal<AllowRecovery>(sax);
+
+        if (result)
+        {
+            if (strict)
+            {
+                // strict mode: next byte must be EOF; after recovering from an
+                // error, the end of the input may already have been read
+                if (last_token != token_type::end_of_input && get_token() != token_type::end_of_input)
+                {
+                    // the value is complete, so there is nothing to recover
+                    static_cast<void>(report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::end_of_input, "value"), nullptr),
+                                                   std::integral_constant<bool, AllowRecovery> {}));
+                    return false;
+                }
+            }
+            else
+            {
+                // the caller keeps using the input: position it right after
+                // the value by leaving the character that terminated it
+                m_lexer.release_lookahead();
+            }
+        }
+
+        return result && !error_reported;
+    }
+
     /// the text of the current float token for @a sax (see sax_ignores_number_text)
     template<typename SAX>
     const string_t& number_float_text(SAX* /*sax*/)
@@ -238,17 +285,26 @@ class parser
     template<typename DomSax>
     bool parse_dom(DomSax& sdp, const bool strict)
     {
-        sax_parse_internal(&sdp);
+        sax_parse_internal<false>(&sdp);
 
         static_cast<void>(finish_input(sdp, strict));
 
         return !sdp.is_errored();
     }
 
-    template<typename SAX>
+    /*!
+    @brief parse a JSON value and pass it to a SAX parser
+
+    @tparam AllowRecovery  whether to recover from an error if the SAX parser's
+                           parse_error() returns true; false for the SAX parsers
+                           of parse() and accept(), which never do, so that no
+                           code for recovering is generated for them
+    */
+    template<bool AllowRecovery, typename SAX>
     JSON_HEDLEY_NON_NULL(2)
     bool sax_parse_internal(SAX* sax)
     {
+        const std::integral_constant<bool, AllowRecovery> allow_recovery{};
         // value to avoid a goto (see comment where set to true)
         bool skip_to_state_evaluation = false;
 
@@ -287,10 +343,18 @@ class parser
                             break;
                         }
 
-                        // parse key
+                        // remember we are now inside an object
+                        states.push_back(0);
+
+                        // parse key (the steps of parse_key(), which are
+                        // repeated here and below for speed)
                         if (JSON_HEDLEY_UNLIKELY(last_token != token_type::value_string))
                         {
-                            return syntax_error(*sax, exception_message(token_type::value_string, "object key"));
+                            if (!continue_after(key_error(sax, allow_recovery, false), skip_to_state_evaluation))
+                            {
+                                return false;
+                            }
+                            continue;
                         }
                         if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string_token())))
                         {
@@ -300,11 +364,12 @@ class parser
                         // parse separator (:)
                         if (JSON_HEDLEY_UNLIKELY(!get_token_expecting(token_type::name_separator)))
                         {
-                            return syntax_error(*sax, exception_message(token_type::name_separator, "object separator"));
+                            if (!continue_after(key_error(sax, allow_recovery, true), skip_to_state_evaluation))
+                            {
+                                return false;
+                            }
+                            continue;
                         }
-
-                        // remember we are now inside an object
-                        states.push_back(0);
 
                         // parse values
                         get_token();
@@ -341,9 +406,11 @@ class parser
 
                         if (JSON_HEDLEY_UNLIKELY(!std::isfinite(res)))
                         {
-                            return sax->parse_error(m_lexer.get_position(),
+                            if (!overflow_error(sax, res, allow_recovery))
-                                                    m_lexer.get_token_string(),
+                            {
-                                                    out_of_range::create(exception_id::number_overflow, concat("number overflow parsing '", m_lexer.get_token_string(), '\''), nullptr));
+                                return false;
+                            }
+                            break;
                         }
 
                         if (JSON_HEDLEY_UNLIKELY(!sax->number_float(res, number_float_text(sax))))
@@ -411,16 +478,63 @@ class parser
                     case token_type::parse_error:
                     {
                         // using "uninitialized" to avoid an "expected" message
-                        return syntax_error(*sax, exception_message(token_type::uninitialized, "value"));
+                        if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::uninitialized, "value"), nullptr), allow_recovery))
+                        {
+                            return false;
+                        }
+
+                        // recover: keep what can be read of the token
+                        recover_token(allow_recovery);
+                        if (last_token != token_type::uninitialized)
+                        {
+                            // a string or a number
+                            continue;
+                        }
+                        if (states.empty())
+                        {
+                            // look for the value after the garbage
+                            if (!skip_to_value(allow_recovery))
+                            {
+                                return false;
+                            }
+                            continue;
+                        }
+                        // nothing could be read
+                        if (JSON_HEDLEY_UNLIKELY(!sax->null()))
+                        {
+                            return false;
+                        }
+                        break;
                     }
                     case token_type::end_of_input:
                     {
                         if (JSON_HEDLEY_UNLIKELY(m_lexer.get_chars_read_total() == 1))
                         {
-                            return syntax_error(*sax, empty_input_message());
+                            // there is nothing to recover
+                            static_cast<void>(report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(),
+                                                           empty_input_message(), nullptr), allow_recovery));
+                            return false;
                         }
 
-                        return syntax_error(*sax, exception_message(token_type::literal_or_value, "value"));
+                        if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::literal_or_value, "value"), nullptr), allow_recovery))
+                        {
+                            return false;
+                        }
+
+                        // recover: the input ends where a value is missing
+                        if (states.empty())
+                        {
+                            // there is no value
+                            return false;
+                        }
+                        if (!recover_missing_value(sax, states, allow_recovery))
+                        {
+                            return false;
+                        }
+                        // the state evaluation reads the token again
+                        unget_token(allow_recovery);
+                        skip_to_state_evaluation = true;
+                        continue;
                     }
                     case token_type::uninitialized:
                     case token_type::end_array:
@@ -430,7 +544,35 @@ class parser
                     case token_type::literal_or_value:
                     default: // the last token was unexpected
                     {
-                        return syntax_error(*sax, exception_message(token_type::literal_or_value, "value"));
+                        if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::literal_or_value, "value"), nullptr), allow_recovery))
+                        {
+                            return false;
+                        }
+
+                        // recover
+                        if (states.empty())
+                        {
+                            // look for the value after the garbage
+                            if (!skip_to_value(allow_recovery))
+                            {
+                                return false;
+                            }
+                            continue;
+                        }
+                        if (last_token == token_type::name_separator)
+                        {
+                            // a stray ':'; the value may follow
+                            get_token();
+                            continue;
+                        }
+                        if (!recover_missing_value(sax, states, allow_recovery))
+                        {
+                            return false;
+                        }
+                        // the state evaluation reads the token again
+                        unget_token(allow_recovery);
+                        skip_to_state_evaluation = true;
+                        continue;
                     }
                 }
             }
@@ -481,7 +623,30 @@ class parser
                     continue;
                 }
 
-                return syntax_error(*sax, exception_message(token_type::end_array, "array"));
+                if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::end_array, "array"), nullptr), allow_recovery))
+                {
+                    return false;
+                }
+
+                // recover
+                if (last_token == token_type::end_of_input)
+                {
+                    // the input ends inside the array
+                    return close_containers(sax, states, allow_recovery);
+                }
+                if (last_token == token_type::end_object)
+                {
+                    // a wrong closing bracket closes the innermost container
+                    if (JSON_HEDLEY_UNLIKELY(!sax->end_array()))
+                    {
+                        return false;
+                    }
+                    states.pop_back();
+                    skip_to_state_evaluation = true;
+                }
+                // otherwise, a missing ',' (or a stray ':', which value
+                // parsing drops): the next value begins here
+                continue;
             }
 
             // states.back() is false -> object
@@ -498,9 +663,12 @@ class parser
                     // parse key
                     if (JSON_HEDLEY_UNLIKELY(last_token != token_type::value_string))
                     {
-                        return syntax_error(*sax, exception_message(token_type::value_string, "object key"));
+                        if (!continue_after(key_error(sax, allow_recovery, false), skip_to_state_evaluation))
+                        {
+                            return false;
+                        }
+                        continue;
                     }
-
                     if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string_token())))
                     {
                         return false;
@@ -509,7 +677,11 @@ class parser
                     // parse separator (:)
                     if (JSON_HEDLEY_UNLIKELY(!get_token_expecting(token_type::name_separator)))
                     {
-                        return syntax_error(*sax, exception_message(token_type::name_separator, "object separator"));
+                        if (!continue_after(key_error(sax, allow_recovery, true), skip_to_state_evaluation))
+                        {
+                            return false;
+                        }
+                        continue;
                     }
 
                     // parse values
@@ -537,8 +709,544 @@ class parser
                 continue;
             }
 
-            return syntax_error(*sax, exception_message(token_type::end_object, "object"));
+            if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::end_object, "object"), nullptr), allow_recovery))
+            {
+                return false;
             }
+
+            // recover
+            if (last_token == token_type::end_of_input)
+            {
+                // the input ends inside the object
+                return close_containers(sax, states, allow_recovery);
+            }
+            if (last_token == token_type::end_array)
+            {
+                // a wrong closing bracket closes the innermost container
+                if (JSON_HEDLEY_UNLIKELY(!sax->end_object()))
+                {
+                    return false;
+                }
+                states.pop_back();
+                skip_to_state_evaluation = true;
+                continue;
+            }
+            if (!continue_after(recover_member(sax, allow_recovery), skip_to_state_evaluation))
+            {
+                return false;
+            }
+        }
+    }
+
+    /*!
+    @brief continue sax_parse_internal() after a recovery
+    @return whether to continue parsing
+    */
+    bool continue_after(const next_step step, bool& skip_to_state_evaluation)
+    {
+        if (step == next_step::evaluate_state)
+        {
+            // the state evaluation reads the token again
+            m_lexer.unget_token();
+            skip_to_state_evaluation = true;
+        }
+        return step != next_step::stop;
+    }
+
+    /// the parser for parse() and accept() never recovers: stop parsing
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type continue_after(std::false_type /*step*/, bool& /*skip_to_state_evaluation*/) noexcept
+    {
+        return {};
+    }
+
+    /*!
+    @brief parse an object key and the name separator (:) after it
+
+    last_token is the token where the key is expected. sax_parse_internal()
+    repeats these steps rather than calling this function, which is used
+    when recovering from an error.
+
+    @return next_step::parse_value if the value follows, with last_token its
+            first token; next_step::evaluate_state if the object's state is
+            to be evaluated after recovering from an error; next_step::stop
+            to stop parsing
+    */
+    template<typename SAX>
+    next_step parse_key(SAX* sax)
+    {
+        const std::true_type allow_recovery{};
+
+        if (JSON_HEDLEY_UNLIKELY(last_token != token_type::value_string))
+        {
+            return key_error(sax, allow_recovery, false);
+        }
+
+        if (JSON_HEDLEY_UNLIKELY(!sax->key(m_lexer.get_string_token())))
+        {
+            return next_step::stop;
+        }
+
+        // parse separator (:)
+        if (JSON_HEDLEY_UNLIKELY(get_token() != token_type::name_separator))
+        {
+            return key_error(sax, allow_recovery, true);
+        }
+
+        // the value begins with the next token
+        get_token();
+        return next_step::parse_value;
+    }
+
+    /*!
+    @brief report a number that is too large for number_float_t, and recover
+           from the error by passing the value on; the SAX parser gets the
+           number's text as well
+
+    This is a separate function, as reading other numbers is measurably
+    slower if the error is handled where they are read.
+
+    @param[in] sax    the SAX parser
+    @param[in] value  the value that is not finite
+    @return whether to continue parsing
+    */
+    template<typename SAX>
+    bool overflow_error(SAX* sax, const number_float_t value, std::true_type allow_recovery)
+    {
+        if (!report_error(sax, out_of_range::create(exception_id::number_overflow, concat("number overflow parsing '", m_lexer.get_token_string(), '\''), nullptr), allow_recovery))
+        {
+            return false;
+        }
+        return sax->number_float(value, number_float_text(sax));
+    }
+
+    /// @copydoc overflow_error
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    std::false_type overflow_error(SAX* sax, const number_float_t /*value*/, std::false_type allow_recovery)
+    {
+        return report_error(sax, out_of_range::create(exception_id::number_overflow, concat("number overflow parsing '", m_lexer.get_token_string(), '\''), nullptr), allow_recovery);
+    }
+
+    /*!
+    @brief report a missing key, or a missing name separator (:) after the
+           key; the parser for parse() and accept() never recovers
+
+    @param[in] key_read  whether the key was read, so that the name separator
+                         is missing
+    @return std::false_type, see report_error()
+    */
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    std::false_type key_error(SAX* sax, std::false_type allow_recovery, const bool key_read)
+    {
+        return report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), key_read
+                            ? exception_message(token_type::name_separator, "object separator")
+                            : exception_message(token_type::value_string, "object key"), nullptr), allow_recovery);
+    }
+
+    /*!
+    @brief report a missing key, or a missing name separator (:) after the
+           key, and recover from it
+
+    @param[in] key_read  whether the key was read, so that the name separator
+                         is missing
+    */
+    template<typename SAX>
+    next_step key_error(SAX* sax, std::true_type allow_recovery, const bool key_read)
+    {
+        if (!key_read)
+        {
+            if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::value_string, "object key"), nullptr), allow_recovery))
+            {
+                return next_step::stop;
+            }
+            return recover_key(sax);
+        }
+
+        if (!report_error(sax, parse_error::create(exception_id::syntax_error, m_lexer.get_position(), exception_message(token_type::name_separator, "object separator"), nullptr), allow_recovery))
+        {
+            return next_step::stop;
+        }
+        return recover_name_separator(sax);
+    }
+
+    /////////////////////
+    // error recovery
+    /////////////////////
+
+    /*
+    The functions below repair an error after the SAX parser's parse_error()
+    returned true (see #3989). Each mistake is repaired by the smallest local
+    edit: a missing ',' or ':' is inserted, a stray token is removed, what can
+    be read of an invalid string or number is kept (see
+    lexer::recover_token()), a missing value becomes null, a wrong closing
+    bracket closes the innermost container, and the end of the input closes
+    all of them. The events stay balanced, and every key() is followed by
+    exactly one value.
+
+    A repair hands a token to the state evaluation, by returning it to the
+    lexer (lexer::unget_token()) so that the state evaluation reads it again,
+    only if it is ',', ']', '}', or the end of the input. The state evaluation
+    hands a token to value or key parsing only if it is none of them, so a
+    token is never handed back and forth. Every other step reads a token or
+    closes a container, so parsing always ends.
+    */
+
+    /*!
+    @brief report an error to the SAX parser; the parser for parse() and
+           accept() never recovers
+
+    @return std::false_type rather than false: its value is known where the
+            function is called even if the call is not inlined, so the code
+            for recovering is not generated
+    */
+    template<typename SAX, typename Exception>
+    JSON_INTERNAL_ALWAYS_INLINE
+    std::false_type report_error(SAX* sax, const Exception& ex, std::false_type /*allow_recovery*/)
+    {
+        static_cast<void>(sax); // MSVC 2015 does not count calling a static parse_error() as using it
+        error_reported = true;
+        static_cast<void>(sax->parse_error(m_lexer.get_position(), m_lexer.get_token_string(), ex));
+        return {};
+    }
+
+    /*!
+    @brief report an error to the SAX parser
+    @return whether to recover from the error
+    */
+    template<typename SAX, typename Exception>
+    bool report_error(SAX* sax, const Exception& ex, std::true_type /*allow_recovery*/)
+    {
+        static_cast<void>(sax); // MSVC 2015 does not count calling a static parse_error() as using it
+        const std::size_t position = m_lexer.get_position().chars_read_total;
+        if (error_reported && position == last_error_position && last_token == last_error_token)
+        {
+            // a repair handed on the token of the error it repaired; the
+            // token was reported already, and the SAX parser asked to recover
+            return true;
+        }
+
+        error_reported = true;
+        last_error_position = position;
+        last_error_token = last_token;
+
+        if (!sax->parse_error(m_lexer.get_position(), m_lexer.get_token_string(), ex))
+        {
+            return false;
+        }
+
+        // the token string of the next error begins here
+        m_lexer.restart_token_string();
+        return true;
+    }
+
+    /*!
+    @brief keep what can be read of the token that the lexer rejected
+
+    The error was reported for the rejected token, so it is not reported again
+    for the token it is repaired to (see lexer::recover_token()).
+    */
+    token_type recover_token()
+    {
+        last_token = m_lexer.recover_token();
+        last_error_position = m_lexer.get_position().chars_read_total;
+        last_error_token = last_token;
+        return last_token;
+    }
+
+    // The functions below are called from sax_parse_internal() after an error
+    // was reported. The parser for parse() and accept() stopped there, so for
+    // it they are stubs: the code for recovering is then not even referenced,
+    // and unoptimized builds do not emit it either.
+
+    /// @copydoc recover_token()
+    JSON_INTERNAL_ALWAYS_INLINE
+    token_type recover_token(std::true_type /*allow_recovery*/)
+    {
+        return recover_token();
+    }
+
+    /// @copydoc recover_token()
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type recover_token(std::false_type /*allow_recovery*/) noexcept
+    {
+        return {};
+    }
+
+    /// return the token that was read last to the lexer (see lexer::unget_token())
+    JSON_INTERNAL_ALWAYS_INLINE
+    void unget_token(std::true_type /*allow_recovery*/)
+    {
+        m_lexer.unget_token();
+    }
+
+    /// @copydoc unget_token(std::true_type)
+    JSON_INTERNAL_ALWAYS_INLINE
+    static void unget_token(std::false_type /*allow_recovery*/) noexcept {}
+
+    /// @copydoc skip_to_value
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type skip_to_value(std::false_type /*allow_recovery*/) noexcept
+    {
+        return {};
+    }
+
+    /// @copydoc recover_missing_value
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type recover_missing_value(SAX* /*sax*/, const std::vector<std::uint8_t>& /*states*/, std::false_type /*allow_recovery*/) noexcept
+    {
+        return {};
+    }
+
+    /// pass the end events of all open containers
+    template<typename SAX>
+    bool close_containers(SAX* sax, std::vector<std::uint8_t>& states, std::true_type /*allow_recovery*/)
+    {
+        while (!states.empty())
+        {
+            const bool is_array = states.back() != 0;
+            states.pop_back();
+            if (JSON_HEDLEY_UNLIKELY(is_array ? !sax->end_array() : !sax->end_object()))
+            {
+                return false;
+            }
+        }
+        return true;
+    }
+
+    /// @copydoc close_containers
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    static std::false_type close_containers(SAX* /*sax*/, std::vector<std::uint8_t>& /*states*/, std::false_type /*allow_recovery*/) noexcept
+    {
+        return {};
+    }
+
+    /*!
+    @brief read tokens until one begins a value, skipping everything before
+           the top-level value
+    @return whether a value begins with last_token
+    */
+    bool skip_to_value(std::true_type /*allow_recovery*/)
+    {
+        while (true)
+        {
+            switch (get_token())
+            {
+                case token_type::begin_array:
+                case token_type::begin_object:
+                case token_type::literal_false:
+                case token_type::literal_null:
+                case token_type::literal_true:
+                case token_type::value_float:
+                case token_type::value_integer:
+                case token_type::value_string:
+                case token_type::value_unsigned:
+                    return true;
+
+                case token_type::end_of_input:
+                    return false;
+
+                case token_type::parse_error:
+                    recover_token();
+                    if (last_token != token_type::uninitialized)
+                    {
+                        return true;
+                    }
+                    break;
+
+                case token_type::uninitialized:
+                case token_type::end_array:
+                case token_type::end_object:
+                case token_type::name_separator:
+                case token_type::value_separator:
+                case token_type::literal_or_value:
+                default:
+                    break;
+            }
+        }
+    }
+
+    /*!
+    @brief skip the rest of an object member that cannot be read
+
+    Reads tokens, beginning with last_token, until a ',', '}', or ']' that is
+    not inside a container that begins in the skipped tokens, or the end of
+    the input.
+    */
+    void skip_member()
+    {
+        std::size_t depth = 0;
+        while (true)
+        {
+            switch (last_token)
+            {
+                case token_type::begin_array:
+                case token_type::begin_object:
+                    ++depth;
+                    break;
+
+                case token_type::end_array:
+                case token_type::end_object:
+                    if (depth == 0)
+                    {
+                        return;
+                    }
+                    --depth;
+                    break;
+
+                case token_type::value_separator:
+                    if (depth == 0)
+                    {
+                        return;
+                    }
+                    break;
+
+                case token_type::end_of_input:
+                    return;
+
+                case token_type::parse_error:
+                    recover_token();
+                    break;
+
+                case token_type::uninitialized:
+                case token_type::literal_true:
+                case token_type::literal_false:
+                case token_type::literal_null:
+                case token_type::value_string:
+                case token_type::value_unsigned:
+                case token_type::value_integer:
+                case token_type::value_float:
+                case token_type::name_separator:
+                case token_type::literal_or_value:
+                default:
+                    break;
+            }
+            get_token();
+        }
+    }
+
+    /*!
+    @brief pass a value where it is missing
+
+    last_token is ',', ']', '}', or the end of the input, where a value was
+    expected. In an object, the key gets null; in an array, a ',' where a
+    value is missing stands for null (as in JavaScript), while an array that
+    ends there just ends.
+    */
+    template<typename SAX>
+    bool recover_missing_value(SAX* sax, const std::vector<std::uint8_t>& states, std::true_type /*allow_recovery*/)
+    {
+        JSON_ASSERT(!states.empty());
+        if (states.back() == 0 || last_token == token_type::value_separator)
+        {
+            return sax->null();
+        }
+        return true;
+    }
+
+    /// recover from a missing key; last_token is where it was expected
+    template<typename SAX>
+    next_step recover_key(SAX* sax)
+    {
+        switch (last_token)
+        {
+            case token_type::value_separator:
+            case token_type::end_object:
+            case token_type::end_array:
+            case token_type::end_of_input:
+                // no member: the object's state handles the token
+                return next_step::evaluate_state;
+
+            case token_type::parse_error:
+                recover_token();
+                if (last_token == token_type::value_string)
+                {
+                    // a key that could be repaired
+                    return parse_key(sax);
+                }
+                skip_member();
+                return next_step::evaluate_state;
+
+            case token_type::uninitialized:
+            case token_type::literal_true:
+            case token_type::literal_false:
+            case token_type::literal_null:
+            case token_type::value_string:
+            case token_type::value_unsigned:
+            case token_type::value_integer:
+            case token_type::value_float:
+            case token_type::begin_array:
+            case token_type::begin_object:
+            case token_type::name_separator:
+            case token_type::literal_or_value:
+            default:
+                // a member without a key
+                skip_member();
+                return next_step::evaluate_state;
+        }
+    }
+
+    /// recover from a missing name separator (:) after the key; last_token
+    /// is where it was expected
+    template<typename SAX>
+    next_step recover_name_separator(SAX* sax)
+    {
+        switch (last_token)
+        {
+            case token_type::value_separator:
+            case token_type::end_object:
+            case token_type::end_array:
+            case token_type::end_of_input:
+                // the value is missing as well
+                return sax->null() ? next_step::evaluate_state : next_step::stop;
+
+            case token_type::uninitialized:
+            case token_type::literal_true:
+            case token_type::literal_false:
+            case token_type::literal_null:
+            case token_type::value_string:
+            case token_type::value_unsigned:
+            case token_type::value_integer:
+            case token_type::value_float:
+            case token_type::begin_array:
+            case token_type::begin_object:
+            case token_type::name_separator:
+            case token_type::parse_error:
+            case token_type::literal_or_value:
+            default:
+                // a missing ':'; the value begins here
+                return next_step::parse_value;
+        }
+    }
+
+    /// recover from a token after an object member that is neither ',' nor
+    /// '}' (nor ']' or the end of the input, which the caller handles)
+    template<typename SAX>
+    next_step recover_member(SAX* sax, std::true_type /*allow_recovery*/)
+    {
+        if (last_token == token_type::parse_error)
+        {
+            recover_token();
+        }
+        if (last_token == token_type::value_string)
+        {
+            // a missing ','; the next key begins here
+            return parse_key(sax);
+        }
+        skip_member();
+        return next_step::evaluate_state;
+    }
+
+    /// the parser for parse() and accept() never recovers (and does not come
+    /// here, as report_error() returned false)
+    template<typename SAX>
+    JSON_INTERNAL_ALWAYS_INLINE
+    std::false_type recover_member(SAX* /*sax*/, std::false_type /*allow_recovery*/) const noexcept
+    {
+        return {};
     }
 
     /// get next token from lexer
@@ -604,6 +1312,12 @@ class parser
     const bool allow_exceptions = true;
     /// whether trailing commas in objects and arrays should be ignored (true) or signaled as errors (false)
     const bool ignore_trailing_commas = false;
+    /// whether an error was reported to the SAX parser
+    bool error_reported = false;
+    /// the position of the last reported error
+    std::size_t last_error_position = 0;
+    /// the token of the last reported error
+    token_type last_error_token = token_type::uninitialized;
 };
 
 }  // namespace detail

include/nlohmann/detail/macro_scope.hpp

@@ -215,6 +215,17 @@
     #define JSON_NO_UNIQUE_ADDRESS
 #endif
 
+// Inlines small functions even in unoptimized builds, so that they are not
+// emitted. The parsers for parse(), accept(), and from_*() use it for the
+// functions that stand in for the code recovering from errors (see #3989).
+// MSVC is left to decide, as it warns (C4714) where it does not inline a
+// __forceinline function.
+#if defined(_MSC_VER) && !defined(__clang__)
+    #define JSON_INTERNAL_ALWAYS_INLINE
+#else
+    #define JSON_INTERNAL_ALWAYS_INLINE JSON_HEDLEY_ALWAYS_INLINE
+#endif
+
 // Clang targeting MinGW does not survive the thread_local storage the copy
 // constructor uses to bound its descent: every test that copies a value
 // segfaults with clang 11.0.1 and clang 18.1.8, while the same tests pass with

include/nlohmann/detail/macro_unscope.hpp

@@ -19,9 +19,11 @@
 #undef NLOHMANN_CAN_CALL_STD_FUNC_IMPL
 #undef JSON_INLINE_VARIABLE
 #undef JSON_NO_UNIQUE_ADDRESS
+#undef JSON_INTERNAL_ALWAYS_INLINE
 #undef JSON_DISABLE_ENUM_SERIALIZATION
 #undef JSON_DISABLE_TUPLE_REFERENCE_CONVERSION
 #undef JSON_UNLOCKED_FILE_IO
+#undef JSON_FILE_WINDOW
 #undef JSON_DTOA_SSE2
 #undef JSON_DTOA_NEON
 #undef JSON_LITTLE_ENDIAN_TARGET

include/nlohmann/detail/output/binary_writer.hpp

@@ -3126,22 +3126,19 @@ class binary_writer
                 // negative fixnum
                 write_number(static_cast<std::int8_t>(j.m_data.m_value.number_integer));
             }
-            else if (j.m_data.m_value.number_integer >= (std::numeric_limits<std::int8_t>::min)() &&
+            else if (j.m_data.m_value.number_integer >= (std::numeric_limits<std::int8_t>::min)())
-                     j.m_data.m_value.number_integer <= (std::numeric_limits<std::int8_t>::max)())
             {
                 // int 8
                 oa.write_character(to_char_type(0xD0));
                 write_number(static_cast<std::int8_t>(j.m_data.m_value.number_integer));
             }
-            else if (j.m_data.m_value.number_integer >= (std::numeric_limits<std::int16_t>::min)() &&
+            else if (j.m_data.m_value.number_integer >= (std::numeric_limits<std::int16_t>::min)())
-                     j.m_data.m_value.number_integer <= (std::numeric_limits<std::int16_t>::max)())
             {
                 // int 16
                 oa.write_character(to_char_type(0xD1));
                 write_number(static_cast<std::int16_t>(j.m_data.m_value.number_integer));
             }
-            else if (j.m_data.m_value.number_integer >= (std::numeric_limits<std::int32_t>::min)() &&
+            else if (j.m_data.m_value.number_integer >= (std::numeric_limits<std::int32_t>::min)())
-                     j.m_data.m_value.number_integer <= (std::numeric_limits<std::int32_t>::max)())
             {
                 // int 32
                 oa.write_character(to_char_type(0xD2));
@@ -4057,7 +4054,7 @@ class binary_writer
     Containers, strings, high-precision numbers, booleans and null cannot be
     declared as the single type of an optimized container in BJData; such a
     container is written unoptimized. The reader rejects them with the same
-    list (binary_reader::is_bjd_excluded_optimized_type()).
+    list (binary_reader::is_valid_optimized_type()).
     */
     static constexpr bool is_bjdata_excluded_type_marker(const CharType marker) noexcept
     {

include/nlohmann/detail/output/output_adapters.hpp

@@ -116,6 +116,10 @@ inline void copy_short(CharType* dst, const CharType* src, const std::size_t len
 /// the number of characters an empty vector is grown to by output_vector_sink
 constexpr std::size_t min_vector_sink_capacity = 64;
 
+/// the number of characters output_vector_sink makes writable ahead of its
+/// position at a time: 64 KiB of the vector's capacity, not all of it
+constexpr std::size_t vector_sink_window = 65536;
+
 /// @brief non-virtual output sink writing into a std::vector
 ///
 /// This sink is not part of the virtual output_adapter_protocol hierarchy: it is
@@ -129,7 +133,8 @@ constexpr std::size_t min_vector_sink_capacity = 64;
 /// encoding them: insert() is an out-of-line call into a routine that can insert
 /// anywhere, and resize() is one as well with libstdc++. The sink therefore keeps
 /// its own write position: it grows the vector geometrically ahead of the
-/// position and copies each write to it, so a write is a capacity check and a
+/// position (making @ref vector_sink_window characters of it writable at a
+/// time) and copies each write to it, so a write is a capacity check and a
 /// short copy. The destructor shrinks the vector to the bytes actually written,
 /// also when a write throws; shrinking a vector never allocates. The sink owns
 /// that trim, so it can be moved but not copied, and the vector must not be used
@@ -277,9 +282,17 @@ class output_vector_sink
         // reallocate for every few bytes
         const std::size_t written = position();
         const std::size_t needed = written + length;
-        v->resize((std::max)(min_vector_sink_capacity, needed <= v->capacity()
+        const std::size_t capacity = (std::max)(min_vector_sink_capacity, needed <= v->capacity()
                                                 ? v->capacity()
-                             : (std::max)(needed, v->capacity() * 2)));
+                                                : (std::max)(needed, v->capacity() * 2));
+        v->reserve(capacity);
+        // resize() zeroes the characters it adds. Zeroing all of a doubled
+        // capacity, megabytes of a fresh allocation, at once faults in its
+        // pages and evicts them from the cache before the writer gets there,
+        // so its stores miss again; a window ahead of the position is still
+        // cached when they come (the binary writers on a long array of
+        // numbers: 4-26 % faster on x86, 1-2.5 % on Apple M1)
+        v->resize((std::min)(capacity, (std::max)(needed, written + vector_sink_window)));
         cursor = v->data() + written;
         limit = v->data() + v->size();
     }

include/nlohmann/detail/output/serializer.hpp

@@ -8,7 +8,7 @@
 
 #pragma once
 
-#include <algorithm> // remove, fill, find, none_of, min
+#include <algorithm> // copy, none_of, min, search
 #include <array> // array
 #include <clocale> // localeconv, lconv
 #include <cmath> // isfinite
@@ -2080,31 +2080,35 @@ class serializer
         // check if the buffer was large enough
         JSON_ASSERT(static_cast<std::size_t>(len) < number_buffer.size());
 
-        // look up the locale's thousands separator and decimal point now,
+        // look up the locale's decimal point now, matching what
-        // matching what snprintf_float() just used (see lexer::get_decimal_point())
+        // snprintf_float() just used (see get_decimal_point() in
+        // number_parse.hpp); it is a string, which can be longer than one
+        // byte (U+066B in ps_AF, #5993).
+        // A thousands separator needs no handling: %g never groups digits.
         const auto* loc = std::localeconv();
         JSON_ASSERT(loc != nullptr);
-        const char thousands_sep = (loc->thousands_sep == nullptr) ? '\0' : *loc->thousands_sep;
+        const char* const decimal_point = (loc->decimal_point == nullptr) ? "" : loc->decimal_point;
-        const char decimal_point = (loc->decimal_point == nullptr) ? '\0' : *loc->decimal_point;
+        const std::size_t decimal_point_len = std::char_traits<char>::length(decimal_point);
-
-        // erase thousands separators
-        if (thousands_sep != '\0')
-        {
-            // NOLINTNEXTLINE(readability-qualified-auto,llvm-qualified-auto): std::remove returns an iterator, see https://github.com/nlohmann/json/issues/3081
-            const auto end = std::remove(number_buffer.begin(), number_buffer.begin() + len, thousands_sep);
-            std::fill(end, number_buffer.end(), '\0');
-            JSON_ASSERT((end - number_buffer.begin()) <= len);
-            len = (end - number_buffer.begin());
-        }
 
         // convert decimal point to '.'
-        if (decimal_point != '\0' && decimal_point != '.')
+        if ((decimal_point_len > 1) || (decimal_point_len == 1 && decimal_point[0] != '.'))
         {
-            // NOLINTNEXTLINE(readability-qualified-auto,llvm-qualified-auto): std::find returns an iterator, see https://github.com/nlohmann/json/issues/3081
+            const auto first = number_buffer.begin();
-            const auto dec_pos = std::find(number_buffer.begin(), number_buffer.end(), decimal_point);
+            const auto last = first + len;
-            if (dec_pos != number_buffer.end())
+            // NOLINTNEXTLINE(readability-qualified-auto,llvm-qualified-auto): std::search returns an iterator, see https://github.com/nlohmann/json/issues/3081
+            const auto dec_pos = std::search(first, last, decimal_point, decimal_point + decimal_point_len);
+            if (dec_pos != last)
             {
                 *dec_pos = '.';
+                if (decimal_point_len > 1)
+                {
+                    // move the fraction and exponent behind the '.' (std::copy
+                    // allows the overlap, as the destination is before the source)
+                    // NOLINTNEXTLINE(readability-qualified-auto,llvm-qualified-auto): std::copy returns an iterator, see https://github.com/nlohmann/json/issues/3081
+                    const auto end = std::copy(dec_pos + static_cast<std::ptrdiff_t>(decimal_point_len), last, dec_pos + 1);
+                    *end = '\0';
+                    len = end - first;
+                }
             }
         }
 

include/nlohmann/detail/string_utils.hpp

@@ -13,6 +13,7 @@
 #include <cstddef> // size_t
 #include <cstdint> // uint8_t, uint32_t
 #include <string> // string, to_string
+#include <utility> // move
 
 #include <nlohmann/detail/abi_macros.hpp>
 #include <nlohmann/detail/macro_scope.hpp>

include/nlohmann/detail/view/builder.hpp

@@ -1168,20 +1168,20 @@ indent_done:
             if (negative)
             {
                 ++p;
+                if (NLOHMANN_VIEW_UNLIKELY(cur() < '0' || cur() > '9'))
+                {
+                    return fail(error_code::number_after_minus);
+                }
             }
             const unsigned char* const int_start = p;
             if (cur() == '0')
             {
                 ++p;
             }
-            else if (!negative || NLOHMANN_VIEW_LIKELY(cur() >= '1' && cur() <= '9'))
-            {
-                // (without a sign, the dispatch saw a digit)
-                p = skip_digits(p + 1, e);
-            }
             else
             {
-                return fail(error_code::number_after_minus);
+                // (a digit: the dispatch or the check after the sign saw it)
+                p = skip_digits(p + 1, e);
             }
             const auto int_digits = static_cast<std::size_t>(p - int_start);
             std::size_t frac_digits = 0;

include/nlohmann/detail/view/errors.hpp

@@ -251,8 +251,12 @@ std::string diagnostics(const error_site site)
     std::string context;
 #if JSON_DIAGNOSTICS
     std::vector<std::string> tokens;
-    // a value that is not found leaves no tokens, like the root
+    const bool found = edited ? value_path<true>(d, site.value(), tokens) : value_path<false>(d, site.value(), tokens);
-    static_cast<void>(edited ? value_path<true>(d, site.value(), tokens) : value_path<false>(d, site.value(), tokens));
+    if (!found)
+    {
+        // a value that is not found has no path, and with it no bytes
+        return context;
+    }
     if (!tokens.empty())
     {
         // the root value has no path, but still has a byte range
@@ -449,11 +453,21 @@ struct throwing_sax
 
     /// (a template, as in json_sax_dom_parser: the parser passes the concrete
     /// exception type, which is thrown as is)
+    ///
+    /// It is not [[noreturn]]: sax_parse() recovers from an error if
+    /// parse_error() returns true, and MSVC reports that code as unreachable
+    /// (C4702) for a handler that never returns. The flag hides that from it.
     template<class Exception>
-    [[noreturn]] bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const Exception& ex)
+    bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const Exception& ex)
+    {
+        if (throwing)
         {
             NLOHMANN_VIEW_THROW(ex);
         }
+        return false; // LCOV_EXCL_LINE
+    }
+
+    bool throwing = true;
 };
 
 /*!

include/nlohmann/json.hpp

@@ -157,7 +157,7 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
     friend class ::nlohmann::detail::iter_impl;
     template<typename BasicJsonType, typename CharType, typename OutputSinkType>
     friend class ::nlohmann::detail::binary_writer;
-    template<typename BasicJsonType, typename InputType, typename SAX>
+    template<typename BasicJsonType, typename InputType, typename SAX, bool AllowRecovery>
     friend class ::nlohmann::detail::binary_reader;
     template<typename BasicJsonType, typename InputAdapterType, bool Collect, typename LexerType>
     friend class ::nlohmann::detail::json_sax_dom_parser;
@@ -3137,7 +3137,8 @@ public:
         {
             return parser(std::move(ia), nullptr, true, ignore_comments, ignore_trailing_commas).sax_parse(sax, strict);
         }
-        return detail::binary_reader<basic_json, InputAdapterType, SAX>(std::move(ia), format).sax_parse(sax, strict, tag_handler);
+        // the SAX parser may recover from errors (see parse_error())
+        return detail::binary_reader<basic_json, InputAdapterType, SAX, true>(std::move(ia), format).sax_parse(sax, strict, tag_handler);
     }
 
     /// serialize @a j into a new vector; @a write receives the vector-backed writer

single_include/nlohmann/json_view.hpp

@@ -2473,20 +2473,20 @@ indent_done:
             if (negative)
             {
                 ++p;
+                if (NLOHMANN_VIEW_UNLIKELY(cur() < '0' || cur() > '9'))
+                {
+                    return fail(error_code::number_after_minus);
+                }
             }
             const unsigned char* const int_start = p;
             if (cur() == '0')
             {
                 ++p;
             }
-            else if (!negative || NLOHMANN_VIEW_LIKELY(cur() >= '1' && cur() <= '9'))
-            {
-                // (without a sign, the dispatch saw a digit)
-                p = skip_digits(p + 1, e);
-            }
             else
             {
-                return fail(error_code::number_after_minus);
+                // (a digit: the dispatch or the check after the sign saw it)
+                p = skip_digits(p + 1, e);
             }
             const auto int_digits = static_cast<std::size_t>(p - int_start);
             std::size_t frac_digits = 0;
@@ -5080,8 +5080,12 @@ std::string diagnostics(const error_site site)
     std::string context;
 #if JSON_DIAGNOSTICS
     std::vector<std::string> tokens;
-    // a value that is not found leaves no tokens, like the root
+    const bool found = edited ? value_path<true>(d, site.value(), tokens) : value_path<false>(d, site.value(), tokens);
-    static_cast<void>(edited ? value_path<true>(d, site.value(), tokens) : value_path<false>(d, site.value(), tokens));
+    if (!found)
+    {
+        // a value that is not found has no path, and with it no bytes
+        return context;
+    }
     if (!tokens.empty())
     {
         // the root value has no path, but still has a byte range
@@ -5278,11 +5282,21 @@ struct throwing_sax
 
     /// (a template, as in json_sax_dom_parser: the parser passes the concrete
     /// exception type, which is thrown as is)
+    ///
+    /// It is not [[noreturn]]: sax_parse() recovers from an error if
+    /// parse_error() returns true, and MSVC reports that code as unreachable
+    /// (C4702) for a handler that never returns. The flag hides that from it.
     template<class Exception>
-    [[noreturn]] bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const Exception& ex)
+    bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const Exception& ex)
+    {
+        if (throwing)
         {
             NLOHMANN_VIEW_THROW(ex);
         }
+        return false; // LCOV_EXCL_LINE
+    }
+
+    bool throwing = true;
 };
 
 /*!

tests/src/fuzzer-parse_bjdata.cpp

@@ -47,6 +47,10 @@ dumps is stable under exactly the same values that break operator==.
 The unit tests run the same checks on a fixed corpus (see the "BJData round-trip
 invariants" test case), so keep both in sync.
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_bjdata() fails (see #3989).
+
 Finally, it reads the data from a stream, a non-contiguous iterator range, and
 a FILE*, which read byte by byte, and checks that each gives the same value or
 error (id, message, byte position) as the pointer range, which takes the bulk
@@ -57,6 +61,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -81,6 +86,9 @@ static bool is_value_stable(const json& lhs, const json& rhs)
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::bjdata).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -99,6 +107,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2.1: round trip without adding size annotations to container types
@@ -132,6 +143,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -142,6 +154,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors may happen if provided sizes are excessive
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_bon8.cpp

@@ -17,6 +17,10 @@ array data, it performs the following steps:
 - j2 = from_bon8(vec)
 - assert(to_bon8(j2) == vec)
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_bon8() fails (see #3989).
+
 Finally, it reads the data from a stream, a non-contiguous iterator range, and
 a FILE*, which read byte by byte, and checks that each gives the same value or
 error (id, message, byte position) as the pointer range, which takes the bulk
@@ -27,6 +31,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -44,6 +49,9 @@ struct read_bon8
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::bon8).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -62,6 +70,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2: round trip
@@ -83,6 +94,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -93,6 +105,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors may happen if provided sizes are excessive
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_bson.cpp

@@ -17,6 +17,10 @@ array data, it performs the following steps:
 - j2 = from_bson(vec)
 - assert(to_bson(j2) == vec)
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_bson() fails (see #3989).
+
 Finally, it reads the data from a stream, a non-contiguous iterator range, and
 a FILE*, which read byte by byte, and checks that each gives the same value or
 error (id, message, byte position) as the pointer range, which takes the bulk
@@ -27,6 +31,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -44,6 +49,9 @@ struct read_bson
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::bson).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -62,6 +70,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2: round trip
@@ -83,6 +94,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -93,6 +105,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors can occur during parsing, too
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_cbor.cpp

@@ -17,6 +17,10 @@ array data, it performs the following steps:
 - j2 = from_cbor(vec)
 - assert(to_cbor(j2) == vec)
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_cbor() fails (see #3989).
+
 Finally, it reads the data from a stream, a non-contiguous iterator range, and
 a FILE*, which read byte by byte, and checks that each gives the same value or
 error (id, message, byte position) as the pointer range, which takes the bulk
@@ -27,6 +31,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -44,6 +49,9 @@ struct read_cbor
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::cbor).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -62,6 +70,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2: round trip
@@ -83,6 +94,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -93,6 +105,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors can occur during parsing, too
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_json.cpp

@@ -24,11 +24,16 @@ array data, it performs the following steps:
 The pointer range takes the bulk fast paths, the other input adapters read byte
 by byte; see check_input_adapters().
 
+Furthermore, it parses data with a SAX parser that recovers from every error
+and checks that the events are balanced, that parsing ends, and that valid
+input is parsed without errors (see #3989).
+
 The provided function `LLVMFuzzerTestOneInput` can be used in different fuzzer
 drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -59,6 +64,13 @@ struct accept_json
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    {
+        const auto checker = check_recovering_parse(data, size, json::input_format_t::json);
+        assert(checker.events <= (4 * size) + 4);
+        assert((checker.errors == 0) == json::accept(data, data + size));
+    }
+
     // step 0: parse input without exceptions
     bool noexcept_threw = false;
     json const j_noexcept = parse_without_exceptions([&] { return json::parse(data, data + size, nullptr, false); }, noexcept_threw);

tests/src/fuzzer-parse_msgpack.cpp

@@ -17,6 +17,10 @@ array data, it performs the following steps:
 - j2 = from_msgpack(vec)
 - assert(to_msgpack(j2) == vec)
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_msgpack() fails (see #3989).
+
 Finally, it reads the data from a stream, a non-contiguous iterator range, and
 a FILE*, which read byte by byte, and checks that each gives the same value or
 error (id, message, byte position) as the pointer range, which takes the bulk
@@ -27,6 +31,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -44,6 +49,9 @@ struct read_msgpack
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::msgpack).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -62,6 +70,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2: round trip
@@ -83,6 +94,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -93,6 +105,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors may happen if provided sizes are excessive
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-parse_ubjson.cpp

@@ -26,6 +26,10 @@ array data, it performs the following steps:
 The unit tests run the same checks on a fixed corpus (see the "UBJSON round-trip
 invariants" test case), so keep both in sync.
 
+Furthermore, it reads data with a SAX parser that recovers from every error
+and checks that the events are balanced, that reading ends, and that it
+reports an error exactly when from_ubjson() fails (see #3989).
+
 Finally, it reads the data from a stream, a non-contiguous iterator range, and
 a FILE*, which read byte by byte, and checks that each gives the same value or
 error (id, message, byte position) as the pointer range, which takes the bulk
@@ -36,6 +40,7 @@ drivers.
 */
 
 #include "fuzzer_common.hpp"
+#include "fuzzer-recovering_checker.hpp"
 
 namespace
 {
@@ -53,6 +58,9 @@ struct read_ubjson
 // see http://llvm.org/docs/LibFuzzer.html
 extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
 {
+    // recover from all errors, reading from memory and from a stream
+    const bool recovered_without_errors = check_recovering_parse(data, size, json::input_format_t::ubjson).errors == 0;
+
     std::vector<uint8_t> const vec1(data, data + size);
 
     // step 0: parse input without exceptions
@@ -71,6 +79,9 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
         // without exceptions, the same input must give the same value
         assert(!noexcept_threw && !j_noexcept.is_discarded() && same_value(j_noexcept, j1));
 
+        // the recovering parser must not have reported an error either
+        assert(recovered_without_errors);
+
         try
         {
             // step 2.1: round trip without adding size annotations to container types
@@ -102,6 +113,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // parse errors are ok, because input may be random bytes
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
     catch (const json::type_error&)
     {
@@ -112,6 +124,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size)
     {
         // out of range errors may happen if provided sizes are excessive
         assert(parsed || noexcept_threw || j_noexcept.is_discarded());
+        assert(parsed || !recovered_without_errors);
     }
 
     // step 3: the other input adapters must read the data alike

tests/src/fuzzer-recovering_checker.hpp

@@ -0,0 +1,154 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include <cassert>
+#include <cstddef>
+#include <cstdint>
+#include <sstream>
+#include <string>
+#include <vector>
+#include <nlohmann/json.hpp>
+
+namespace
+{
+// a SAX parser that recovers from every error and checks that the events are
+// balanced and that every key is followed by exactly one value
+class recovering_checker : public nlohmann::json_sax<nlohmann::json>
+{
+  public:
+    bool null() override
+    {
+        return value();
+    }
+
+    bool boolean(bool /*val*/) override
+    {
+        return value();
+    }
+
+    bool number_integer(number_integer_t /*val*/) override
+    {
+        return value();
+    }
+
+    bool number_unsigned(number_unsigned_t /*val*/) override
+    {
+        return value();
+    }
+
+    bool number_float(number_float_t /*val*/, const string_t& /*s*/) override
+    {
+        return value();
+    }
+
+    bool string(string_t& /*val*/) override
+    {
+        return value();
+    }
+
+    bool binary(binary_t& /*val*/) override
+    {
+        return value();
+    }
+
+    bool start_object(std::size_t /*elements*/) override
+    {
+        value();
+        stack.push_back('o');
+        return true;
+    }
+
+    bool key(string_t& /*val*/) override
+    {
+        ++events;
+        assert(!stack.empty() && stack.back() == 'o');
+        stack.back() = 'v';
+        return true;
+    }
+
+    bool end_object() override
+    {
+        ++events;
+        assert(!stack.empty() && stack.back() == 'o');
+        stack.pop_back();
+        return true;
+    }
+
+    bool start_array(std::size_t /*elements*/) override
+    {
+        value();
+        stack.push_back('a');
+        return true;
+    }
+
+    bool end_array() override
+    {
+        ++events;
+        assert(!stack.empty() && stack.back() == 'a');
+        stack.pop_back();
+        return true;
+    }
+
+    bool parse_error(std::size_t /*position*/, const std::string& /*last_token*/, const nlohmann::detail::exception& /*ex*/) override
+    {
+        ++errors;
+        return true;
+    }
+
+    bool complete() const
+    {
+        return stack.empty();
+    }
+
+    std::size_t events = 0;
+    std::size_t errors = 0;
+
+  private:
+    bool value()
+    {
+        ++events;
+        if (!stack.empty())
+        {
+            // an array element, or the value of a key
+            assert(stack.back() != 'o');
+            if (stack.back() == 'v')
+            {
+                stack.back() = 'o';
+            }
+        }
+        return true;
+    }
+
+    // 'a' for an array, 'o' for an object that expects a key, 'v' for an
+    // object that expects the value of a key
+    std::vector<char> stack {}; // NOLINT(readability-redundant-member-init)
+};
+
+/// parses @a data with a recovering_checker from memory and from a stream,
+/// checks that both see the same, that the events are balanced, and that the
+/// number of errors is bounded, and returns the checker (see #3989)
+inline recovering_checker check_recovering_parse(const std::uint8_t* data, const std::size_t size, const nlohmann::json::input_format_t format)
+{
+    recovering_checker checker;
+    const bool ok = nlohmann::json::sax_parse(data, data + size, &checker, format);
+    assert(checker.complete());
+    assert(checker.errors <= size + 1);
+    assert(ok == (checker.errors == 0));
+
+    std::istringstream stream(std::string(reinterpret_cast<const char*>(data), size));
+    recovering_checker stream_checker;
+    assert(nlohmann::json::sax_parse(stream, &stream_checker, format) == ok);
+    assert(stream_checker.complete());
+    assert(stream_checker.events == checker.events);
+    assert(stream_checker.errors == checker.errors);
+
+    return checker;
+}
+} // namespace

tests/src/sax_event_loggers.hpp

@@ -131,11 +131,13 @@ struct SaxEventLogger : public nlohmann::json_sax<nlohmann::json>
     {
         errored = true;
         events.push_back("parse_error(" + std::to_string(position) + ")");
-        return false;
+        return recover;
     }
 
     std::vector<std::string> events {}; // NOLINT(readability-redundant-member-init)
     bool errored = false;
+    /// whether parse_error() asks the parser to recover from the error (see #3989)
+    bool recover = false;
 };
 
 struct SaxEventLoggerExitAfterStartObject : public SaxEventLogger

tests/src/unit-32bit.cpp

@@ -97,3 +97,23 @@ TEST_CASE("CBOR")
         }
     }
 }
+
+TEST_CASE("MessagePack")
+{
+    SECTION("parse errors")
+    {
+        SECTION("array/map size equal to the unknown-size sentinel")
+        {
+            // on 32-bit platforms an array 32/map 32 length of 0xFFFFFFFF aliases unknown_size()
+            std::vector<uint8_t> const varr = {0xDD, 0xFF, 0xFF, 0xFF, 0xFF};
+            std::vector<uint8_t> const vmap = {0xDF, 0xFF, 0xFF, 0xFF, 0xFF};
+
+            json _;
+            CHECK_THROWS_WITH_AS(_ = json::from_msgpack(varr), "[json.exception.out_of_range.408] syntax error while parsing MessagePack size: excessive array size", json::out_of_range&);
+            CHECK(json::from_msgpack(varr, true, false).is_discarded());
+
+            CHECK_THROWS_WITH_AS(_ = json::from_msgpack(vmap), "[json.exception.out_of_range.408] syntax error while parsing MessagePack size: excessive map size", json::out_of_range&);
+            CHECK(json::from_msgpack(vmap, true, false).is_discarded());
+        }
+    }
+}

tests/src/unit-allocator.cpp

@@ -1038,6 +1038,14 @@ template<class T>
 struct sweep_allocator
 {
     using value_type = T;
+    // GCC 4.8's std::map takes these from the allocator instead of from
+    // std::allocator_traits
+    using pointer = T*;
+    using const_pointer = const T*;
+    using reference = T&;
+    using const_reference = const T&;
+    using size_type = std::size_t;
+    using difference_type = std::ptrdiff_t;
 
     sweep_allocator() = default;
     template<class U> sweep_allocator(const sweep_allocator<U>& /*unused*/) noexcept {}
@@ -1071,6 +1079,11 @@ struct sweep_allocator
         ::new (static_cast<void*>(p)) U(std::forward<Args>(args)...);
     }
 
+    // libc++ probes whether construct() is usable for const objects (when
+    // checking for trivial copies); this must not be viable then
+    template<class U, class... Args>
+    void construct(const U* /*unused*/, const Args& ... /*unused*/) = delete;
+
     template<class U>
     void destroy(U* p) noexcept
     {
@@ -1155,9 +1168,14 @@ TEST_CASE("parse() survives a failing allocation at every point (#5985)")
 
     // With iterator debugging (MSVC STL debug builds, also used by clang-cl),
     // containers allocate a debug proxy through the allocator inside their
-    // noexcept move constructors, so failing that allocation terminates the
+    // noexcept default and move constructors, so failing that allocation
-    // program instead of throwing std::bad_alloc. Nothing to check there.
+    // terminates the program instead of throwing std::bad_alloc. Nothing to
+    // check there.
 #if !(defined(_ITERATOR_DEBUG_LEVEL) && _ITERATOR_DEBUG_LEVEL > 0)
+    // Clang 3.4 and 3.5 leak memory when an element constructor throws
+    // inside libstdc++'s containers (a std::map copy, a std::vector
+    // reallocation), without any json value involved.
+#if !(defined(__clang__) && !defined(__apple_build_version__) && __clang_major__ == 3 && __clang_minor__ < 6)
     SECTION("parser callback")
     {
         const auto keep_all = [](int /*depth*/, sweep_json::parse_event_t /*event*/, sweep_json& /*parsed*/) noexcept
@@ -1178,7 +1196,6 @@ TEST_CASE("parse() survives a failing allocation at every point (#5985)")
             return sweep_json::parse(text, drop_d);
         });
     }
-#endif
 
     SECTION("sax_parse into an existing value")
     {
@@ -1192,5 +1209,7 @@ TEST_CASE("parse() survives a failing allocation at every point (#5985)")
             return value;
         });
     }
+#endif
+#endif
 }
 #endif

tests/src/unit-alt-string.cpp

@@ -433,6 +433,46 @@ TEST_CASE("alternative string type")
         CHECK(j2.dump() == R"({"/foo/0":"bar","/foo/1":"baz"})");
     }
 
+    SECTION("error recovery")
+    {
+        // a SAX parser that recovers from every error (see #3989)
+        struct recovering_parser : nlohmann::detail::json_sax_dom_parser<alt_json>
+        {
+            explicit recovering_parser(alt_json& j)
+                : nlohmann::detail::json_sax_dom_parser<alt_json>(j, false)
+            {}
+
+            // sax_parse() calls the SAX parser's own parse_error(), so hiding
+            // the one of the base class is what recovering takes
+            // NOLINTNEXTLINE(bugprone-derived-method-shadowing-base-method)
+            bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const nlohmann::detail::exception& /*unused*/)
+            {
+                ++errors;
+                return true;
+            }
+
+            std::size_t errors = 0;
+        };
+
+        alt_json j;
+        recovering_parser sax(j);
+        // not inside CHECK(): MSVC reads the escape in a stringized raw string
+        const std::string input = R"([1., "a\qb", tru, {"k" 2}])";
+        CHECK(!alt_json::sax_parse(input, &sax));
+        CHECK(sax.errors == 4);
+        CHECK(j.dump() == R"([1,"aqb",null,{"k":2}])");
+
+        // a UBJSON high-precision number, a CBOR key that is not a string
+        alt_json u;
+        recovering_parser ubjson_sax(u);
+        CHECK(!alt_json::sax_parse(std::vector<std::uint8_t> {'[', 'H', 'i', 2, '1', '.', ']'}, &ubjson_sax, alt_json::input_format_t::ubjson));
+        CHECK(u.dump() == "[1]");
+        alt_json c;
+        recovering_parser cbor_sax(c);
+        CHECK(!alt_json::sax_parse(std::vector<std::uint8_t> {0xA2, 0x01, 0x02, 0x61, 'a', 0x03}, &cbor_sax, alt_json::input_format_t::cbor));
+        CHECK(c.dump() == R"({"a":3})");
+    }
+
     SECTION("conversion between basic_json specializations (#2649)")
     {
         // explicit conversions are always possible

tests/src/unit-binary_writer_sinks.cpp

@@ -50,6 +50,14 @@ std::vector<json> test_values()
         large_payloads.push_back(json::binary(std::vector<std::uint8_t>(static_cast<std::size_t>(100 * i), static_cast<std::uint8_t>(i))));
     }
 
+    // several times vector_sink_window characters in every format: the vector
+    // sink makes the capacity it grows to writable a window at a time
+    json many_windows = json::array();
+    for (std::uint64_t i = 0; i < 30000; ++i)
+    {
+        many_windows.push_back((i * 0x9E3779B97F4A7C15u) >> 1u);
+    }
+
     return
     {
         json(nullptr), json(true), json(false),
@@ -61,7 +69,7 @@ std::vector<json> test_values()
         json::array(), json::object(),
         json::array({1, 2, 3}), json({{"a", 1}, {"b", nullptr}}),
         json({{"nested", {{"deep", json::array({1, "two", 3.0, nullptr})}}}}),
-        big_array, big_object, large_payloads,
+        big_array, big_object, large_payloads, many_windows,
         json::binary(std::vector<std::uint8_t>(5000, 0x42))
     };
 }

tests/src/unit-bjdata.cpp

@@ -78,23 +78,25 @@ TEST_CASE("BJData")
 {
     SECTION("binary_reader BJData lookup tables")
     {
-        // both lookups are static member functions
+        // bjd_type_name() is static; is_valid_optimized_type() needs a BJData reader
         std::vector<std::uint8_t> const data;
         using reader_t = nlohmann::detail::binary_reader<json, decltype(nlohmann::detail::input_adapter(data))>;
 
-        // the excluded optimized-type markers must match binary_writer's
+        // the markers BJData excludes as optimized types must match
-        // is_bjdata_excluded_type_marker(), which encodes the same 8 markers
+        // binary_writer's is_bjdata_excluded_type_marker(), which encodes the
+        // same 8 markers
+        const reader_t reader(nlohmann::detail::input_adapter(data), json::input_format_t::bjdata);
         for (const char marker :
-                {'[', '{', 'S', 'H', 'T', 'F', 'N', 'Z'
+                {'[', '{', 'S', 'H', 'T', 'F', 'N', 'Z', 'x'
                 })
         {
-            CHECK(reader_t::is_bjd_excluded_optimized_type(static_cast<unsigned char>(marker)));
+            CHECK(!reader.is_valid_optimized_type(static_cast<unsigned char>(marker)));
         }
         for (const char marker :
-                {'U', 'i', 'u', 'I', 'm', 'l', 'M', 'L', 'd', 'D', 'C', 'B', 'x'
+                {'U', 'i', 'u', 'I', 'm', 'l', 'M', 'L', 'd', 'D', 'C', 'B', 'h'
                 })
         {
-            CHECK(!reader_t::is_bjd_excluded_optimized_type(static_cast<unsigned char>(marker)));
+            CHECK(reader.is_valid_optimized_type(static_cast<unsigned char>(marker)));
         }
 
         // every dtype marker must round-trip to its ND-array type name
@@ -3237,14 +3239,34 @@ TEST_CASE("BJData")
 #endif
                 CHECK(json::from_bjdata(vM, true, false).is_discarded());
 
-#if SIZE_MAX != 0xffffffff
+                // the first dimension is 2^64-1, which is rejected as a count (see below)
-                CHECK_THROWS_WITH_AS(_ = json::from_bjdata(vMX), "[json.exception.out_of_range.408] syntax error while parsing BJData size: excessive ndarray size caused overflow", json::out_of_range&);
-#else
                 CHECK_THROWS_WITH_AS(_ = json::from_bjdata(vMX), "[json.exception.out_of_range.408] syntax error while parsing BJData size: integer value overflow", json::out_of_range&);
-#endif
                 CHECK(json::from_bjdata(vMX, true, false).is_discarded());
             }
 
+            SECTION("optimized array: count equal to SIZE_MAX")
+            {
+                // the reader marks a container without a count by SIZE_MAX, so
+                // a count of that value must not be read as "no count"
+                // (it used to parse as [1])
+                json _;
+#if SIZE_MAX != 0xffffffff
+                std::vector<uint8_t> const vM = {'[', '#', 'M', 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 'i', 0x01, ']'};
+                CHECK_THROWS_WITH_AS(_ = json::from_bjdata(vM), "[json.exception.out_of_range.408] syntax error while parsing BJData size: integer value overflow", json::out_of_range&);
+                CHECK(json::from_bjdata(vM, true, false).is_discarded());
+                // one less is an ordinary excessive count
+                std::vector<uint8_t> const vM1 = {'[', '#', 'M', 0xFE, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 'i', 0x01, ']'};
+                CHECK_THROWS_WITH_AS(_ = json::from_bjdata(vM1), "[json.exception.out_of_range.408] excessive array size: 18446744073709551614", json::out_of_range&);
+#else
+                std::vector<uint8_t> const vm = {'[', '#', 'm', 0xFF, 0xFF, 0xFF, 0xFF, 'i', 0x01, ']'};
+                CHECK_THROWS_WITH_AS(_ = json::from_bjdata(vm), "[json.exception.out_of_range.408] syntax error while parsing BJData size: integer value overflow", json::out_of_range&);
+                CHECK(json::from_bjdata(vm, true, false).is_discarded());
+                std::vector<uint8_t> const vL = {'[', '#', 'L', 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 'i', 0x01, ']'};
+                CHECK_THROWS_WITH_AS(_ = json::from_bjdata(vL), "[json.exception.out_of_range.408] syntax error while parsing BJData size: integer value overflow", json::out_of_range&);
+                CHECK(json::from_bjdata(vL, true, false).is_discarded());
+#endif
+            }
+
             SECTION("optimized array: integer value overflow")
             {
 #if SIZE_MAX == 0xffffffff

tests/src/unit-class_lexer.cpp

@@ -2222,9 +2222,10 @@ TEST_CASE("serialization of short strings")
     {
         CAPTURE(s)
         const json j = s;
-        CHECK(j.dump() == "\"" + s + "\"");
+        const std::string quoted = "\"" + s + "\""; // NOLINT(performance-inefficient-string-concatenation)
-        CHECK(j.dump(-1, ' ', true) == "\"" + s + "\"");
+        CHECK(j.dump() == quoted);
-        CHECK(json({{s, s}}).dump() == "{\"" + s + "\":\"" + s + "\"}");
+        CHECK(j.dump(-1, ' ', true) == quoted);
+        CHECK(json({{s, s}}).dump() == "{" + quoted + ":" + quoted + "}"); // NOLINT(performance-inefficient-string-concatenation)
     }
 
     // a character that needs an escape

tests/src/unit-class_parser.cpp

@@ -2953,6 +2953,651 @@ TEST_CASE("diagnostic positions: value lifetime, input adapters, and SAX")
 }
 #endif
 
+namespace
+{
+/// builds a value like json::parse(), but asks the parser to recover from
+/// errors (see #3989), and checks that the events it receives are balanced
+class RecoveringDomParser
+{
+  public:
+    explicit RecoveringDomParser(json& j, std::size_t max_errors_ = static_cast<std::size_t>(-1))
+        : dom(j, false)
+        , max_errors(max_errors_)
+    {}
+
+    bool null()
+    {
+        value();
+        return dom.null();
+    }
+
+    bool boolean(bool val)
+    {
+        value();
+        return dom.boolean(val);
+    }
+
+    bool number_integer(json::number_integer_t val)
+    {
+        value();
+        return dom.number_integer(val);
+    }
+
+    bool number_unsigned(json::number_unsigned_t val)
+    {
+        value();
+        return dom.number_unsigned(val);
+    }
+
+    bool number_float(json::number_float_t val, const std::string& s)
+    {
+        value();
+        return dom.number_float(val, s);
+    }
+
+    bool string(std::string& val)
+    {
+        value();
+        return dom.string(val);
+    }
+
+    bool binary(json::binary_t& val)
+    {
+        value();
+        return dom.binary(val);
+    }
+
+    bool start_object(std::size_t elements)
+    {
+        value();
+        stack.push_back('o');
+        return dom.start_object(elements);
+    }
+
+    bool key(std::string& val)
+    {
+        ++events;
+        if (stack.empty() || stack.back() != 'o')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.back() = 'v';
+        return dom.key(val);
+    }
+
+    bool end_object()
+    {
+        ++events;
+        if (stack.empty() || stack.back() != 'o')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.pop_back();
+        return dom.end_object();
+    }
+
+    bool start_array(std::size_t elements)
+    {
+        value();
+        stack.push_back('a');
+        return dom.start_array(elements);
+    }
+
+    bool end_array()
+    {
+        ++events;
+        if (stack.empty() || stack.back() != 'a')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.pop_back();
+        return dom.end_array();
+    }
+
+    bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const json::exception& ex)
+    {
+        errors.emplace_back(ex.what());
+        return errors.size() < max_errors;
+    }
+
+    /// whether the events were balanced and every key was followed by a value
+    bool balanced() const
+    {
+        return well_formed && stack.empty();
+    }
+
+    /// builds the value
+    nlohmann::detail::json_sax_dom_parser<json> dom;
+    std::vector<std::string> errors {}; // NOLINT(readability-redundant-member-init)
+    std::size_t events = 0;
+    /// the open containers: 'a' for an array, 'o' for an object that expects
+    /// a key, 'v' for an object that expects the value of a key
+    std::vector<char> stack {}; // NOLINT(readability-redundant-member-init)
+    bool well_formed = true;
+    std::size_t max_errors;
+
+  private:
+    /// a value is passed: it is an array element, or the value of a key
+    void value()
+    {
+        ++events;
+        if (!stack.empty())
+        {
+            if (stack.back() == 'v')
+            {
+                stack.back() = 'o';
+            }
+            else if (stack.back() == 'o')
+            {
+                // a value without a key
+                well_formed = false;
+            }
+        }
+    }
+
+};
+
+struct RecoveryResult
+{
+    json value;
+    std::vector<std::string> errors;
+    std::size_t events;
+    bool ok;
+    bool balanced;
+};
+
+template<typename InputType>
+RecoveryResult parse_recovering(InputType&& input, const bool strict = true,
+                                const bool ignore_comments = false, const bool ignore_trailing_commas = false)
+{
+    json j;
+    RecoveringDomParser sax(j);
+    const bool ok = json::sax_parse(std::forward<InputType>(input), &sax, json::input_format_t::json,
+                                    strict, ignore_comments, ignore_trailing_commas);
+    return {j, sax.errors, sax.events, ok, sax.balanced()};
+}
+
+/// stops after a number of events, but recovers from errors
+class RecoveringCountdown
+{
+  public:
+    explicit RecoveringCountdown(const int count)
+        : countdown(count)
+    {}
+
+    bool null()
+    {
+        return countdown.null();
+    }
+
+    bool boolean(bool val)
+    {
+        return countdown.boolean(val);
+    }
+
+    bool number_integer(json::number_integer_t val)
+    {
+        return countdown.number_integer(val);
+    }
+
+    bool number_unsigned(json::number_unsigned_t val)
+    {
+        return countdown.number_unsigned(val);
+    }
+
+    bool number_float(json::number_float_t val, const std::string& s)
+    {
+        return countdown.number_float(val, s);
+    }
+
+    bool string(std::string& val)
+    {
+        return countdown.string(val);
+    }
+
+    bool binary(json::binary_t& val)
+    {
+        return countdown.binary(val);
+    }
+
+    bool start_object(std::size_t elements)
+    {
+        return countdown.start_object(elements);
+    }
+
+    bool key(std::string& val)
+    {
+        return countdown.key(val);
+    }
+
+    bool end_object()
+    {
+        return countdown.end_object();
+    }
+
+    bool start_array(std::size_t elements)
+    {
+        return countdown.start_array(elements);
+    }
+
+    bool end_array()
+    {
+        return countdown.end_array();
+    }
+
+    bool parse_error(std::size_t /*position*/, const std::string& /*last_token*/, const json::exception& /*ex*/) // NOLINT(readability-convert-member-functions-to-static)
+    {
+        return true;
+    }
+
+  private:
+    SaxCountdown countdown;
+};
+
+/// a repaired input: the value it is repaired to, and the number of errors
+struct Repair
+{
+    const char* input;
+    const char* expected;
+    std::size_t errors;
+};
+} // namespace
+
+TEST_CASE("parser error recovery (#3989)")
+{
+    SECTION("repairs")
+    {
+        const std::vector<Repair> repairs =
+        {
+            // a missing separator is inserted
+            {"[1 2]", "[1,2]", 1},
+            {R"({"a":1 "b":2})", R"({"a":1,"b":2})", 1},
+            {R"({"a" 1})", R"({"a":1})", 1},
+            {"[1 tru 2]", "[1,null,2]", 2},
+            {R"({"a" "b": 1})", R"({"a":"b"})", 2},
+
+            // a missing value is null in an object; in an array, a ',' stands
+            // for null, while an array that ends there just ends
+            {R"({"a":})", R"({"a":null})", 1},
+            {R"({"a"})", R"({"a":null})", 1},
+            {R"({"a","b":1})", R"({"a":null,"b":1})", 1},
+            {"[1,,2]", "[1,null,2]", 1},
+            {"[,1]", "[null,1]", 1},
+            {"[1,]", "[1]", 1},
+            {"[1,2,3,]", "[1,2,3]", 1},
+            {R"({"a":1,})", R"({"a":1})", 1},
+
+            // a broken string keeps what can be read
+            {R"(["a\qb"])", R"(["aqb"])", 1},
+            {R"({"na\me":1})", R"({"name":1})", 1},
+            {"[\"\xFF\"]", R"(["\uFFFD"])", 1},
+            {"[\"a\xC3(\"]", R"(["a\uFFFD("])", 1},
+            {"[\"\xE2\x82\"]", R"(["\uFFFD"])", 1},
+            {"[\"\xC3\\\\\", 1]", R"(["\uFFFD\\",1])", 1},
+            {R"(["\u12"])", R"(["\uFFFD"])", 1},
+            {R"(["\u12G4"])", R"(["\uFFFDG4"])", 1},
+            {R"(["\uDC00x"])", R"(["\uFFFDx"])", 1},
+            {R"(["\uD800x"])", R"(["\uFFFDx"])", 1},
+            {R"(["\uD800\u0041"])", R"(["\uFFFDA"])", 1},
+            {R"(["\uD800\uD800\uDC00"])", R"(["\uFFFD\uD800\uDC00"])", 1},
+            {R"(["\uD800\uD800\uD800x"])", R"(["\uFFFD\uFFFD\uFFFDx"])", 1},
+            {
+                R"(["\uD800\"x", 1])", R"(["\uFFFD\"x",1])", 1
+            },
+            {R"(["\uD800\q"])", R"(["\uFFFDq"])", 1},
+            {"[\"a\tb\"]", R"(["a\tb"])", 1},
+            {R"(["a\qb\u0041\x"])", R"(["aqbAx"])", 1},
+
+            // a broken number keeps its longest valid prefix
+            {"[1.]", "[1]", 1},
+            {"[-2.]", "[-2]", 1},
+            {"[1.5e]", "[1.5]", 1},
+            {"[1e+]", "[1]", 1},
+            {"[1.x2, 3]", "[1,3]", 1},
+
+            // what cannot be read at all is null
+            {"[1,NaN,3]", "[1,null,3]", 1},
+            {"[tru]", "[null]", 1},
+            {"[-]", "[null]", 1},
+            {R"({"a":Infinity})", R"({"a":null})", 1},
+
+            // a stray token is dropped
+            {"[:1]", "[1]", 1},
+            {R"(["a":1])", R"(["a",1])", 1},
+            {R"({"a"::1})", R"({"a":1})", 1},
+
+            // a member that cannot be read is skipped
+            {R"({1:2,"b":3})", R"({"b":3})", 1},
+            {R"({"a":1 2})", R"({"a":1})", 1},
+            {R"({,"a":1})", R"({"a":1})", 1},
+            {R"({"a":1,,"b":2})", R"({"a":1,"b":2})", 1},
+            {"{a:1}", "{}", 1},
+            {R"({"a":1 [1,{"b":2}], "c":3})", R"({"a":1,"c":3})", 1},
+            {R"([{1}, "a"])", R"([{},"a"])", 1},
+
+            // a wrong closing bracket closes the innermost container
+            {R"({"a":[1,2}, "b":3})", R"({"a":[1,2],"b":3})", 1},
+            {R"([{"a":1], 2])", R"([{"a":1},2])", 1},
+            {"{]", "{}", 1},
+            {"[}", "[]", 1},
+
+            // the end of the input closes all containers
+            {R"({"a":[1,2)", R"({"a":[1,2]})", 1},
+            {"[", "[]", 1},
+            {"{", "{}", 1},
+            {R"({"a")", R"({"a":null})", 1},
+            {R"({"a":)", R"({"a":null})", 1},
+            {"[1,", "[1]", 1},
+            {"[[[1", "[[[1]]]", 1},
+            {
+                R"(["abc)", R"(["abc"])", 2
+            },
+            {"[1,tr", "[1,null]", 2},
+            {"\"abc", "\"abc\"", 1},
+            {"[\"ab\ncd\"]", R"(["ab",null,"]"])", 4},
+
+            // what comes before the top-level value is skipped
+            {")]}'\n{\"a\":1}", R"({"a":1})", 1},
+            {R"(data: {"a":1})", R"({"a":1})", 1},
+            {"\xEF\xBB[1]", "[1]", 1},
+
+            // what comes after it is an error that ends parsing
+            {R"({"a":1}})", R"({"a":1})", 1},
+            {"[1}]", "[1]", 2},
+            {"[1] [2]", "[1]", 1},
+        };
+
+        for (const auto& repair : repairs)
+        {
+            CAPTURE(repair.input)
+            const auto result = parse_recovering(std::string(repair.input));
+            CHECK(!result.ok);
+            CHECK(result.balanced);
+            CHECK(result.value == json::parse(repair.expected));
+            CHECK(result.errors.size() == repair.errors);
+        }
+    }
+
+    SECTION("number overflow")
+    {
+        const auto result = parse_recovering(std::string("[1e999,-1e999]"));
+        CHECK(!result.ok);
+        CHECK(result.balanced);
+        CHECK(result.errors.size() == 2);
+        CHECK(result.errors[0] == "[json.exception.out_of_range.406] number overflow parsing '1e999'");
+        REQUIRE(result.value.size() == 2);
+        CHECK(result.value[0].is_number_float());
+        CHECK(result.value[0].get<double>() == std::numeric_limits<double>::infinity());
+        CHECK(result.value[1].get<double>() == -std::numeric_limits<double>::infinity());
+
+        // the SAX parser gets the number's text
+        SaxEventLogger logger;
+        logger.recover = true;
+        CHECK(!json::sax_parse("1e999", &logger));
+        CHECK(logger.events == std::vector<std::string>({"parse_error(5)", "number_float(1e999)"}));
+    }
+
+    SECTION("nothing to recover")
+    {
+        for (const std::string s :
+                {
+                    "", "   ", "]", "tru", "NaN", ",:", "/* comment"
+                })
+        {
+            CAPTURE(s)
+            const auto result = parse_recovering(s, true, true);
+            CHECK(!result.ok);
+            CHECK(result.balanced);
+            CHECK(result.events == 0);
+            CHECK(result.value == nullptr);
+            CHECK(result.errors.size() == 1);
+        }
+    }
+
+    SECTION("error messages")
+    {
+        // the first error is reported as without recovery
+        for (const std::string s :
+                {
+                    "[1 2]", R"({"a":1 "b":2})", R"({"a" 1})", R"({"a":})", "[1,]", "[1.]",
+                    R"(["a\qb"])", "[1e999]", "{1:2}", R"({"a":[1,2}})", "[1,", "[1] [2]", "{a:1}"
+                })
+        {
+            CAPTURE(s)
+            const auto result = parse_recovering(s);
+            REQUIRE(!result.errors.empty());
+            json _;
+            CHECK_THROWS_WITH_STD_STR(_ = json::parse(s), result.errors.front());
+        }
+
+        // the token of an error begins where the previous error was
+        const auto result = parse_recovering(std::string("[tru, fals, nul]"));
+        CHECK(result.errors == std::vector<std::string>(
+        {
+            "[json.exception.parse_error.101] parse error at line 1, column 5: syntax error while parsing value - invalid literal; last read: '[tru,'",
+            "[json.exception.parse_error.101] parse error at line 1, column 11: syntax error while parsing value - invalid literal; last read: ', fals,'",
+            "[json.exception.parse_error.101] parse error at line 1, column 16: syntax error while parsing value - invalid literal; last read: ', nul]'"
+        }));
+        CHECK(result.value == json::parse("[null,null,null]"));
+    }
+
+    SECTION("events")
+    {
+        // see #4522
+        SaxEventLogger logger;
+        logger.recover = true;
+        CHECK(!json::sax_parse(R"([{1}, "a"])", &logger));
+        CHECK(logger.events == std::vector<std::string>(
+        {
+            "start_array()", "start_object()", "parse_error(3)", "end_object()", "string(a)", "end_array()"
+        }));
+    }
+
+    SECTION("options")
+    {
+        SECTION("strict")
+        {
+            const auto result = parse_recovering(std::string("[1 2] [3]"), false);
+            CHECK(!result.ok);
+            CHECK(result.value == json::parse("[1,2]"));
+            CHECK(result.errors.size() == 1);
+        }
+
+        SECTION("ignore_trailing_commas")
+        {
+            for (const std::string s :
+                    {
+                        "[1,]", R"({"a":1,})", "[[1,],]"
+                    })
+            {
+                CAPTURE(s)
+                const auto result = parse_recovering(s, true, false, true);
+                CHECK(result.ok);
+                CHECK(result.errors.empty());
+            }
+
+            auto result = parse_recovering(std::string("[1,,]"), true, false, true);
+            CHECK(result.value == json::parse("[1,null]"));
+            CHECK(result.errors.size() == 1);
+
+            result = parse_recovering(std::string(R"({"a":1,,})"), true, false, true);
+            CHECK(result.value == json::parse(R"({"a":1})"));
+            CHECK(result.errors.size() == 1);
+        }
+
+        SECTION("ignore_comments")
+        {
+            auto result = parse_recovering(std::string("[1 /* one */ 2]"), true, true);
+            CHECK(result.value == json::parse("[1,2]"));
+            CHECK(result.errors.size() == 1);
+
+            // a comment that is not closed runs to the end of the input, which
+            // is not reported again
+            result = parse_recovering(std::string("[1, 2 /* unterminated"), true, true);
+            CHECK(result.balanced);
+            CHECK(result.value == json::parse("[1,2]"));
+            CHECK(result.errors.size() == 1);
+
+            // a '/' that does not begin a comment is garbage
+            result = parse_recovering(std::string("[1, /x, 2]"), true, true);
+            CHECK(result.balanced);
+            CHECK(result.value == json::parse("[1,null,2]"));
+            CHECK(result.errors.size() == 1);
+        }
+    }
+
+    SECTION("null bytes")
+    {
+        // a null byte ends the input, unless JSON_STRICT_NUL_HANDLING is set
+        const auto result = parse_recovering(std::string("[1,\0x", 5));
+        CHECK(result.balanced);
+        CHECK(!result.ok);
+#ifdef JSON_TEST_STRICT_NUL_HANDLING_ENABLED
+        CHECK(result.value == json::parse("[1,null]"));
+#else
+        CHECK(result.value == json::parse("[1]"));
+        CHECK(result.errors.size() == 1);
+#endif
+
+        const auto in_string = parse_recovering(std::string("[\"a\0b\"]", 7));
+        CHECK(in_string.balanced);
+#ifdef JSON_TEST_STRICT_NUL_HANDLING_ENABLED
+        CHECK(in_string.value == json::array({std::string("a\0b", 3)}));
+#else
+        CHECK(in_string.value == json::parse(R"(["a"])"));
+#endif
+    }
+
+    SECTION("the SAX parser stops recovering")
+    {
+        json j;
+        RecoveringDomParser sax(j, 2);
+        CHECK(!json::sax_parse("[1 2 3 4 5]", &sax));
+        CHECK(sax.errors.size() == 2);
+
+        // an error at a delimiter that an invalid token consumed is reported
+        // to the SAX parser, too
+        json j2;
+        RecoveringDomParser sax2(j2, 2);
+        CHECK(!json::sax_parse("[tru}, 1]", &sax2));
+        CHECK(sax2.errors.size() == 2);
+    }
+
+    SECTION("an event stops parsing during a repair")
+    {
+        // start_object() and key() are passed, then null() for the missing
+        // value returns false
+        RecoveringCountdown countdown(2);
+        CHECK(!json::sax_parse(R"({"a":})", &countdown));
+
+        // the end of the input: end_array() for the second array returns false
+        RecoveringCountdown countdown2(4);
+        CHECK(!json::sax_parse("[[1", &countdown2));
+    }
+
+    SECTION("input adapters")
+    {
+        // the lexer reads contiguous and streaming input differently, and it
+        // puts back a character that ended an invalid token
+        for (const std::string s :
+                {
+                    "[1 2]", "[tru}, 1]", R"({"a" "b\q", "c":[1.x, 2}})", "[\"\xFF\xC3(\", -, 1e+]", "{a:1,\"b\":2", ")]}' [1]"
+                })
+        {
+            CAPTURE(s)
+            const auto reference = parse_recovering(s);
+            CHECK(reference.balanced);
+
+            const auto from_c_string = parse_recovering(s.c_str());
+            CHECK(from_c_string.value == reference.value);
+            CHECK(from_c_string.errors == reference.errors);
+
+            const std::list<char> l(s.begin(), s.end());
+            json j;
+            RecoveringDomParser sax(j);
+            CHECK(!json::sax_parse(l.begin(), l.end(), &sax));
+            CHECK(j == reference.value);
+            CHECK(sax.errors == reference.errors);
+
+            std::istringstream ss(s);
+            const auto from_stream = parse_recovering(ss);
+            CHECK(from_stream.value == reference.value);
+            CHECK(from_stream.errors == reference.errors);
+        }
+    }
+
+    SECTION("long runs of errors")
+    {
+        // no error may copy all the input read before it
+        const auto closing = parse_recovering("[" + std::string(100000, '}'));
+        CHECK(closing.balanced);
+        CHECK(closing.value == json::array());
+
+        const auto garbage = parse_recovering("[" + std::string(100000, 'x') + "]");
+        CHECK(garbage.balanced);
+        CHECK(garbage.errors.size() == 1);
+
+        const auto commas = parse_recovering("{" + std::string(100000, ',') + "}");
+        CHECK(commas.balanced);
+        CHECK(commas.value == json::object());
+    }
+
+    SECTION("mutations of valid input")
+    {
+        // whatever the input, the events are balanced, every error is reported
+        // at most once, and valid input is parsed as usual
+        const std::vector<std::string> documents =
+        {
+            R"({"name": "value", "list": [1, -2.5, true, null, {"x": [[]]}], "e": "\u00e9"})",
+            R"([{"a": [1, 2, {"b": "c"}]}, [], {}, "\ud83d\ude00", 1e10])",
+            "{\"\xC3\xA9\": \"\xF0\x9F\x98\x80\"}",
+            R"(  {"k" : [ "v" , 0 ] }  )",
+        };
+        // each character that can be inserted, including a null byte
+        const std::string insertions("[]{},:\"x\\\0\xFF", 11);
+
+        std::vector<std::string> inputs;
+        for (const auto& doc : documents)
+        {
+            for (std::size_t i = 0; i <= doc.size(); ++i)
+            {
+                inputs.push_back(doc.substr(0, i));
+                if (i < doc.size())
+                {
+                    inputs.push_back(doc.substr(0, i) + doc.substr(i + 1));
+                }
+                for (const char c : insertions)
+                {
+                    inputs.push_back(doc.substr(0, i) + c + doc.substr(i));
+                }
+            }
+        }
+
+        for (const auto& s : inputs)
+        {
+            CAPTURE(s)
+            const auto result = parse_recovering(s);
+            CHECK(result.balanced);
+            CHECK(result.errors.size() <= s.size() + 1);
+            CHECK(result.events <= (4 * s.size()) + 4);
+            if (json::accept(s))
+            {
+                CHECK(result.ok);
+                CHECK(result.errors.empty());
+                CHECK(result.value == json::parse(s));
+            }
+            else
+            {
+                CHECK(!result.ok);
+                CHECK(!result.errors.empty());
+            }
+        }
+    }
+}
+
 TEST_CASE("accept() agrees with parse() on floats around the overflow limit")
 {
     // accept() does not convert a float that cannot overflow; the tokens

tests/src/unit-custom-object-key-type.cpp

@@ -160,12 +160,12 @@ class strict_key
     }
 
   private:
-    std::string m_value{};
+    std::string m_value;
 };
 
 strict_key make_strict_key(const std::string& key)
 {
-    return strict_key(key.c_str());
+    return {key.c_str()};
 }
 
 template<class Key, class Value, class Compare, class Allocator>
@@ -224,7 +224,9 @@ TEST_CASE("converting a value with keys that cannot be made from a std::string n
         }
     }
 
-    const strict_json converted = value;
+    // direct-initialization: a copy-initialization makes GCC 4.8 instantiate the
+    // copy constructor, which needs a range constructor strict_object lacks
+    const strict_json converted(value);
 
     // at() needs to make a std::string of a key for its error message
     const auto member = [](const strict_json & object, const char* name) -> const strict_json &

tests/src/unit-file_input.cpp

@@ -49,6 +49,14 @@ file_ptr open_with(const char* name, const std::string& content)
     return {std::fopen(name, "rb"), &std::fclose};
 }
 
+// the smallest size setvbuf() accepts for _IOFBF: the Microsoft C runtime
+// rejects sizes below 2 as an invalid parameter, which ends the process
+#if defined(_WIN32)
+    constexpr int min_full_buffer = 2;
+#else
+    constexpr int min_full_buffer = 1;
+#endif
+
 // The stdio lock is recursive, so whether parsing released it can only be
 // seen from another thread: ftrylockfile() fails there while this thread
 // holds the lock. ftrylockfile() is POSIX; elsewhere, the check is skipped.
@@ -225,6 +233,131 @@ TEST_CASE("parsing from a FILE*")
         check_unlocked(f.get());
     }
 
+#if !defined(JSON_NOEXCEPTION)
+    SECTION("tokens across the edges of the file's buffer")
+    {
+        // the adapter reads the buffer of the FILE directly where it can (see
+        // JSON_FILE_WINDOW): with buffers of 1 to 70 bytes and none, every
+        // token crosses an edge somewhere; the results, the error messages
+        // and their positions must be those of the string
+        const std::vector<std::string> documents =
+        {
+            input,
+            R"({"long": "a string that is longer than the buffer of the file, with \"escapes\" and \u00e4", "n": [123456789012345678901, -0.000123e-45, 1.5E+300, 0, -0]})",
+            "[\n    \"indented\",\n    \"caf\xc3\xa9 \xe2\x82\xac \xf0\x9f\x98\x80\",\n    {\n        \"key\": 1\n    }\n]",
+            R"({"a": [1, 2, 3], "b": "unterminated)",
+            R"([1, 2, 3, 1e999])",
+            R"({"a": "invalid escape \x"})",
+            "[\"invalid UTF-8 \xc3\x28\"]",
+            "[1, 2, 3] trailing",
+            R"([truex])",
+        };
+        for (const auto& document : documents)
+        {
+            std::string expected;
+            try
+            {
+                expected = json::parse(document).dump();
+            }
+            catch (const json::exception& e)
+            {
+                expected = e.what();
+            }
+            const bool expected_accept = json::accept(document);
+            for (int size = 0; size <= 70; ++size)
+            {
+                if (size != 0 && size < min_full_buffer)
+                {
+                    continue;
+                }
+                CAPTURE(document)
+                CAPTURE(size)
+                std::vector<char> buffer(static_cast<std::size_t>(size) + 1);
+                for (int pass = 0; pass < 2; ++pass)
+                {
+                    const file_ptr f = open_with(name, document);
+                    REQUIRE(f != nullptr);
+                    // size 0: unbuffered
+                    REQUIRE(std::setvbuf(f.get(), size == 0 ? nullptr : buffer.data(), size == 0 ? _IONBF : _IOFBF, static_cast<std::size_t>(size)) == 0);
+                    if (pass == 0)
+                    {
+                        std::string actual;
+                        try
+                        {
+                            actual = json::parse(f.get()).dump();
+                        }
+                        catch (const json::exception& e)
+                        {
+                            actual = e.what();
+                        }
+                        CHECK(actual == expected);
+                    }
+                    else
+                    {
+                        CHECK(json::accept(f.get()) == expected_accept);
+                    }
+                    check_unlocked(f.get());
+                }
+            }
+        }
+    }
+#endif
+
+    // the C standard guarantees only one pushed-back character (and the
+    // Windows C runtime fails further calls to ungetc()), so this is limited
+    // to the C libraries that keep them in a separate buffer
+#if !defined(JSON_NOEXCEPTION) && ((defined(__GLIBC__) && !defined(__UCLIBC__)) || defined(__APPLE__))
+    SECTION("characters pushed back with ungetc() before parsing")
+    {
+        const file_ptr f = open_with(name, "2, 3]");
+        REQUIRE(f != nullptr);
+        REQUIRE(std::ungetc(',', f.get()) == ',');
+        REQUIRE(std::ungetc('1', f.get()) == '1');
+        REQUIRE(std::ungetc('[', f.get()) == '[');
+        CHECK(json::parse(f.get()) == json({1, 2, 3}));
+    }
+#endif
+
+    SECTION("SAX events across the edges of the file's buffer")
+    {
+        const std::string document = R"({"key": "a value that is longer than the buffer", "n": [1, -2.5e3, true, null]})";
+        std::vector<long> expected;
+        {
+            const file_ptr f = open_with(name, document);
+            REQUIRE(f != nullptr);
+            position_reading_sax sax(f.get());
+            CHECK(json::sax_parse(f.get(), &sax));
+            expected = sax.positions();
+        }
+        for (int size = min_full_buffer; size <= 20; ++size)
+        {
+            CAPTURE(size)
+            std::vector<char> buffer(static_cast<std::size_t>(size));
+            const file_ptr f = open_with(name, document);
+            REQUIRE(f != nullptr);
+            REQUIRE(std::setvbuf(f.get(), buffer.data(), _IOFBF, buffer.size()) == 0);
+            position_reading_sax sax(f.get());
+            CHECK(json::sax_parse(f.get(), &sax));
+            // the events happen at the same positions of the file
+            CHECK(sax.positions() == expected);
+        }
+    }
+
+    SECTION("binary strings across the edges of the file's buffer")
+    {
+        const json value = {{"a string that is longer than the buffer of the file", "and a value that is longer, too"}, {"n", {1, 2}}};
+        const std::vector<std::uint8_t> cbor = json::to_cbor(value);
+        for (int size = min_full_buffer; size <= 20; ++size)
+        {
+            CAPTURE(size)
+            std::vector<char> buffer(static_cast<std::size_t>(size));
+            const file_ptr f = open_with(name, std::string(reinterpret_cast<const char*>(cbor.data()), cbor.size()));
+            REQUIRE(f != nullptr);
+            REQUIRE(std::setvbuf(f.get(), buffer.data(), _IOFBF, buffer.size()) == 0);
+            CHECK(json::from_cbor(f.get()) == value);
+        }
+    }
+
     SECTION("binary formats read from a FILE*")
     {
         const json j = {{"string", std::string(300, 'x')}, {"array", {1, 2, 3}}, {"number", 1.5}};

tests/src/unit-json_view.cpp

@@ -105,19 +105,53 @@ template<typename T>
 struct failing_allocator
 {
     using value_type = T;
+    // GCC 4.8's std::map takes these from the allocator instead of from
+    // std::allocator_traits
+    using pointer = T*;
+    using const_pointer = const T*;
+    using reference = T&;
+    using const_reference = const T&;
+    using size_type = std::size_t;
+    using difference_type = std::ptrdiff_t;
+
+    template<typename U>
+    struct rebind
+    {
+        using other = failing_allocator<U>;
+    };
 
     failing_allocator() = default;
 
     template<typename U>
     explicit failing_allocator(const failing_allocator<U>& /*other*/) noexcept {}
 
-    T* allocate(std::size_t /*n*/) // NOLINT(readability-convert-member-functions-to-static)
+    [[noreturn]] T* allocate(std::size_t /*n*/) // NOLINT(readability-convert-member-functions-to-static)
     {
         throw std::bad_alloc();
     }
 
     void deallocate(T* /*p*/, std::size_t /*n*/) noexcept {} // NOLINT(readability-convert-member-functions-to-static)
 
+    // GCC 4.8's std::map calls these on the allocator directly instead of
+    // going through std::allocator_traits
+    std::size_t max_size() const noexcept // NOLINT(readability-convert-member-functions-to-static)
+    {
+        return (std::numeric_limits<std::size_t>::max)() / sizeof(T);
+    }
+
+    template<typename U, typename... Args>
+    void construct(U* p, Args&& ... args) // NOLINT(readability-convert-member-functions-to-static)
+    {
+        ::new (static_cast<void*>(p)) U(std::forward<Args>(args)...);
+    }
+
+    template<typename U>
+    void destroy(U* p) noexcept // NOLINT(readability-convert-member-functions-to-static)
+    {
+        static_cast<void>(p); // unused for trivially destructible types
+        p->~U();
+    }
+
     template<typename U>
     bool operator==(const failing_allocator<U>& /*other*/) const noexcept
     {
@@ -270,7 +304,7 @@ TEST_CASE("json_view")
                     "\"", "\"abc", "\"\\x\"", "\"\\u12\"", "\"\\ud800\"", "\"\\udc00\"", "\"\x01\"", "\"\xff\"", "\"\xc3\"", "[1]x", "/", "/*", "[\n  1,\n  x\n]", // NOLINT(modernize-raw-string-literal)
                     "1e400", "-1e400", "[1.7976931348623159e308]", "{\"a\":\n{\"b\": [1, 2,\n 3 x]}}",
                     // (an error after every kind of value: the error is found by replaying the parser)
-                    "[null, true, -1, 1, 1.5, \"s\", {\"a\": 1} 2]", "{\"a\": null, \"b\": -1, \"c\": 1.5, \"d\": \"s\", \"e\": {}, 1}"
+                    R"([null, true, -1, 1, 1.5, "s", {"a": 1} 2])", R"({"a": null, "b": -1, "c": 1.5, "d": "s", "e": {}, 1})"
                 })
         {
             CAPTURE(text)
@@ -1556,7 +1590,7 @@ TEST_CASE("json_view values")
         CHECK_THROWS_WITH_AS(not_an_object.root().get<view_map>(), json_view_test::message("type_error.302", "", 0, 5, "type must be object, but is array").c_str(), json::type_error&);
 #if !defined(JSON_NOEXCEPTION)
         // (and one that cannot allocate its nodes)
-        using failing_map = std::map<std::string, json_view, std::less<std::string>, failing_allocator<std::pair<const std::string, json_view>>>;
+        using failing_map = std::map<std::string, json_view, std::less<std::string>, failing_allocator<std::pair<const std::string, json_view>>>; // NOLINT(modernize-use-transparent-functors)
         const json_document one_member = json_document::parse(R"({"a":1})");
         CHECK_THROWS_AS(one_member.root().get<failing_map>(), std::bad_alloc&);
 #endif

tests/src/unit-locale-cpp.cpp

@@ -397,6 +397,14 @@ TEST_CASE("locale with a multi-byte decimal point")
 
         // a value the locale-independent paths convert is not affected
         CHECK(json::parse("12.5") == 12.5);
+
+        // dump() of a long double that takes the snprintf path replaces the
+        // whole decimal point, not only its first byte (#5993)
+        CHECK(long_double_json(1.5L).dump() == "1.5");
+        CHECK(long_double_json(2.0L).dump() == "2.0");
+        CHECK(long_double_json(-0.25L).dump() == "-0.25");
+        CHECK(long_double_json::parse(long_double_json(1.5e-7L).dump()) == long_double_json(1.5e-7L));
+        CHECK(long_double_json::parse(ld.dump()) == expected_long_double);
     }
     if (!tested)
     {
@@ -505,8 +513,8 @@ struct locale_switching_token
 {
     using value_type = char;
 
-    locale_switching_token(std::string text_, const char* switch_to_)
+    locale_switching_token(std::string text_, std::string switch_to_)
-        : text(std::move(text_)), switch_to(switch_to_)
+        : text(std::move(text_)), switch_to(std::move(switch_to_))
     {}
 
     std::size_t size() const
@@ -523,13 +531,13 @@ struct locale_switching_token
     {
         if (!switched)
         {
-            switched = std::setlocale(LC_NUMERIC, switch_to) != nullptr;
+            switched = std::setlocale(LC_NUMERIC, switch_to.c_str()) != nullptr;
         }
-        return &text[0];
+        return &text[0]; // NOLINT(readability-container-data-pointer): std::string::data() is const before C++17
     }
 
     std::string text;
-    const char* switch_to;
+    std::string switch_to;
     bool switched = false;
 };
 
@@ -647,7 +655,7 @@ TEST_CASE("convert_float_locale_aware")
 
             // "C" -> comma locale: strtod stops at the '.', so the conversion is repeated
             {
-                locale_switching_token token("1.5", comma_locale.c_str());
+                locale_switching_token token("1.5", comma_locale);
                 double value = 0;
                 nlohmann::detail::convert_float_locale_aware(token, 1, value);
                 if (token.switched)
@@ -660,7 +668,7 @@ TEST_CASE("convert_float_locale_aware")
             // the same for the other floating-point types
             REQUIRE(std::setlocale(LC_NUMERIC, "C") != nullptr);
             {
-                locale_switching_token token("2.25", comma_locale.c_str());
+                locale_switching_token token("2.25", comma_locale);
                 float value = 0;
                 nlohmann::detail::convert_float_locale_aware(token, 1, value);
                 if (token.switched)
@@ -670,7 +678,7 @@ TEST_CASE("convert_float_locale_aware")
             }
             REQUIRE(std::setlocale(LC_NUMERIC, "C") != nullptr);
             {
-                locale_switching_token token("2.25", comma_locale.c_str());
+                locale_switching_token token("2.25", comma_locale);
                 long double value = 0;
                 nlohmann::detail::convert_float_locale_aware(token, 1, value);
                 if (token.switched)

tests/src/unit-msgpack-cpp17.cpp

@@ -122,7 +122,8 @@ TEST_CASE("JSON text with std::byte")
     SECTION("invalid text")
     {
         CHECK_FALSE(json::accept(to_bytes("[1,\n2,")));
-        CHECK_THROWS_WITH_AS(json::parse(to_bytes("[1,\n2,")),
+        json _;
+        CHECK_THROWS_WITH_AS(_ = json::parse(to_bytes("[1,\n2,")),
                              "[json.exception.parse_error.101] parse error at line 2, column 3: syntax error while parsing value - unexpected end of input; expected '[', '{', or a literal", json::parse_error&);
     }
 }

tests/src/unit-msgpack.cpp

@@ -1648,30 +1648,6 @@ TEST_CASE("MessagePack")
             }
         }
     }
-
-    SECTION("SAX aborts")
-    {
-        SECTION("start_array(len)")
-        {
-            std::vector<uint8_t> const v = {0x93, 0x01, 0x02, 0x03};
-            SaxCountdown scp(0);
-            CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
-        }
-
-        SECTION("start_object(len)")
-        {
-            std::vector<uint8_t> const v = {0x81, 0xa3, 0x66, 0x6F, 0x6F, 0xc2};
-            SaxCountdown scp(0);
-            CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
-        }
-
-        SECTION("key()")
-        {
-            std::vector<uint8_t> const v = {0x81, 0xa3, 0x66, 0x6F, 0x6F, 0xc2};
-            SaxCountdown scp(1);
-            CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
-        }
-    }
 }
 
 TEST_CASE("issue #5405 - array reserve for definite-length MessagePack arrays")
@@ -1742,21 +1718,6 @@ TEST_CASE("issue #5405 - array reserve for definite-length MessagePack arrays")
             CHECK(json::from_msgpack(packed) == j);
         }
     }
-
-    SECTION("a user-defined SAX consumer is unaffected by the internal DOM reserve optimization")
-    {
-        // the reserve() call is local to json_sax_dom_parser / json_sax_dom_callback_parser;
-        // a custom SAX consumer that does not touch a DOM array sees identical events
-        json j = json::array();
-        for (int i = 0; i < 100; ++i)
-        {
-            j.push_back(i);
-        }
-        const auto packed = json::to_msgpack(j);
-
-        SaxCountdown scp(1000000); // large enough to never trigger an abort
-        CHECK(json::sax_parse(packed, &scp, json::input_format_t::msgpack));
-    }
 }
 
 TEST_CASE("regression test - MessagePack ext type rejects a subtype that doesn't fit a single byte")
@@ -1795,15 +1756,6 @@ TEST_CASE("MessagePack nesting does not consume the call stack")
         CHECK(json::from_msgpack(input, true, false).is_discarded());
     }
 
-    SECTION("a well-formed deep value is read through the SAX interface")
-    {
-        std::vector<uint8_t> input(300000, 0x91);
-        input.push_back(0x01); // innermost value
-
-        SaxCountdown accept_all(600001);
-        CHECK(json::sax_parse(input, &accept_all, json::input_format_t::msgpack));
-    }
-
     SECTION("a well-formed deep value is read into a value")
     {
         const std::size_t depth = 10000;
@@ -1851,31 +1803,6 @@ TEST_CASE("MessagePack input that cannot be read is discarded by every overload"
 #endif
 }
 
-TEST_CASE("MessagePack SAX parsing stops at every event")
-{
-    // Containers are opened and closed by the loop that reads them; a SAX
-    // handler that rejects any event - including the end of a nested
-    // container - must stop the parse right there.
-    const auto count_events = [](const std::vector<std::uint8_t>& input)
-    {
-        int events = 0;
-        while (true)
-        {
-            SaxCountdown scp(events);
-            if (json::sax_parse(input, &scp, json::input_format_t::msgpack))
-            {
-                return events;
-            }
-            ++events;
-            REQUIRE(events < 1000);
-        }
-    };
-
-    // 20 events: every container kind closes inside another one
-    const json j = json::parse(R"({"a": [1, {"b": []}], "c": {"d": [[2]]}})");
-    CHECK(count_events(json::to_msgpack(j)) == 20);
-}
-
 TEST_CASE("single MessagePack roundtrip")
 {
     SECTION("sample.json")

tests/src/unit-ordered_json.cpp

@@ -541,6 +541,7 @@ TEST_CASE("object_key_index")
         };
 
         std::vector<std::size_t> hashes;
+        hashes.reserve(initial);
         for (std::size_t i = 0; i < initial; ++i)
         {
             hashes.push_back(hash_of(i));

tests/src/unit-regression1.cpp

@@ -1533,7 +1533,7 @@ TEST_CASE("regression tests 1")
         {
             my_json::object_t top = orig.get_ref<const my_json::object_t&>();
             top.at("x") = my_json::parse(R"({"a":{"b":1},"n":null,"arr":[10,20],"objs":[{"m":5},{"m":6}]})");
-            return my_json(std::move(top));
+            return my_json(std::move(top)); // NOLINT(modernize-return-braced-init-list): {top} would be an array
         };
 
         SECTION("existing members")

tests/src/unit-regression2.cpp

@@ -1009,7 +1009,7 @@ TEST_CASE("regression test - excessive binary container size honors allow_except
     // allow_exceptions=true (the default) must still throw exactly as before.
     // The exact message text is not checked here: on platforms where
     // std::size_t is 32-bit, the CBOR reader's own length-narrowing check
-    // (get_cbor_container_size(), unrelated to this fix) intercepts a
+    // (get_container_size(), unrelated to this fix) intercepts a
     // declared length of 2^63 before it ever reaches the check this test
     // targets, with different (but equally valid, and already correct)
     // wording -- see unit-cbor.cpp for coverage of that message.

tests/src/unit-regression3.cpp

@@ -22,14 +22,6 @@
 // scoped enum, so get<std::byte>() (needed below to get<std::vector<std::byte>>()
 // from a plain JSON array, not just from an already-binary value) relies on
 // enum serialization being enabled
-// capture whether JSON_DELETE_DEPRECATED_FUNCTIONS was enabled on the command
-// line *before* including json.hpp, since the library #undefs it once the header
-// has been fully processed (see include/nlohmann/detail/macro_unscope.hpp); the
-// tests of deprecated functions are skipped if these functions are deleted
-#if defined(JSON_DELETE_DEPRECATED_FUNCTIONS) && (JSON_DELETE_DEPRECATED_FUNCTIONS == 1)
-    #define JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
-#endif
-
 #if defined(JSON_DISABLE_ENUM_SERIALIZATION) && (JSON_DISABLE_ENUM_SERIALIZATION == 1)
     #define SKIP_TESTS_FOR_ENUM_SERIALIZATION
 #endif
@@ -810,50 +802,6 @@ TEST_CASE("issue #5338 - truncated CBOR tagged binary subtype is rejected")
     }
 }
 
-TEST_CASE("issue #5676 - SAX parsing of CBOR tags")
-{
-    const json expected = json::binary({1, 2, 3}, 42);
-    const auto cbor = json::to_cbor(expected);
-
-    nlohmann::detail::json_sax_acceptor<json> acceptor;
-    CHECK_FALSE(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor));
-    CHECK_FALSE(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor,
-                                true, false, false, json::cbor_tag_handler_t::error));
-
-    CHECK(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor,
-                          true, false, false, json::cbor_tag_handler_t::ignore));
-
-    json parsed;
-    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> sax(parsed);
-    CHECK(json::sax_parse(cbor, &sax, json::input_format_t::cbor,
-                          true, false, false, json::cbor_tag_handler_t::store));
-    CHECK(parsed == expected);
-
-    json iterator_parsed;
-    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> iterator_sax(iterator_parsed);
-    CHECK(json::sax_parse(cbor.begin(), cbor.end(), &iterator_sax, json::input_format_t::cbor,
-                          true, false, false, json::cbor_tag_handler_t::store));
-    CHECK(iterator_parsed == expected);
-
-#ifndef JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
-    json span_parsed;
-    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> span_sax(span_parsed);
-    CHECK(json::sax_parse(nlohmann::detail::span_input_adapter(cbor.data(), cbor.size()), &span_sax,
-                          json::input_format_t::cbor, true, false, false, json::cbor_tag_handler_t::store));
-    CHECK(span_parsed == expected);
-#endif
-
-    const std::string text = "null";
-    CHECK(json::sax_parse(text, &acceptor, json::input_format_t::json,
-                          true, false, false, json::cbor_tag_handler_t::store));
-    CHECK(json::sax_parse(text.begin(), text.end(), &acceptor, json::input_format_t::json,
-                          true, false, false, json::cbor_tag_handler_t::store));
-#ifndef JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
-    CHECK(json::sax_parse(nlohmann::detail::span_input_adapter(text.data(), text.size()), &acceptor,
-                          json::input_format_t::json, true, false, false, json::cbor_tag_handler_t::store));
-#endif
-}
-
 TEST_CASE("issue #5402 - update(merge_objects=true) overwrites a primitive with an object")
 {
     json t = {{"k", 1}};

tests/src/unit-sax_parse.cpp

@@ -0,0 +1,784 @@
+//     __ _____ _____ _____
+//  __|  |   __|     |   | |  JSON for Modern C++ (supporting code)
+// |  |  |__   |  |  | | | |  version 3.12.0
+// |_____|_____|_____|_|___|  https://github.com/nlohmann/json
+//
+// SPDX-FileCopyrightText: 2013-2026 Niels Lohmann <https://nlohmann.me>
+// SPDX-License-Identifier: MIT
+
+/////////////////////////////////////////////////////////////////////
+// Tests that call basic_json::sax_parse have a file of their own: every
+// sax_parse call instantiates the parser and binary reader that recover from
+// errors (see #3989), and in unit-regression2.cpp, unit-regression3.cpp, and
+// unit-msgpack.cpp this made the objects too large for the MinGW linker to
+// relocate (see #5511).
+/////////////////////////////////////////////////////////////////////
+
+#include "doctest_compatibility.h"
+
+// capture whether JSON_DELETE_DEPRECATED_FUNCTIONS was enabled on the command
+// line *before* including json.hpp, since the library #undefs it once the header
+// has been fully processed (see include/nlohmann/detail/macro_unscope.hpp); the
+// tests of deprecated functions are skipped if these functions are deleted
+#if defined(JSON_DELETE_DEPRECATED_FUNCTIONS) && (JSON_DELETE_DEPRECATED_FUNCTIONS == 1)
+    #define JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
+#endif
+
+#include <nlohmann/json.hpp>
+using json = nlohmann::json;
+
+#include <cmath>
+#include <cstddef>
+#include <cstdint>
+#include <initializer_list>
+#include <map>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "sax_countdown.hpp"
+using utils::SaxCountdown;
+
+// a narrow number_float_t, so that a double read from binary input can
+// overflow it
+using float_json = nlohmann::basic_json<std::map, std::vector, std::string, bool, std::int64_t, std::uint64_t, float>;
+
+DOCTEST_CLANG_SUPPRESS_WARNING_PUSH
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wexit-time-destructors")
+
+namespace
+{
+/// builds a value from SAX events, asks the parser to recover from its first
+/// 100 errors, and checks that the events are balanced (see #3989)
+template<typename BasicJsonType>
+class BasicRecoveringParser
+{
+  public:
+    explicit BasicRecoveringParser(BasicJsonType& j)
+        : dom(j, false)
+    {}
+
+    bool null()
+    {
+        value();
+        return dom.null();
+    }
+
+    bool boolean(bool val)
+    {
+        value();
+        return dom.boolean(val);
+    }
+
+    bool number_integer(typename BasicJsonType::number_integer_t val)
+    {
+        value();
+        return dom.number_integer(val);
+    }
+
+    bool number_unsigned(typename BasicJsonType::number_unsigned_t val)
+    {
+        value();
+        return dom.number_unsigned(val);
+    }
+
+    bool number_float(typename BasicJsonType::number_float_t val, const std::string& s)
+    {
+        value();
+        return dom.number_float(val, s);
+    }
+
+    bool string(std::string& val)
+    {
+        value();
+        return dom.string(val);
+    }
+
+    bool binary(typename BasicJsonType::binary_t& val)
+    {
+        value();
+        return dom.binary(val);
+    }
+
+    bool start_object(std::size_t elements)
+    {
+        value();
+        stack.push_back('o');
+        return dom.start_object(elements);
+    }
+
+    bool key(std::string& val)
+    {
+        if (stack.empty() || stack.back() != 'o')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.back() = 'v';
+        return dom.key(val);
+    }
+
+    bool end_object()
+    {
+        if (stack.empty() || stack.back() != 'o')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.pop_back();
+        return dom.end_object();
+    }
+
+    bool start_array(std::size_t elements)
+    {
+        value();
+        stack.push_back('a');
+        return dom.start_array(elements);
+    }
+
+    bool end_array()
+    {
+        if (stack.empty() || stack.back() != 'a')
+        {
+            well_formed = false;
+            return false;
+        }
+        stack.pop_back();
+        return dom.end_array();
+    }
+
+    bool parse_error(std::size_t /*unused*/, const std::string& /*unused*/, const json::exception& ex)
+    {
+        messages.emplace_back(ex.what());
+        // a limit, so that a reader that does not stop fails the test
+        // instead of making it hang
+        return ++errors < 100;
+    }
+
+    /// whether the events were balanced and every key was followed by a value
+    bool balanced() const
+    {
+        return well_formed && stack.empty();
+    }
+
+    /// builds the value
+    nlohmann::detail::json_sax_dom_parser<BasicJsonType> dom;
+    std::size_t errors = 0;
+    std::vector<std::string> messages {}; // NOLINT(readability-redundant-member-init)
+    std::vector<char> stack {}; // NOLINT(readability-redundant-member-init)
+    bool well_formed = true;
+
+  private:
+    void value()
+    {
+        if (!stack.empty())
+        {
+            if (stack.back() == 'v')
+            {
+                stack.back() = 'o';
+            }
+            else if (stack.back() == 'o')
+            {
+                well_formed = false;
+            }
+        }
+    }
+
+};
+
+using RecoveringParser = BasicRecoveringParser<json>;
+
+struct BinaryParseResult
+{
+    json value;
+    std::size_t errors;
+    std::vector<std::string> messages;
+    bool ok;
+    bool balanced;
+};
+
+BinaryParseResult parse_binary_recovering(const std::vector<std::uint8_t>& input, const json::input_format_t format)
+{
+    json j;
+    RecoveringParser sax(j);
+    const bool ok = json::sax_parse(input, &sax, format);
+    return {j, sax.errors, sax.messages, ok, sax.balanced()};
+}
+
+#if !defined(JSON_NOEXCEPTION)
+/// the message of the exception that reading @a input into a JSON value
+/// throws, or an empty string if reading succeeds
+std::string binary_error_message(const std::vector<std::uint8_t>& input, const json::input_format_t format)
+{
+    try
+    {
+        json _;
+        switch (format)
+        {
+            case json::input_format_t::cbor:
+                _ = json::from_cbor(input);
+                break;
+            case json::input_format_t::msgpack:
+                _ = json::from_msgpack(input);
+                break;
+            case json::input_format_t::ubjson:
+                _ = json::from_ubjson(input);
+                break;
+            case json::input_format_t::bjdata:
+                _ = json::from_bjdata(input);
+                break;
+            case json::input_format_t::bson:
+                _ = json::from_bson(input);
+                break;
+            case json::input_format_t::bon8:
+                _ = json::from_bon8(input);
+                break;
+            case json::input_format_t::json:
+            default:
+                break;
+        }
+    }
+    catch (const json::exception& e)
+    {
+        return e.what();
+    }
+    return "";
+}
+#endif
+
+/// a BSON element: its type, its name, and its value
+std::vector<std::uint8_t> bson_element(const std::uint8_t type, const std::string& name, const std::vector<std::uint8_t>& value)
+{
+    std::vector<std::uint8_t> result = {type};
+    result.insert(result.end(), name.begin(), name.end());
+    result.push_back(0x00);
+    result.insert(result.end(), value.begin(), value.end());
+    return result;
+}
+
+/// a BSON document of the given elements; @a size_offset is added to the
+/// size it declares
+std::vector<std::uint8_t> bson_document(const std::vector<std::vector<std::uint8_t>>& elements, const int size_offset = 0)
+{
+    std::vector<std::uint8_t> body;
+    for (const auto& element : elements)
+    {
+        body.insert(body.end(), element.begin(), element.end());
+    }
+    const auto size = static_cast<std::uint32_t>(static_cast<int>(body.size()) + 5 + size_offset);
+    std::vector<std::uint8_t> result = {static_cast<std::uint8_t>(size & 0xFFu), static_cast<std::uint8_t>((size >> 8u) & 0xFFu),
+                                        static_cast<std::uint8_t>((size >> 16u) & 0xFFu), static_cast<std::uint8_t>((size >> 24u) & 0xFFu)
+                                       };
+    result.insert(result.end(), body.begin(), body.end());
+    result.push_back(0x00);
+    return result;
+}
+
+/// a BSON int32 value
+std::vector<std::uint8_t> bson_int32(const std::int32_t value)
+{
+    const auto u = static_cast<std::uint32_t>(value);
+    return {static_cast<std::uint8_t>(u & 0xFFu), static_cast<std::uint8_t>((u >> 8u) & 0xFFu),
+            static_cast<std::uint8_t>((u >> 16u) & 0xFFu), static_cast<std::uint8_t>((u >> 24u) & 0xFFu)};
+}
+
+/// a BSON string value, whose length is @a length_offset off
+std::vector<std::uint8_t> bson_string(const std::string& value, const std::int32_t length_offset = 0)
+{
+    auto result = bson_int32(static_cast<std::int32_t>(value.size() + 1) + length_offset);
+    result.insert(result.end(), value.begin(), value.end());
+    result.push_back(0x00);
+    return result;
+}
+
+/// @a count bytes of value 0xAB
+std::vector<std::uint8_t> bytes(const std::size_t count)
+{
+    return std::vector<std::uint8_t>(count, 0xAB);
+}
+
+/// the bytes of @a parts, one after the other
+std::vector<std::uint8_t> concatenated(std::initializer_list<std::vector<std::uint8_t>> parts)
+{
+    std::vector<std::uint8_t> result;
+    for (const auto& part : parts)
+    {
+        result.insert(result.end(), part.begin(), part.end());
+    }
+    return result;
+}
+
+/// U+FFFD REPLACEMENT CHARACTER
+std::string replacement_character()
+{
+    return "\xEF\xBF\xBD";
+}
+}  // namespace
+
+TEST_CASE("regression test - #3989 SAX parse_error() returning true")
+{
+    SECTION("binary formats complete what was read before the input ends")
+    {
+        const json j = {{"a", {1, -2, {{"b", "c"}}, json::array()}}, {"d", {{"e", nullptr}, {"f", true}}}, {"g", 1.5}, {"h", json::binary({1, 2, 3})}};
+
+        const std::vector<std::pair<json::input_format_t, std::vector<std::uint8_t>>> encodings =
+        {
+            {json::input_format_t::cbor, json::to_cbor(j)},
+            {json::input_format_t::msgpack, json::to_msgpack(j)},
+            {json::input_format_t::ubjson, json::to_ubjson(j)},
+            {json::input_format_t::ubjson, json::to_ubjson(j, true, true)},
+            {json::input_format_t::bjdata, json::to_bjdata(j)},
+            {json::input_format_t::bjdata, json::to_bjdata(j, true, true)},
+            {json::input_format_t::bson, json::to_bson(j)},
+            {json::input_format_t::bon8, json::to_bon8(j)},
+        };
+
+        for (const auto& encoding : encodings)
+        {
+            const auto format = encoding.first;
+            const auto& bytes = encoding.second;
+            CAPTURE(format)
+
+            // every prefix is truncated input
+            for (std::size_t length = 0; length < bytes.size(); ++length)
+            {
+                CAPTURE(length)
+                const auto result = parse_binary_recovering(std::vector<std::uint8_t>(bytes.begin(), bytes.begin() + static_cast<std::ptrdiff_t>(length)), format);
+                CHECK(!result.ok);
+                CHECK(result.errors == 1);
+                CHECK(result.balanced);
+            }
+
+            // the complete input is read as usual (binary values do not
+            // round-trip through every format, so compare with a plain parse)
+            json expected;
+            nlohmann::detail::json_sax_dom_parser<json> dom(expected);
+            CHECK(json::sax_parse(bytes, &dom, format));
+            const auto complete = parse_binary_recovering(bytes, format);
+            CHECK(complete.ok);
+            CHECK(complete.errors == 0);
+            CHECK(complete.value == expected);
+
+            // a byte after the value
+            auto trailing_bytes = bytes;
+            trailing_bytes.push_back(0x01);
+            const auto trailing = parse_binary_recovering(trailing_bytes, format);
+            CHECK(!trailing.ok);
+            CHECK(trailing.errors == 1);
+            CHECK(trailing.value == expected);
+        }
+    }
+
+    SECTION("containers without an end")
+    {
+        // these made the readers loop, or read on, after the error
+        const auto cbor_array = parse_binary_recovering({0x9F}, json::input_format_t::cbor);
+        CHECK(cbor_array.errors == 1);
+        CHECK(cbor_array.value == json::array());
+
+        const auto cbor_map = parse_binary_recovering({0xBF, 0x61, 'a'}, json::input_format_t::cbor);
+        CHECK(cbor_map.errors == 1);
+        CHECK(cbor_map.value == json({{"a", nullptr}}));
+
+        // 2^32-1 elements: on a 32-bit target this exceeds the array's max_size(),
+        // so the array is rejected before it is started and the value stays null
+        const auto msgpack_array = parse_binary_recovering({0xDD, 0xFF, 0xFF, 0xFF, 0xFF}, json::input_format_t::msgpack);
+        const bool msgpack_array_fits = json::array_t().max_size() >= 0xFFFFFFFFu;
+        CHECK(msgpack_array.errors == 1);
+        CHECK(msgpack_array.value == (msgpack_array_fits ? json::array() : json(nullptr)));
+
+        const auto msgpack_map = parse_binary_recovering({0x81, 0xA1, 'a', 0x92, 0x01}, json::input_format_t::msgpack);
+        CHECK(msgpack_map.errors == 1);
+        CHECK(msgpack_map.value == json({{"a", {1}}}));
+    }
+
+    SECTION("BJData ndarray")
+    {
+        // a 2x3 int8 array with two of its six elements; the annotated array
+        // format opens an object and two arrays of its own
+        const auto result = parse_binary_recovering({'[', '$', 'i', '#', '[', '$', 'i', '#', 'i', 2, 2, 3, 1, 2}, json::input_format_t::bjdata);
+        CHECK(result.errors == 1);
+        CHECK(result.balanced);
+        CHECK(result.value == json({{"_ArrayType_", "int8"}, {"_ArraySize_", {2, 3}}, {"_ArrayData_", {1, 2}}}));
+    }
+
+    SECTION("binary formats repair items whose end is known")
+    {
+        struct Repair
+        {
+            json::input_format_t format;
+            std::vector<std::uint8_t> input;
+            json expected;
+            std::size_t errors;
+        };
+
+        const std::vector<Repair> repairs =
+        {
+            // CBOR: tags are ignored (here tag 1 and the self-describe tag 55799)
+            {json::input_format_t::cbor, {0x82, 0xC1, 0x05, 0xD9, 0xD9, 0xF7, 0x06}, {5, 6}, 2},
+            // CBOR: undefined and other simple values become null
+            {json::input_format_t::cbor, {0x84, 0xF7, 0xE0, 0xF8, 0x20, 0x01}, {nullptr, nullptr, nullptr, 1}, 3},
+            // CBOR: members whose key is not a string are skipped, whatever their key and value
+            {json::input_format_t::cbor, {0xA4, 0x01, 0x02, 0x82, 0x01, 0x02, 0xA1, 0x61, 'x', 0x9F, 0xFF, 0xC1, 0x01, 0x5F, 0x41, 0x00, 0xFF, 0x61, 'a', 0x03}, {{"a", 3}}, 3},
+            {json::input_format_t::cbor, {0xBF, 0xF5, 0xBF, 0x61, 'x', 0x7F, 0x61, 'y', 0xFF, 0xFF, 0x61, 'a', 0x03, 0xFF}, {{"a", 3}}, 1},
+            // MessagePack: members whose key is not a string are skipped
+            {json::input_format_t::msgpack, {0x84, 0x01, 0x02, 0x81, 0xA1, 'x', 0x01, 0x92, 0x01, 0x02, 0xD4, 0x01, 0x02, 0xC0, 0xA1, 'a', 0x04}, {{"a", 4}}, 3},
+            // UBJSON: a char that is not ASCII becomes U+FFFD
+            {json::input_format_t::ubjson, {'[', 'C', 0x80, 'C', 'A', ']'}, {replacement_character(), "A"}, 1},
+            // UBJSON: the longest beginning of a high-precision number is kept
+            {json::input_format_t::ubjson, {'[', 'H', 'i', 5, '1', '2', 'a', 'b', 'c', 'H', 'i', 2, '1', '.', 'H', 'i', 3, 'a', 'b', 'c', 'H', 'i', 3, '4', '.', '5', ']'}, {12, 1, nullptr, 4.5}, 3},
+            // BJData, too
+            {json::input_format_t::bjdata, {'[', 'C', 0xFF, 'H', 'i', 2, '-', '1', 'H', 'i', 2, '-', 'x', ']'}, {replacement_character(), -1, nullptr}, 2},
+            // a NUL ends a high-precision number, as it ends JSON text
+            {json::input_format_t::ubjson, {'[', 'H', 'i', 4, '1', '2', 0, '9', 'H', 'i', 2, 0, '1', 'i', 3, ']'}, {12, nullptr, 3}, 2},
+            // BON8: members whose key is not a string are skipped
+            {json::input_format_t::bon8, {0x89, 0x91, 0x92, 0xC9, 0x40, 0x82, 0x91, 0x92, 0x61, 0x93}, {{"a", 3}}, 2},
+            {json::input_format_t::bon8, {0x8B, 0x91, 0x85, 0x91, 0xFE, 0xFA, 0x8B, 'x', 0x91, 0xFE, 0x61, 0x93, 0xFE}, {{"a", 3}}, 2},
+            // BSON: elements of types the library does not read become null
+            {
+                json::input_format_t::bson, bson_document(
+                {
+                    bson_element(0x07, "_id", bytes(12)),                         // ObjectId
+                    bson_element(0x09, "date", bytes(8)),                         // UTC datetime
+                    bson_element(0x13, "decimal", bytes(16)),                     // 128-bit decimal
+                    bson_element(0x0B, "regex", {'a', '+', 0, 'i', 0}),           // regular expression
+                    bson_element(0x0D, "code", bson_string("f()")),               // JavaScript code
+                    bson_element(0x0E, "symbol", bson_string("s")),               // symbol
+                    bson_element(0x0C, "pointer", concatenated({bson_string("c"), bytes(12)})), // DBPointer
+                    bson_element(0x0F, "scope", concatenated({bson_int32(15), bson_string("g"), bson_document({})})), // code with scope
+                    bson_element(0x06, "undefined", {}),                          // undefined
+                    bson_element(0xFF, "min", {}),                                // min key
+                    bson_element(0x7F, "max", {}),                                // max key
+                    bson_element(0x10, "z", bson_int32(7)),
+                }),
+                {{"_id", nullptr}, {"date", nullptr}, {"decimal", nullptr}, {"regex", nullptr}, {"code", nullptr}, {"symbol", nullptr}, {"pointer", nullptr}, {"scope", nullptr}, {"undefined", nullptr}, {"min", nullptr}, {"max", nullptr}, {"z", 7}},
+                11
+            },
+            // BSON: an element of an unknown type becomes null, and the rest of its document is skipped
+            {
+                json::input_format_t::bson, bson_document(
+                {
+                    bson_element(0x03, "inner", bson_document({bson_element(0x10, "a", bson_int32(1)), bson_element(0x42, "x", bytes(3)), bson_element(0x10, "b", bson_int32(2))})),
+                    bson_element(0x04, "array", bson_document({bson_element(0x10, "0", bson_int32(1)), bson_element(0x42, "1", bytes(3))})),
+                    bson_element(0x10, "after", bson_int32(3)),
+                }),
+                {{"inner", {{"a", 1}, {"x", nullptr}}}, {"array", {1, nullptr}}, {"after", 3}},
+                2
+            },
+            // BSON: so does a string or byte array whose length cannot be right
+            {
+                json::input_format_t::bson, bson_document(
+                {
+                    bson_element(0x03, "inner", bson_document({bson_element(0x02, "s", bson_string("abc", -10)), bson_element(0x10, "b", bson_int32(2))})),
+                    bson_element(0x03, "bin", bson_document({bson_element(0x05, "b", concatenated({bson_int32(-1), bytes(1)})), bson_element(0x10, "b", bson_int32(2))})),
+                    bson_element(0x10, "after", bson_int32(3)),
+                }),
+                {{"inner", {{"s", nullptr}}}, {"bin", {{"b", nullptr}}}, {"after", 3}},
+                2
+            },
+            // BSON: a string without its terminator, and a document whose size does not match, are kept
+            {
+                json::input_format_t::bson, bson_document(
+                {
+                    bson_element(0x02, "s", {2, 0, 0, 0, 'a', 'X'}),
+                    bson_element(0x03, "inner", bson_document({bson_element(0x10, "a", bson_int32(1))}, 1)),
+                }),
+                {{"s", "a"}, {"inner", {{"a", 1}}}},
+                2
+            },
+            // BSON: so is a document whose size is too small for its content,
+            // which is read up to its terminator
+            {
+                json::input_format_t::bson, bson_document(
+                {
+                    bson_element(0x03, "inner", bson_document({bson_element(0x10, "a", bson_int32(1)), bson_element(0x10, "b", bson_int32(2))}, -7)),
+                    bson_element(0x10, "after", bson_int32(3)),
+                }),
+                {{"inner", {{"a", 1}, {"b", 2}}}, {"after", 3}},
+                1
+            },
+        };
+
+        for (const auto& repair : repairs)
+        {
+            CAPTURE(repair.format)
+            CAPTURE(repair.input)
+            const auto result = parse_binary_recovering(repair.input, repair.format);
+            CHECK(!result.ok);
+            CHECK(result.balanced);
+            CHECK(result.errors == repair.errors);
+            CHECK(result.value == repair.expected);
+            REQUIRE(!result.messages.empty());
+#if !defined(JSON_NOEXCEPTION)
+            // the first error is the one reported without recovering; under
+            // JSON_NOEXCEPTION, reading without recovering aborts instead of
+            // throwing, so there is no message to compare with
+            CHECK(result.messages.front() == binary_error_message(repair.input, repair.format));
+#endif
+        }
+    }
+
+    SECTION("binary formats repair numbers that are out of range")
+    {
+        // CBOR: a double too large for a float number_float_t
+        float_json cbor;
+        BasicRecoveringParser<float_json> sax(cbor);
+        const std::vector<std::uint8_t> cbor_input = {0x82, 0xFB, 0x7E, 0x37, 0xE4, 0x3C, 0x88, 0x00, 0x75, 0x9C, 0x01}; // [1e300, 1]
+        CHECK(!float_json::sax_parse(cbor_input, &sax, float_json::input_format_t::cbor));
+        CHECK(sax.errors == 1);
+        CHECK(sax.messages.front() == "[json.exception.out_of_range.406] syntax error while parsing CBOR value: number overflow");
+        REQUIRE(cbor.size() == 2);
+        CHECK(std::isinf(cbor[0].get<float>()));
+        CHECK(cbor[1] == 1);
+
+        // UBJSON: a high-precision number too large for number_float_t
+        const auto ubjson = parse_binary_recovering({'H', 'i', 5, '1', 'e', '9', '9', '9'}, json::input_format_t::ubjson);
+        CHECK(ubjson.errors == 1);
+        CHECK(ubjson.value.is_number_float());
+        CHECK(std::isinf(ubjson.value.get<double>()));
+    }
+
+    SECTION("binary formats stop where the end of an item is not known")
+    {
+        // a byte that begins no item
+        const auto cbor = parse_binary_recovering({0x82, 0x01, 0x1C, 0x02}, json::input_format_t::cbor);
+        CHECK(cbor.errors == 1);
+        CHECK(cbor.value == json({1}));
+
+        // a key that is no item: the unused MessagePack byte, a CBOR break
+        // in a map of known size, and the end of a BON8 container
+        const auto msgpack = parse_binary_recovering({0x82, 0xA1, 'a', 0x01, 0xC1, 0x02}, json::input_format_t::msgpack);
+        CHECK(msgpack.errors == 1);
+        CHECK(msgpack.value == json({{"a", 1}}));
+        const auto cbor_break = parse_binary_recovering({0xA2, 0x61, 'a', 0x01, 0xFF, 0x02}, json::input_format_t::cbor);
+        CHECK(cbor_break.errors == 1);
+        CHECK(cbor_break.value == json({{"a", 1}}));
+        const auto bon8 = parse_binary_recovering({0x88, 0x61, 0x91, 0xFE}, json::input_format_t::bon8);
+        CHECK(bon8.errors == 1);
+        CHECK(bon8.value == json({{"a", 1}}));
+
+        // an indefinite-length string inside an indefinite-length string
+        const auto nested = parse_binary_recovering({0x82, 0x01, 0x7F, 0x7F, 0x61, 'a', 0xFF, 0xFF}, json::input_format_t::cbor);
+        CHECK(nested.errors == 1);
+        CHECK(nested.value == json({1}));
+
+        // a BJData ndarray whose element type is not valid: the type is
+        // checked before the object that would hold the ndarray is begun
+        const auto ndarray = parse_binary_recovering({'[', '$', 0x01, '#', '[', '$', 'i', '#', 'i', 2, 2, 3}, json::input_format_t::bjdata);
+        CHECK(ndarray.errors == 1);
+        CHECK(ndarray.balanced);
+        CHECK(ndarray.value == nullptr);
+
+        // a skipped member that the input ends in
+        const auto truncated = parse_binary_recovering({0xA2, 0x01, 0x82, 0x01}, json::input_format_t::cbor);
+        CHECK(truncated.errors == 2);
+        CHECK(truncated.balanced);
+        CHECK(truncated.value == json::object());
+
+        // a BSON element of an unknown type in a document whose size cannot be
+        // right: the size is reported when it is read, and the element ends
+        // the document, as its end is not known
+        const auto bson = parse_binary_recovering(bson_document({bson_element(0x10, "a", bson_int32(1)), bson_element(0x42, "x", bytes(3))}, -10), json::input_format_t::bson);
+        CHECK(bson.errors == 2);
+        CHECK(bson.value == json({{"a", 1}, {"x", nullptr}}));
+    }
+
+    SECTION("changed bytes in binary input")
+    {
+        const json j = {{"a", {1, -2, {{"b", "c"}}, json::array()}}, {"d", {{"e", nullptr}, {"f", true}}}, {"g", 1.5}, {"h", json::binary({1, 2, 3})}, {"i", "\xC3\xA4"}};
+
+        const std::vector<std::pair<json::input_format_t, std::vector<std::uint8_t>>> encodings =
+        {
+            {json::input_format_t::cbor, json::to_cbor(j)},
+            {json::input_format_t::msgpack, json::to_msgpack(j)},
+            {json::input_format_t::ubjson, json::to_ubjson(j)},
+            {json::input_format_t::ubjson, json::to_ubjson(j, true, true)},
+            {json::input_format_t::bjdata, json::to_bjdata(j)},
+            {json::input_format_t::bjdata, json::to_bjdata(j, true, true)},
+            {json::input_format_t::bson, json::to_bson(j)},
+            {json::input_format_t::bon8, json::to_bon8(j)},
+        };
+        const std::vector<std::uint8_t> replacements = {0x00, 0x01, 0x7F, 0x80, 0xC1, 0xD9, 0xE0, 0xF7, 0xFE, 0xFF};
+
+        for (const auto& encoding : encodings)
+        {
+            const auto format = encoding.first;
+            const auto& original = encoding.second;
+            CAPTURE(format)
+
+            std::vector<std::vector<std::uint8_t>> inputs;
+            for (std::size_t position = 0; position < original.size(); ++position)
+            {
+                for (const auto replacement : replacements)
+                {
+                    auto changed = original;
+                    changed[position] = replacement;
+                    inputs.push_back(changed);
+                }
+                auto removed = original;
+                removed.erase(removed.begin() + static_cast<std::ptrdiff_t>(position));
+                inputs.push_back(removed);
+            }
+
+            for (const auto& input : inputs)
+            {
+                CAPTURE(input)
+                const auto result = parse_binary_recovering(input, format);
+                CHECK(result.balanced);
+                CHECK(result.errors <= input.size() + 1);
+#if !defined(JSON_NOEXCEPTION)
+                // an error is reported exactly if reading into a JSON value
+                // fails, and the first one is the same (under JSON_NOEXCEPTION,
+                // that reading aborts instead of throwing)
+                const auto message = binary_error_message(input, format);
+                CHECK(result.ok == message.empty());
+                if (!result.ok && result.errors < 100)
+                {
+                    CHECK(result.messages.front() == message);
+                }
+#endif
+            }
+        }
+    }
+
+    SECTION("JSON text")
+    {
+        // the parser stopped, but reported success
+        json j;
+        RecoveringParser sax(j);
+        CHECK(!json::sax_parse("[1,2,3,]", &sax));
+        CHECK(sax.errors == 1);
+        CHECK(j == json({1, 2, 3}));
+    }
+
+    SECTION("the SAX parsers of the library stop")
+    {
+        json _;
+        CHECK(json::from_cbor(std::vector<std::uint8_t> {0x9F}, true, false).is_discarded());
+        CHECK_THROWS_WITH_AS(_ = json::from_cbor(std::vector<std::uint8_t> {0x9F}), "[json.exception.parse_error.110] parse error at byte 2: syntax error while parsing CBOR value: unexpected end of input", json::parse_error&);
+        CHECK(json::parse("[1,2,3,]", nullptr, false).is_discarded());
+        CHECK(!json::accept("[1,2,3,]"));
+    }
+}
+
+
+TEST_CASE("issue #5676 - SAX parsing of CBOR tags")
+{
+    const json expected = json::binary({1, 2, 3}, 42);
+    const auto cbor = json::to_cbor(expected);
+
+    nlohmann::detail::json_sax_acceptor<json> acceptor;
+    CHECK_FALSE(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor));
+    CHECK_FALSE(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor,
+                                true, false, false, json::cbor_tag_handler_t::error));
+
+    CHECK(json::sax_parse(cbor, &acceptor, json::input_format_t::cbor,
+                          true, false, false, json::cbor_tag_handler_t::ignore));
+
+    json parsed;
+    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> sax(parsed);
+    CHECK(json::sax_parse(cbor, &sax, json::input_format_t::cbor,
+                          true, false, false, json::cbor_tag_handler_t::store));
+    CHECK(parsed == expected);
+
+    json iterator_parsed;
+    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> iterator_sax(iterator_parsed);
+    CHECK(json::sax_parse(cbor.begin(), cbor.end(), &iterator_sax, json::input_format_t::cbor,
+                          true, false, false, json::cbor_tag_handler_t::store));
+    CHECK(iterator_parsed == expected);
+
+#ifndef JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
+    json span_parsed;
+    nlohmann::detail::json_sax_dom_parser<json, nlohmann::detail::string_input_adapter_type> span_sax(span_parsed);
+    CHECK(json::sax_parse(nlohmann::detail::span_input_adapter(cbor.data(), cbor.size()), &span_sax,
+                          json::input_format_t::cbor, true, false, false, json::cbor_tag_handler_t::store));
+    CHECK(span_parsed == expected);
+#endif
+
+    const std::string text = "null";
+    CHECK(json::sax_parse(text, &acceptor, json::input_format_t::json,
+                          true, false, false, json::cbor_tag_handler_t::store));
+    CHECK(json::sax_parse(text.begin(), text.end(), &acceptor, json::input_format_t::json,
+                          true, false, false, json::cbor_tag_handler_t::store));
+#ifndef JSON_TEST_DEPRECATED_FUNCTIONS_DELETED
+    CHECK(json::sax_parse(nlohmann::detail::span_input_adapter(text.data(), text.size()), &acceptor,
+                          json::input_format_t::json, true, false, false, json::cbor_tag_handler_t::store));
+#endif
+}
+
+
+TEST_CASE("MessagePack SAX aborts")
+{
+    SECTION("start_array(len)")
+    {
+        std::vector<uint8_t> const v = {0x93, 0x01, 0x02, 0x03};
+        SaxCountdown scp(0);
+        CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
+    }
+
+    SECTION("start_object(len)")
+    {
+        std::vector<uint8_t> const v = {0x81, 0xa3, 0x66, 0x6F, 0x6F, 0xc2};
+        SaxCountdown scp(0);
+        CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
+    }
+
+    SECTION("key()")
+    {
+        std::vector<uint8_t> const v = {0x81, 0xa3, 0x66, 0x6F, 0x6F, 0xc2};
+        SaxCountdown scp(1);
+        CHECK(!json::sax_parse(v, &scp, json::input_format_t::msgpack));
+    }
+}
+
+TEST_CASE("issue #5405 - a user-defined SAX consumer is unaffected by the internal DOM reserve optimization")
+{
+    // the reserve() call is local to json_sax_dom_parser / json_sax_dom_callback_parser;
+    // a custom SAX consumer that does not touch a DOM array sees identical events
+    json j = json::array();
+    for (int i = 0; i < 100; ++i)
+    {
+        j.push_back(i);
+    }
+    const auto packed = json::to_msgpack(j);
+
+    SaxCountdown scp(1000000); // large enough to never trigger an abort
+    CHECK(json::sax_parse(packed, &scp, json::input_format_t::msgpack));
+}
+
+TEST_CASE("MessagePack nesting does not consume the call stack - SAX interface")
+{
+    // see the test case of the same name in unit-msgpack.cpp (#5104)
+    std::vector<uint8_t> input(300000, 0x91);
+    input.push_back(0x01); // innermost value
+
+    SaxCountdown accept_all(600001);
+    CHECK(json::sax_parse(input, &accept_all, json::input_format_t::msgpack));
+}
+
+TEST_CASE("MessagePack SAX parsing stops at every event")
+{
+    // Containers are opened and closed by the loop that reads them; a SAX
+    // handler that rejects any event - including the end of a nested
+    // container - must stop the parse right there.
+    const auto count_events = [](const std::vector<std::uint8_t>& input)
+    {
+        int events = 0;
+        while (true)
+        {
+            SaxCountdown scp(events);
+            if (json::sax_parse(input, &scp, json::input_format_t::msgpack))
+            {
+                return events;
+            }
+            ++events;
+            REQUIRE(events < 1000);
+        }
+    };
+
+    // 20 events: every container kind closes inside another one
+    const json j = json::parse(R"({"a": [1, {"b": []}], "c": {"d": [[2]]}})");
+    CHECK(count_events(json::to_msgpack(j)) == 20);
+}
+
+DOCTEST_CLANG_SUPPRESS_WARNING_POP

tests/src/unit-stream-bulk-scan.cpp

@@ -120,6 +120,7 @@ class oversized_get_area_streambuf : public std::streambuf
   private:
     std::array<char, 8> m_data{{'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h'}};
 };
+
 #endif
 
 // how a test reads from the stream
@@ -463,15 +464,18 @@ TEST_CASE("parsing from a std::istream scans the get area in bulk")
         CHECK(token.empty());
     }
 
+#if SIZE_MAX > UINT32_MAX
+    // a get area larger than an int needs a 64-bit std::size_t (checked by the
+    // preprocessor, as a constant condition triggers MSVC's C4127)
     SECTION("skipping more characters than gbump() takes at once")
     {
         // gbump() takes an int, but a get area can be larger
-        const std::size_t claimed = sizeof(std::size_t) < 8 ? 0 : std::size_t{3} << 30u;
+        const std::size_t claimed = std::size_t{3} << 30u;
         oversized_get_area_streambuf buf(claimed);
-        if (claimed == 0 || static_cast<std::size_t>(buf.in_avail()) != claimed)
+        if (static_cast<std::size_t>(buf.in_avail()) != claimed)
         {
-            // a 32-bit size_t, or a standard library whose streambuf stores
+            // a standard library whose streambuf stores the size of the get
-            // the size of the get area in an int (MSVC)
+            // area in an int (MSVC)
             MESSAGE("a get area larger than an int is not supported");
         }
         else
@@ -491,6 +495,7 @@ TEST_CASE("parsing from a std::istream scans the get area in bulk")
             buf.restore_get_area();
         }
     }
+#endif
 #endif
 
     SECTION("SAX parsing from a stream")

tests/src/unit-ubjson.cpp

@@ -2385,6 +2385,18 @@ TEST_CASE("UBJSON optimized arrays of a valueless type are bounded")
         }
     }
 
+    SECTION("a count equal to SIZE_MAX is rejected")
+    {
+        // the reader marks a container without a count by SIZE_MAX, so a
+        // count of that value must not be read as "no count"; UBJSON's
+        // largest count type (int64) can only reach it where size_t is 32 bits
+#if SIZE_MAX == 0xffffffff
+        const std::vector<uint8_t> input = {'[', '#', 'L', 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 'i', 0x01, ']'};
+        CHECK_THROWS_WITH_AS(_ = json::from_ubjson(input), "[json.exception.out_of_range.408] syntax error while parsing UBJSON size: integer value overflow", json::out_of_range&);
+        CHECK(json::from_ubjson(input, true, false).is_discarded());
+#endif
+    }
+
     SECTION("ordinary counts are unaffected")
     {
         CHECK(json::from_ubjson(std::vector<uint8_t>({'[', '$', 'Z', '#', 'i', 3})) == json({nullptr, nullptr, nullptr}));