Pin CI pip installs by hash and update KaTeX/DOMPurify

The pip installs in check_amalgamation.yml, ubuntu.yml (Meson) and semgrep.yml now use --require-hashes with requirements files compiled by pip-compile --generate-hashes (Python 3.14, as on ubuntu-26.04). The astyle and generate_natvis requirements move to requirements.in, and Meson and Semgrep get theirs in .github/requirements, which Dependabot now watches. check_amalgamation.yml installs the natvis requirements from the PR tree, as it already does for astyle. The Mermaid checker overrides KaTeX to 0.18.11 (GHSA-238p-pmpm-9mq7; Mermaid 11 still requires ^0.16.47) and updates DOMPurify to 3.4.17 (GHSA-cg2g-xh64-v4q7, GHSA-mrv2-3pqr-j5jr, GHSA-c6gr-qvv4-5v3f). Signed-off-by: Niels Lohmann <mail@nlohmann.me>
Bump jsdom from 30.1.2 to 30.1.3 in /docs/mkdocs/scripts/mermaid (#5898 )
2026-10-23 11:44:43 +00:00 · 2026-10-23 07:20:44 +02:00 · 2026-10-23 06:54:23 +02:00 · 2026-10-23 06:54:07 +02:00 · 2026-10-23 06:53:50 +02:00 · 2026-10-23 06:53:32 +02:00
30 changed files with 8392 additions and 1190 deletions
@@ -29,6 +29,13 @@ updates:
      - dependency-name: mermaid
        update-types: ["version-update:semver-major"]

+  - package-ecosystem: pip
+    directory: /.github/requirements
+    schedule:
+      interval: daily
+    cooldown:
+      default-days: 7
+
  - package-ecosystem: pip
    directory: /tools/astyle
    schedule:
@@ -36,7 +43,7 @@ updates:
    cooldown:
      default-days: 7
    ignore:
-      # astyle is deliberately pinned (see tools/astyle/requirements.txt): newer versions
+      # astyle is deliberately pinned (see tools/astyle/requirements.in): newer versions
      # reformat the code, and the pinned version defines the formatting CI enforces
      - dependency-name: astyle

@@ -0,0 +1 @@
+meson==1.12.1
@@ -0,0 +1,10 @@
+#
+# This file is autogenerated by pip-compile with Python 3.14
+# by the following command:
+#
+#    pip-compile --generate-hashes --no-emit-index-url --output-file=.github/requirements/meson.txt --strip-extras .github/requirements/meson.in
+#
+meson==1.12.1 \
+    --hash=sha256:930bc7542cbd9f57009e182fd014eba48cf1a0180a7b9006d2d32d8f168d8b02 \
+    --hash=sha256:ab0a6ca09f8ef70c564c8241fb5a23957886a0b53fb58412b5e07eaf07dba743
+    # via -r .github/requirements/meson.in
@@ -0,0 +1 @@
+semgrep==1.168.0
@@ -67,10 +67,10 @@ jobs:
      - name: Install astyle
        run: |
          python3 -mvenv venv
-          venv/bin/pip3 install -r $MAIN_DIR/tools/astyle/requirements.txt
+          venv/bin/pip3 install --require-hashes -r $MAIN_DIR/tools/astyle/requirements.txt

      - name: Install generate_natvis dependencies
-        run: pip3 install -r $NATVIS_TOOL_DIR/requirements.txt
+        run: pip3 install --require-hashes -r $MAIN_DIR/tools/generate_natvis/requirements.txt

      - name: Regenerate the tools/macro_builder tables in macro_scope.hpp
        run: |
@@ -38,14 +38,14 @@ jobs:

    # Initializes the CodeQL tools for scanning.
    - name: Initialize CodeQL
-      uses: github/codeql-action/init@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+      uses: github/codeql-action/init@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
      with:
        languages: c-cpp

    # Autobuild attempts to build any compiled languages  (C/C++, C#, or Java).
    # If this step fails, then you should remove it and run the build manually (see below)
    - name: Autobuild
-      uses: github/codeql-action/autobuild@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+      uses: github/codeql-action/autobuild@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0

    - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+      uses: github/codeql-action/analyze@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
@@ -47,6 +47,6 @@ jobs:
          output: 'flawfinder_results.sarif'

      - name: Upload analysis results to GitHub Security tab
-        uses: github/codeql-action/upload-sarif@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+        uses: github/codeql-action/upload-sarif@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
        with:
          sarif_file: ${{github.workspace}}/flawfinder_results.sarif
@@ -54,7 +54,7 @@ jobs:
    timeout-minutes: 60
    steps:
      - name: Harden Runner
-        uses: step-security/harden-runner@e14015d583714f6e62063499dc959a02595150a1 # v2.21.1
+        uses: step-security/harden-runner@ccd8616d44fd3846e67624a50d5aad6d37bf2d25 # v2.22.1
        with:
          egress-policy: audit

@@ -78,7 +78,7 @@ jobs:
      - name: Summary
        run: cat tests/benchmarks/json_view/results/*.md >> "$GITHUB_STEP_SUMMARY"

-      - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+      - uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
        with:
          name: json_view-benchmarks-${{ matrix.runner }}-${{ inputs.compiler || 'g++' }}
          path: tests/benchmarks/json_view/results/
@@ -80,6 +80,6 @@ jobs:

      # Upload the results to GitHub's code scanning dashboard.
      - name: "Upload to code-scanning"
-        uses: github/codeql-action/upload-sarif@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+        uses: github/codeql-action/upload-sarif@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
        with:
          sarif_file: results.sarif
@@ -53,7 +53,7 @@ jobs:
      # SEMGREP_APP_TOKEN is still passed through so registry auth works if a
      # token is ever added.
      - name: Install Semgrep
-        run: python3 -m pip install --user semgrep==1.168.0
+        run: python3 -m pip install --user --require-hashes -r .github/requirements/semgrep.txt

      # `semgrep scan --sarif` always exits 0 even with findings; continue-on-error
      # is a safety net so the SARIF upload still runs if the scan itself errors.
@@ -65,7 +65,7 @@ jobs:

      # Upload SARIF file generated in previous step
      - name: Upload SARIF file
-        uses: github/codeql-action/upload-sarif@24c54180a607b1449ed407dd24f251e4e9147c8d # v4.38.3
+        uses: github/codeql-action/upload-sarif@6f2ac81f5cd81fe351a197ba0b00d91d03cd2aee # v4.39.0
        with:
          sarif_file: semgrep.sarif
        if: always()
@@ -42,7 +42,7 @@ jobs:
      - name: Check that Meson and pkg-config offer the CMake options
        run: make check_build_options
      - name: Install Meson
-        run: pip install meson
+        run: pip install --require-hashes -r .github/requirements/meson.txt
      - name: Install with Meson
        run: |
          meson setup build-meson --prefix=${{ github.workspace }}/install
@@ -339,6 +339,11 @@ The `serializer` (used by [`dump`](../api/basic_json/dump.md) and [`operator<<`]
 - `output_stream_adapter` writes to a `std::ostream`.
 - `output_string_adapter` appends to a string.

+The `binary_writer` does not call an adapter for every byte. The `to_*` functions that return a `std::vector` write
+into the vector through a sink (`output_vector_sink`) that is not virtual. The functions that take an output adapter
+write through a sink (`output_adapter_sink`) that collects the output in a buffer of one kibibyte and passes it to the
+adapter in blocks, so that the adapter is called (and a stream locked) once per block instead of once per value.
+
 ## Value conversion

 Values are converted from and to other types with the `JSONSerializer` template parameter. The default,
@@ -6,7 +6,7 @@
    "": {
      "name": "check-mermaid",
      "dependencies": {
-        "jsdom": "30.1.2",
+        "jsdom": "30.1.3",
        "mermaid": "11.17.2"
      }
    },
@@ -40,9 +40,9 @@
      }
    },
    "node_modules/@asamuzakjp/dom-selector": {
-      "version": "9.2.4",
-      "resolved": "https://registry.npmjs.org/@asamuzakjp/dom-selector/-/dom-selector-9.2.4.tgz",
-      "integrity": "sha512-YnVzxDxLaqL0t7Q4wfcgHZjg55Wm6y5EaO7Bhu3q9J7wq/8iw3PhsnWiOQhYekx4VexnbkgWpTT6n+QIV7kczw==",
+      "version": "10.0.3",
+      "resolved": "https://registry.npmjs.org/@asamuzakjp/dom-selector/-/dom-selector-10.0.3.tgz",
+      "integrity": "sha512-3+777rEPsOOjoy5/sRmIzHnoRDaLpm3XWY2dmWn/N7f/XwDG0bcpnFaKL3X9FYJUbNwLmpcJC1bUYexJRxsFlw==",
      "license": "MIT",
      "dependencies": {
        "bidi-js": "^1.1.0",
@@ -1105,9 +1105,9 @@
      }
    },
    "node_modules/dompurify": {
-      "version": "3.4.16",
-      "resolved": "https://registry.npmjs.org/dompurify/-/dompurify-3.4.16.tgz",
-      "integrity": "sha512-sqo+pNp3qRhCIpbgRi1y8Tgk27Bo2Ry7w0dC1NBeNTdZChWjz9Xb/KOoZbRP/R6pQZ80Qw8YhXw13hWWBbMRnQ==",
+      "version": "3.4.17",
+      "resolved": "https://registry.npmjs.org/dompurify/-/dompurify-3.4.17.tgz",
+      "integrity": "sha512-B/pSlYVHQ58kqISD/yDjuJ1yfzI34m1gF4lAod2bwSRhiN4K5Cl3H8Af7hqB2PmTL911tdEudnnXn5FnhRoz0w==",
      "license": "(MPL-2.0 OR Apache-2.0)",
      "optionalDependencies": {
        "@types/trusted-types": "^2.0.7"
@@ -1202,13 +1202,13 @@
      "license": "MIT"
    },
    "node_modules/jsdom": {
-      "version": "30.1.2",
-      "resolved": "https://registry.npmjs.org/jsdom/-/jsdom-30.1.2.tgz",
-      "integrity": "sha512-0FFE/jE1rppmVfUrJUgxqXjcwolZYIGtAgj1pTussMhNZxIxi7W/PvfWaMNroGi2B36X1IKHbexpZ9DhkoOPiQ==",
+      "version": "30.1.3",
+      "resolved": "https://registry.npmjs.org/jsdom/-/jsdom-30.1.3.tgz",
+      "integrity": "sha512-hQMUPvRgN7KxsfIytDQji7bimuJeRQa6Qo5+iuiZQk5CafeRxLIA10/cG5YCBFNpD9RO3oqIiN4CO1Y44kxqXg==",
      "license": "MIT",
      "dependencies": {
        "@asamuzakjp/css-color": "^7.1.3",
-        "@asamuzakjp/dom-selector": "^9.2.4",
+        "@asamuzakjp/dom-selector": "^10.0.3",
        "@bramus/specificity": "^2.4.2",
        "@csstools/css-syntax-patches-for-csstree": "^1.1.15",
        "@exodus/bytes": "^1.16.0",
@@ -1221,7 +1221,7 @@
        "parse5": "^8.0.1",
        "saxes": "^6.0.0",
        "tough-cookie": "^6.0.2",
-        "undici": "^8.11.2",
+        "undici": "^8.12.1",
        "w3c-xmlserializer": "^6.0.0",
        "webidl-conversions": "^8.0.1",
        "whatwg-mimetype": "^5.0.0",
@@ -1241,28 +1241,29 @@
      }
    },
    "node_modules/katex": {
-      "version": "0.16.47",
-      "resolved": "https://registry.npmjs.org/katex/-/katex-0.16.47.tgz",
-      "integrity": "sha512-Eeo8Ys1doU1z+x8AZsPpQu+p/QcZBI5PeOo7QGQdy2x2m0MU/hYagBbGOmXwr5KVbEfVuWv9LpnQWeehogurjg==",
+      "version": "0.18.11",
+      "resolved": "https://registry.npmjs.org/katex/-/katex-0.18.11.tgz",
+      "integrity": "sha512-yvyz/2VMKqoho/35Hat5w37wIEdWJ5ElVMidTAlWrxr5uW4aBGsEpveTmKh18/zw08v/7lGswCPon4OhqmqrTA==",
+      "deprecated": "Accidentally published with breaking changes. Use 0.19.0 instead.",
      "funding": [
        "https://opencollective.com/katex",
        "https://github.com/sponsors/katex"
      ],
      "license": "MIT",
      "dependencies": {
-        "commander": "^8.3.0"
+        "commander": "^15.0.0"
      },
      "bin": {
        "katex": "cli.js"
      }
    },
    "node_modules/katex/node_modules/commander": {
-      "version": "8.3.0",
-      "resolved": "https://registry.npmjs.org/commander/-/commander-8.3.0.tgz",
-      "integrity": "sha512-OkTL9umf+He2DZkUq8f8J9of7yL6RJKI24dVITBmNfZBmri9zYZQrKkuXiKhyfPSu8tUhnVBB1iKXevvnlR4Ww==",
+      "version": "15.0.0",
+      "resolved": "https://registry.npmjs.org/commander/-/commander-15.0.0.tgz",
+      "integrity": "sha512-z67u4ZhzCL/Tydu1lJARtEZYWbWaN7oYLHbsuzocr6y4N6WZAagG3RQ4FW61V1/0+jImpj293XfrcYnd1qxtPg==",
      "license": "MIT",
      "engines": {
-        "node": ">= 12"
+        "node": ">=22.12.0"
      }
    },
    "node_modules/khroma": {
@@ -1521,9 +1522,9 @@
      }
    },
    "node_modules/undici": {
-      "version": "8.11.2",
-      "resolved": "https://registry.npmjs.org/undici/-/undici-8.11.2.tgz",
-      "integrity": "sha512-u4UB2/IrKdU6lFxumHmmo1a3fCQO5tzQllRorfoRS63txhrB7xTpSn1PftwC4qEHkOaqP95fCWW4lJzwErwzhQ==",
+      "version": "8.12.2",
+      "resolved": "https://registry.npmjs.org/undici/-/undici-8.12.2.tgz",
+      "integrity": "sha512-XtuDWVdqTLr3NAxm2ly9/hU6kUsJOgGh1RASzMBbsw6aWCOaQzHdW+QQDfdERjfJ2V5vL5pRPa4RetO9UfivDg==",
      "license": "MIT",
      "engines": {
        "node": ">=22.19.0"
@@ -4,7 +4,10 @@
  "description": "Validate the Mermaid diagrams of the documentation (see check_mermaid.mjs)",
  "type": "module",
  "dependencies": {
-    "jsdom": "30.1.2",
+    "jsdom": "30.1.3",
    "mermaid": "11.17.2"
+  },
+  "overrides": {
+    "katex": "0.18.11"
  }
 }
@@ -8,17 +8,18 @@

 #pragma once

-#include <algorithm> // count, find, min
+#include <algorithm> // count, find, max, min
 #include <array> // array
 #include <cstddef> // size_t
 #include <cstdint> // uint32_t
 #include <cstring> // strlen
 #include <iterator> // begin, end, iterator_traits, random_access_iterator_tag, distance, next, reverse_iterator
 #include <limits> // numeric_limits
+#include <memory> // unique_ptr
 #include <streambuf> // streambuf
 #include <string> // string, char_traits
 #include <type_traits> // enable_if, is_base_of, is_pointer, is_integral, remove_pointer
-#include <utility> // pair, declval
+#include <utility> // declval, move, pair
 #include <vector> // vector

 #ifndef JSON_NO_IO
@@ -72,6 +73,82 @@ enum class input_format_t { json, cbor, msgpack, ubjson, bson, bjdata, bon8 };

 #ifndef JSON_NO_IO
 /*!
+A growable array of characters for the part of a token that an input adapter
+copied out of its window: unlike std::vector<char>::push_back(), appending a
+character takes no reference to a local, and is a few instructions when there
+is room.
+*/
+class token_carry_buffer
+{
+  public:
+    token_carry_buffer() = default;
+    token_carry_buffer(const token_carry_buffer&) = delete;
+    token_carry_buffer& operator=(const token_carry_buffer&) = delete;
+    token_carry_buffer(token_carry_buffer&& rhs) noexcept
+        : m_data(std::move(rhs.m_data)), m_end(rhs.m_end), m_capacity_end(rhs.m_capacity_end)
+    {
+        rhs.m_end = rhs.m_capacity_end = nullptr;
+    }
+    token_carry_buffer& operator=(token_carry_buffer&&) = delete;
+    ~token_carry_buffer() = default;
+
+    void clear() noexcept
+    {
+        m_end = m_data.get();
+    }
+
+    void push_back(char c)
+    {
+        if (JSON_HEDLEY_UNLIKELY(m_end == m_capacity_end))
+        {
+            grow(1);
+        }
+        *m_end++ = c;
+    }
+
+    void append(const char* first, const char* last)
+    {
+        const auto n = static_cast<std::size_t>(last - first);
+        const auto room = static_cast<std::size_t>(m_capacity_end - m_end);
+        if (n > room)
+        {
+            grow(n);
+        }
+        if (n != 0)
+        {
+            std::memcpy(m_end, first, n);
+        }
+        m_end += n;
+    }
+
+    /// replace the contents of @a out
+    void copy_to(std::vector<char>& out) const
+    {
+        out.assign(m_data.get(), m_end);
+    }
+
+  private:
+    /// make room for @a extra more characters
+    JSON_HEDLEY_NEVER_INLINE void grow(std::size_t extra)
+    {
+        const auto size = static_cast<std::size_t>(m_end - m_data.get());
+        const auto capacity = static_cast<std::size_t>(m_capacity_end - m_data.get());
+        const std::size_t new_capacity = (std::max)(capacity < 32 ? static_cast<std::size_t>(32) : 2 * capacity, size + extra);
+        std::unique_ptr<char[]> data(new char[new_capacity]); // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays)
+        if (size != 0)
+        {
+            std::memcpy(data.get(), m_data.get(), size);
+        }
+        m_data = std::move(data);
+        m_end = m_data.get() + size;
+        m_capacity_end = m_data.get() + new_capacity;
+    }
+
+    std::unique_ptr<char[]> m_data{}; // NOLINT(cppcoreguidelines-avoid-c-arrays,hicpp-avoid-c-arrays,modernize-avoid-c-arrays,readability-redundant-member-init)
+    char* m_end = nullptr;
+    char* m_capacity_end = nullptr;
+};
+/*!
 Input adapter for stdio file access. This adapter read only 1 byte and do not use any
 buffer. This adapter is a very low level adapter.
 */
@@ -209,6 +286,7 @@ class input_stream_adapter
        , token_segment(rhs.token_segment)
        , current_in_get_area(rhs.current_in_get_area)
        , tracking_token(rhs.tracking_token)
+        , no_get_area(rhs.no_get_area)
 #endif
    {
        rhs.is = nullptr;
@@ -295,37 +373,20 @@ class input_stream_adapter
    std::char_traits<char>::int_type get_character()
    {
 #if !JSON_DISABLE_STREAM_BULK_SCAN
-        // sbumpc() refills the get area if it is exhausted
-        const bool refill = streambuf_access::get_ptr(sb) == streambuf_access::get_end(sb);
-        if (JSON_HEDLEY_UNLIKELY(refill))
-        {
-            save_token_segment();
-        }
-#endif
-
-        auto res = sb->sbumpc();
-        // set eof manually, as we don't use the istream interface.
-        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof()))
-        {
-#if !JSON_DISABLE_STREAM_BULK_SCAN
-            current_in_get_area = false;
-#endif
-            is->clear(is->rdstate() | std::ios::eofbit);
-        }
-#if !JSON_DISABLE_STREAM_BULK_SCAN
-        else if (JSON_HEDLEY_UNLIKELY(refill))
-        {
-            // the consumed character is at gptr() - 1, unless the streambuf
-            // has no get area and returned it from uflow()
-            const char* const next = streambuf_access::get_ptr(sb);
-            after_refill(res, next != nullptr && next != streambuf_access::get_begin(sb));
-        }
-        else
+        // the character is in the get area: step over it, as sbumpc() does
+        const char* const next = streambuf_access::get_ptr(sb);
+        if (JSON_HEDLEY_LIKELY(next != streambuf_access::get_end(sb)))
        {
            current_in_get_area = true;
+            streambuf_access::bump(sb, 1);
+            return std::char_traits<char>::to_int_type(*next);
+        }
+        if (no_get_area)
+        {
+            return get_character_no_get_area();
        }
 #endif
-        return res;
+        return get_character_slow();
    }
 #endif

@@ -334,6 +395,18 @@ class input_stream_adapter
    {
 #if JSON_PRECISE_STREAM_POSITION
        commit_lookahead();
+#endif
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // the elements are in the get area: copy them, as sgetn() does, but
+        // without the virtual call
+        const char* const first = streambuf_access::get_ptr(sb);
+        const auto available = static_cast<std::size_t>(streambuf_access::get_end(sb) - first);
+        if (JSON_HEDLEY_LIKELY(count * sizeof(T) <= available && count * sizeof(T) <= static_cast<std::size_t>((std::numeric_limits<int>::max)())))
+        {
+            std::memcpy(dest, first, count * sizeof(T));
+            streambuf_access::bump(sb, static_cast<int>(count * sizeof(T)));
+            return count * sizeof(T);
+        }
 #endif
        auto res = static_cast<std::size_t>(sb->sgetn(reinterpret_cast<char*>(dest), static_cast<std::streamsize>(count * sizeof(T))));
        if (JSON_HEDLEY_UNLIKELY(res < count * sizeof(T)))
@@ -421,7 +494,7 @@ class input_stream_adapter
    // last one if @a drop_last.
    void copy_token(std::vector<char>& out, bool drop_last) const
    {
-        out = token_carry;
+        token_carry.copy_to(out);
        if (token_segment != nullptr)
        {
            out.insert(out.end(), token_segment, current_end());
@@ -434,6 +507,47 @@ class input_stream_adapter
 #endif

  private:
+#if !JSON_PRECISE_STREAM_POSITION
+    // the get area is exhausted, or there is none
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+    JSON_HEDLEY_NEVER_INLINE
+#endif
+    std::char_traits<char>::int_type get_character_slow()
+    {
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        // sbumpc() refills the get area if it is exhausted
+        const bool refill = streambuf_access::get_ptr(sb) == streambuf_access::get_end(sb);
+        if (JSON_HEDLEY_UNLIKELY(refill))
+        {
+            save_token_segment();
+        }
+#endif
+
+        auto res = sb->sbumpc();
+        // set eof manually, as we don't use the istream interface.
+        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof()))
+        {
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+            current_in_get_area = false;
+#endif
+            is->clear(is->rdstate() | std::ios::eofbit);
+        }
+#if !JSON_DISABLE_STREAM_BULK_SCAN
+        else if (JSON_HEDLEY_UNLIKELY(refill))
+        {
+            // the consumed character is at gptr() - 1, unless the streambuf
+            // has no get area and returned it from uflow()
+            const char* const next = streambuf_access::get_ptr(sb);
+            after_refill(res, next != nullptr && next != streambuf_access::get_begin(sb));
+        }
+        else
+        {
+            current_in_get_area = true;
+        }
+#endif
+        return res;
+    }
+#endif
 #if !JSON_DISABLE_STREAM_BULK_SCAN
    // Reads the protected get-area pointers of any std::streambuf. Naming the
    // members through a derived class makes forming the member pointers
@@ -457,6 +571,11 @@ class input_stream_adapter
        {
            (s->*(&streambuf_access::gbump))(n);
        }
+        // sbumpc() for an exhausted get area
+        static std::streambuf::int_type next_character(std::streambuf* s)
+        {
+            return (s->*(&streambuf_access::uflow))();
+        }
    };

    // one past the character last returned by get_character(), if that is in
@@ -477,16 +596,53 @@ class input_stream_adapter
    {
        if (token_segment != nullptr)
        {
-            token_carry.insert(token_carry.end(), token_segment, streambuf_access::get_end(sb));
+            token_carry.append(token_segment, streambuf_access::get_end(sb));
            token_segment = nullptr;
        }
    }

+#if !JSON_PRECISE_STREAM_POSITION
+    // The streambuf has no get area (e.g., std::cin synchronized with stdio):
+    // every character comes from uflow(), and the token is kept by value.
+    JSON_HEDLEY_NEVER_INLINE std::char_traits<char>::int_type get_character_no_get_area()
+    {
+        // the get area is exhausted, so this is what sbumpc() does
+        const auto res = streambuf_access::next_character(sb);
+        if (JSON_HEDLEY_UNLIKELY(res == std::char_traits<char>::eof() || streambuf_access::get_end(sb) != nullptr))
+        {
+            return get_character_no_get_area_other(res);
+        }
+        // current_in_get_area is false: only the first character of a get
+        // area, or the end of the input, is read here
+        if (tracking_token)
+        {
+            token_carry.push_back(std::char_traits<char>::to_char_type(res));
+        }
+        return res;
+    }
+
+    // @a res came from uflow() of a streambuf without a get area: it is the
+    // end of the input, or the streambuf has a get area now
+    JSON_HEDLEY_NEVER_INLINE std::char_traits<char>::int_type get_character_no_get_area_other(std::char_traits<char>::int_type res)
+    {
+        if (res == std::char_traits<char>::eof())
+        {
+            current_in_get_area = false;
+            is->clear(is->rdstate() | std::ios::eofbit);
+            return res;
+        }
+        // the consumed character is at gptr() - 1
+        after_refill(res, streambuf_access::get_ptr(sb) != streambuf_access::get_begin(sb));
+        return res;
+    }
+#endif
+
    // character @a c was just read into a new get area (@a in_get_area) or
    // returned without one
    void after_refill(std::char_traits<char>::int_type c, bool in_get_area)
    {
        current_in_get_area = in_get_area;
+        no_get_area = !in_get_area;
        if (!tracking_token)
        {
            return;
@@ -525,13 +681,16 @@ class input_stream_adapter
 #endif
 #if !JSON_DISABLE_STREAM_BULK_SCAN
    /// the part of the current token that was in earlier get areas
-    std::vector<char> token_carry{}; // NOLINT(readability-redundant-member-init)
+    token_carry_buffer token_carry{}; // NOLINT(readability-redundant-member-init)
    /// start of the current token's part in the get area, or nullptr
    const char* token_segment = nullptr;
    /// whether the character last returned by get_character() is in the get area
    bool current_in_get_area = false;
    /// whether a token is tracked at all; the binary readers never start one
    bool tracking_token = false;
+    /// whether the streambuf had no get area when it was last asked for a
+    /// character
+    bool no_get_area = false;
 #endif
 };
 #endif  // JSON_NO_IO
@@ -8,14 +8,14 @@

 #pragma once

-#include <algorithm> // all_of
+#include <algorithm> // all_of, copy
 #include <array> // array
 #include <cctype> // isdigit
 #include <cerrno> // errno, ERANGE
 #include <cstddef> // size_t
 #include <cstdint> // uint64_t
 #include <cstdlib> // strtoull
-#include <cstring> // memcmp, memcpy
+#include <cstring> // memchr, memcmp, memcpy
 #ifndef JSON_NO_IO
    #include <iosfwd> // ostream
 #endif  // JSON_NO_IO
@@ -80,14 +80,38 @@ class json_pointer
    /// @sa https://json.nlohmann.me/api/json_pointer/to_string/
    string_t to_string() const
    {
+        if (reference_tokens.empty())
+        {
+            return string_t();
+        }
+
        // one buffer: every token adds its escaped form (at least its length) and a slash
        std::size_t length = 0;
+        bool plain = true;
        for (const auto& reference_token : reference_tokens)
        {
            length += reference_token.size() + 1;
+            const char* const data = reference_token.data();
+            for (std::size_t i = 0, size = reference_token.size(); plain && i < size; ++i)
+            {
+                plain = data[i] != '~' && data[i] != '/';
+            }
        }

        string_t result;
+        if (JSON_HEDLEY_LIKELY(plain))
+        {
+            // nothing to escape: the tokens are copied into the string
+            result.resize(length);
+            char* out = &result[0];
+            for (const auto& reference_token : reference_tokens)
+            {
+                *out++ = '/';
+                out = std::copy(reference_token.data(), reference_token.data() + reference_token.size(), out);
+            }
+            return result;
+        }
+
        result.reserve(length);
        for (const auto& reference_token : reference_tokens)
        {
@@ -303,6 +327,12 @@ class json_pointer
        exceeds_size_type  ///< @a s converts to an integer that exceeds size_type
    };

+    /// whether a reference token is "-" (a plain comparison with a string literal is not always inlined)
+    static bool is_dash(const string_t& s) noexcept
+    {
+        return s.size() == 1 && s[0] == '-';
+    }
+
    /*!
    @param[in] s    reference token to be converted into an array index
    @param[out] idx the integer representation of @a s if @ref array_index_status::ok
@@ -328,7 +358,7 @@ class json_pointer
        // error condition (cf. RFC 6901, Sect. 4); this also covers single-
        // character tokens, so "/x" and "/xy" fail alike; "-" and the empty
        // token are left to the conversion below and are reported as unresolved
-        if (JSON_HEDLEY_UNLIKELY(!s.empty() && s != "-" && !(s[0] >= '0' && s[0] <= '9')))
+        if (JSON_HEDLEY_UNLIKELY(!s.empty() && !is_dash(s) && !(s[0] >= '0' && s[0] <= '9')))
        {
            return array_index_status::not_a_number;
        }
@@ -475,9 +505,40 @@ class json_pointer
        return pos == size;
    }

+    /// reserve room for @a n elements of an array type that has reserve()
+    template<typename ArrayType>
+    static auto reserve_elements(ArrayType& array, const std::size_t n, detail::priority_tag<1> /*unused*/)
+    -> decltype(array.reserve(n), void())
+    {
+        array.reserve(n);
+    }
+
+    template<typename ArrayType>
+    static void reserve_elements(ArrayType& /*array*/, const std::size_t /*n*/, detail::priority_tag<0> /*unused*/)
+    {}
+
    /// @brief object member access with a key that is only copied if it is inserted
+    ///
+    /// Objects that can find the position of a key and insert there
+    /// (std::map) search once, others look the key up and add it if it is missing.
    template<typename BasicJsonType>
-    static BasicJsonType& object_member(BasicJsonType& object, const string_t& key)
+    static auto object_member_impl(BasicJsonType& object, const string_t& key, detail::priority_tag<1> /*unused*/)
+    -> decltype(object.m_data.m_value.object->emplace_hint(object.m_data.m_value.object->lower_bound(key), key, nullptr), std::declval<BasicJsonType&>())
+    {
+        auto& members = *object.m_data.m_value.object;
+        auto it = members.lower_bound(key);
+        if (it != members.end() && !members.key_comp()(key, it->first))
+        {
+            return it->second;
+        }
+
+        const auto old_capacity = object.container_capacity();
+        it = members.emplace_hint(it, key, nullptr);
+        return object.set_parent_after_insert(it->second, old_capacity);
+    }
+
+    template<typename BasicJsonType>
+    static BasicJsonType& object_member_impl(BasicJsonType& object, const string_t& key, detail::priority_tag<0> /*unused*/)
    {
        auto& members = *object.m_data.m_value.object;
        const auto it = members.find(key);
@@ -488,6 +549,12 @@ class json_pointer
        return object.operator[](key);
    }

+    template<typename BasicJsonType>
+    static BasicJsonType& object_member(BasicJsonType& object, const string_t& key)
+    {
+        return object_member_impl(object, key, detail::priority_tag<1> {});
+    }
+
  JSON_PRIVATE_UNLESS_TESTED:
    json_pointer top() const
    {
@@ -720,48 +787,103 @@ class json_pointer
        }
    }

+    /*!
+    @brief the values along the pointer last resolved by @ref resolve_plain
+
+    A patch from @ref basic_json::diff lists its operations in document order,
+    so consecutive pointers share most of their reference tokens; the values
+    along the previous pointer are kept to resume from the common prefix.
+    `steps[k]` is the value after the first `k` reference tokens together with
+    the length of the prefix of `*path` that holds them. A value stays valid as
+    long as no operation changes its container: callers keep only values up to
+    the parent of the pointer they modify, or call `reset()`.
+    */
+    template<typename BasicJsonType>
+    struct plain_steps
+    {
+        const string_t* path = nullptr;
+        std::vector<std::pair<std::size_t, BasicJsonType*>> steps{};
+
+        void reset() noexcept
+        {
+            path = nullptr;
+            steps.clear();
+        }
+    };
+
    /*!
    @brief resolve a plain JSON pointer without building its reference tokens

-    @param[in] ptr     the value the pointer starts at
+    @param[in] root    the value the pointer starts at
    @param[in] path    a JSON pointer string
    @param[in] length  the number of leading bytes of @a path to resolve; 0 or
                       the position of a slash
    @param[in,out] token  a string to hold the reference tokens, to reuse its storage
+    @param[in,out] cache  the values along the previously resolved pointer;
+                          @a path must outlive the next call

    @return the pointed to value, or `nullptr` if the pointer does not denote
-            an existing value, or does so in a way that is not plain: a
-            reference token with an escape, an array index that is malformed or
-            out of range, or a "-". No exception is thrown for these; callers
-            use @ref json_pointer instead, which handles them.
+            an existing value, or does so in a way that is not plain: an array
+            index that is malformed or out of range, or a "-", or a reference
+            token with an escape. No exception is
+            thrown for these; callers use @ref json_pointer instead, which
+            handles them.
    */
    template<typename BasicJsonType>
-    static BasicJsonType* resolve_plain(BasicJsonType* ptr, const string_t& path, const std::size_t length, string_t& token)
+    static BasicJsonType* resolve_plain(BasicJsonType* root, const string_t& path, const std::size_t length, string_t& token, plain_steps<BasicJsonType>& cache)
    {
-        for (std::size_t pos = 0; pos < length;)
+        auto& steps = cache.steps;
+
+        // resume from the deepest step that is also a prefix of this path
+        std::size_t depth = 1;
+        if (cache.path != nullptr && !steps.empty() && steps[0].second == root)
+        {
+            const string_t& previous = *cache.path;
+            for (depth = steps.size(); depth > 1; --depth)
+            {
+                // a step ends at a token boundary of the previous path
+                const std::size_t prefix = steps[depth - 1].first;
+                if (prefix <= length && (prefix == length || path[prefix] == '/') && std::memcmp(path.data(), previous.data(), prefix) == 0)
+                {
+                    break;
+                }
+            }
+            steps.resize(depth);
+        }
+        else
+        {
+            steps.assign(1, std::make_pair(std::size_t{0}, root));
+        }
+        cache.path = &path;
+
+        BasicJsonType* ptr = steps[depth - 1].second;
+        for (std::size_t pos = steps[depth - 1].first; pos < length;)
        {
            if (JSON_HEDLEY_UNLIKELY(path[pos] != '/'))
            {
                return nullptr;
            }

-            // the reference token between the slash at pos and the next one
-            std::size_t end = path.find_first_of('/', pos + 1);
-            if (end == string_t::npos || end > length)
+            // the reference token between the slash at pos and the next one,
+            // which must not have an escape
+            const char* const begin = path.data() + pos + 1;
+            const char* const limit = path.data() + length;
+            const char* end = begin;
+            while (end < limit && *end != '/')
            {
-                end = length;
-            }
-            token.clear();
-            token.append(path.data() + pos + 1, end - pos - 1);
-            if (JSON_HEDLEY_UNLIKELY(token.find_first_of('~') != string_t::npos))
-            {
-                return nullptr;
+                if (JSON_HEDLEY_UNLIKELY(*end == '~'))
+                {
+                    return nullptr;
+                }
+                ++end;
            }

            switch (ptr->type())
            {
                case detail::value_t::object:
                {
+                    token.clear();
+                    token.append(begin, static_cast<std::size_t>(end - begin));
                    const auto it = ptr->m_data.m_value.object->find(token);
                    if (JSON_HEDLEY_UNLIKELY(it == ptr->m_data.m_value.object->end()))
                    {
@@ -773,8 +895,24 @@ class json_pointer

                case detail::value_t::array:
                {
-                    typename BasicJsonType::size_type idx{};
-                    if (JSON_HEDLEY_UNLIKELY(parse_array_index<BasicJsonType>(token, idx) != array_index_status::ok || idx >= ptr->m_data.m_value.array->size()))
+                    // an index of up to 18 digits without a leading zero fits into size_type
+                    // (9 digits for a 32-bit size_type); everything else is left to json_pointer
+                    constexpr std::size_t max_digits = sizeof(typename BasicJsonType::size_type) >= 8 ? 18 : 9;
+                    const auto digits = static_cast<std::size_t>(end - begin);
+                    if (JSON_HEDLEY_UNLIKELY(digits == 0 || digits > max_digits || (digits > 1 && *begin == '0')))
+                    {
+                        return nullptr;
+                    }
+                    typename BasicJsonType::size_type idx = 0;
+                    for (const char* c = begin; c != end; ++c)
+                    {
+                        if (JSON_HEDLEY_UNLIKELY(*c < '0' || *c > '9'))
+                        {
+                            return nullptr;
+                        }
+                        idx = static_cast<typename BasicJsonType::size_type>((idx * 10) + static_cast<typename BasicJsonType::size_type>(*c - '0'));
+                    }
+                    if (JSON_HEDLEY_UNLIKELY(idx >= ptr->m_data.m_value.array->size()))
                    {
                        return nullptr;
                    }
@@ -793,7 +931,8 @@ class json_pointer
                default:
                    return nullptr;
            }
-            pos = end;
+            pos = static_cast<std::size_t>(end - path.data());
+            steps.emplace_back(pos, ptr);
        }

        return ptr;
@@ -907,7 +1046,7 @@ class json_pointer

                case detail::value_t::array:
                {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                    {
                        // "-" always fails the range check
                        JSON_THROW(detail::out_of_range::create(detail::exception_id::pointer_past_the_end_index, detail::concat(
@@ -978,7 +1117,7 @@ class json_pointer

                case detail::value_t::array:
                {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                    {
                        // "-" cannot be used for const access
                        JSON_THROW(detail::out_of_range::create(detail::exception_id::pointer_past_the_end_index, detail::concat("array index '-' (", std::to_string(ptr->m_data.m_value.array->size()), ") is out of range"), ptr));
@@ -1040,7 +1179,7 @@ class json_pointer

                case detail::value_t::array:
                {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                    {
                        // "-" always fails the range check
                        return nullptr;
@@ -1117,7 +1256,7 @@ class json_pointer

                case detail::value_t::array:
                {
-                    if (JSON_HEDLEY_UNLIKELY(reference_token == "-"))
+                    if (JSON_HEDLEY_UNLIKELY(is_dash(reference_token)))
                    {
                        // "-" always fails the range check
                        return false;
@@ -1198,33 +1337,32 @@ class json_pointer
        }
        result.reserve(token_count);

-        // extract the reference tokens:
-        // - slash: position of the last read slash (or end of string)
-        // - start: position after the previous slash
-        for (
-            // search for the first slash after the first character
-            std::size_t slash = reference_string.find_first_of('/', 1),
-            // set the beginning of the first reference token
-            start = 1;
-            // we can stop if start == 0 (if slash == string_t::npos)
-            start != 0;
-            // set the beginning of the next reference token
-            // (will eventually be 0 if slash == string_t::npos)
-            start = (slash == string_t::npos) ? 0 : slash + 1,
-            // find next slash
-            slash = reference_string.find_first_of('/', start))
+        // extract the reference tokens: every one runs from after a slash to
+        // the next slash or the end of the string; a token without a '~' (the
+        // common case) is stored as it is, others are checked and unescaped
+        const char* const data = reference_string.data();
+        const char* const end = data + size;
+        for (const char* start = data + 1;; ++start)
        {
-            // use the text between the beginning of the reference token
-            // (start) and the last slash (slash).
-            const auto count = (slash == string_t::npos ? reference_string.size() : slash) - start;
-            auto reference_token = string_t(reference_string.data() + start, count);
-
-            // check reference tokens are properly escaped; a token without
-            // a '~' (the common case) needs neither check nor unescaping
-            std::size_t pos = reference_token.find_first_of('~');
-            if (pos != string_t::npos)
+            const char* slash = start;
+            bool escaped = false;
+            while (slash != end && *slash != '/')
            {
-                for (; pos != string_t::npos; pos = reference_token.find_first_of('~', pos + 1))
+                escaped = escaped || *slash == '~';
+                ++slash;
+            }
+
+            const auto count = static_cast<std::size_t>(slash - start);
+            if (JSON_HEDLEY_LIKELY(!escaped))
+            {
+                result.emplace_back(start, count);
+            }
+            else
+            {
+                auto reference_token = string_t(start, count);
+
+                // check reference tokens are properly escaped
+                for (std::size_t pos = reference_token.find_first_of('~'); pos != string_t::npos; pos = reference_token.find_first_of('~', pos + 1))
                {
                    JSON_ASSERT(reference_token[pos] == '~');

@@ -1238,10 +1376,14 @@ class json_pointer
                }

                detail::unescape(reference_token);
+                result.push_back(std::move(reference_token));
            }

-            // finally, store the reference token
-            result.push_back(std::move(reference_token));
+            if (slash == end)
+            {
+                break;
+            }
+            start = slash;
        }

        return result;
@@ -1276,6 +1418,52 @@ class json_pointer
        obj[key] = std::forward<ValueType>(v);
    }

+    /// the number of entries that flattening @a value adds: its primitive
+    /// values and empty containers
+    template<typename BasicJsonType>
+    static std::size_t count_flattened(const BasicJsonType& value)
+    {
+        std::size_t count = 0;
+        std::vector<const BasicJsonType*> pending(1, &value);
+        while (!pending.empty())
+        {
+            const BasicJsonType* const v = pending.back();
+            pending.pop_back();
+            if (v->is_array() && !v->m_data.m_value.array->empty())
+            {
+                for (const auto& element : *v->m_data.m_value.array)
+                {
+                    pending.push_back(&element);
+                }
+            }
+            else if (v->is_object() && !v->m_data.m_value.object->empty())
+            {
+                for (const auto& member : *v->m_data.m_value.object)
+                {
+                    pending.push_back(&member.second);
+                }
+            }
+            else
+            {
+                ++count;
+            }
+        }
+        return count;
+    }
+
+    // Reserve the storage for @a value of objects that have a reserve(), so
+    // that it does not have to be moved while it grows
+    template<typename ObjectType, typename BasicJsonType>
+    static auto reserve_flattened(ObjectType& obj, const BasicJsonType& value, detail::priority_tag<1> /*unused*/)
+    -> decltype(obj.reserve(std::size_t {}), void())
+    {
+        obj.reserve(count_flattened(value));
+    }
+
+    template<typename ObjectType, typename BasicJsonType>
+    static void reserve_flattened(ObjectType& /*obj*/, const BasicJsonType& /*value*/, detail::priority_tag<0> /*unused*/)
+    {}
+
    /// append the decimal representation of @a i to @a out
    static void append_index(string_t& out, std::size_t i)
    {
@@ -1385,6 +1573,7 @@ class json_pointer
            }
        };

+        reserve_flattened(object, value, detail::priority_tag<1> {});
        enter(value);
        while (!stack.empty())
        {
@@ -1392,8 +1581,9 @@ class json_pointer
            // frame, which would invalidate a reference to it
            const BasicJsonType* const container = stack.back().container;

-            // drop the path of the previous child
-            path.resize(stack.back().path_length);
+            // the path of the previous child is replaced by the path of the
+            // next one; for an array element in one resize
+            const std::size_t path_length = stack.back().path_length;

            if (container->is_array())
            {
@@ -1407,8 +1597,20 @@ class json_pointer

                // iterate array and use index as a reference string
                ++stack.back().index;
-                path.push_back('/');
-                append_index(path, i);
+                std::array<char, 24> digits{};
+                std::size_t pos = digits.size();
+                std::size_t rest = i;
+                do
+                {
+                    digits[--pos] = static_cast<char>('0' + (rest % 10));
+                    rest /= 10;
+                }
+                while (rest != 0);
+                const std::size_t count = digits.size() - pos;
+                path.resize(path_length + 1 + count);
+                char* const out = &path[path_length];
+                out[0] = '/';
+                std::memcpy(out + 1, digits.data() + pos, count);
                enter(array[i]);
            }
            else
@@ -1422,6 +1624,7 @@ class json_pointer

                // iterate object and use keys as reference string
                ++stack.back().member;
+                path.resize(path_length);
                path.push_back('/');
                detail::escape_into(path, it->first);
                enter(it->second);
@@ -1517,8 +1720,28 @@ class json_pointer
                {
                    end = key.size();
                }
+                const std::size_t token_size = end - pos - 1;
+
+                // a token that is a plain index of up to 9 digits, which is
+                // the case for almost all array elements, is not copied
+                std::size_t index = 0;
+                bool plain_index = current->type() == detail::value_t::array && token_size != 0 && token_size <= 9 && (token_size == 1 || data[pos + 1] != '0');
+                for (std::size_t k = pos + 1; plain_index && k < end; ++k)
+                {
+                    plain_index = data[k] >= '0' && data[k] <= '9';
+                    index = (index * 10) + static_cast<std::size_t>(data[k] - '0');
+                }
+                if (plain_index)
+                {
+                    current = &current->operator[](static_cast<typename BasicJsonType::size_type>(index));
+                    values.push_back(current);
+                    ends.push_back(end);
+                    pos = end;
+                    continue;
+                }
+
                token.clear();
-                token.append(data + pos + 1, end - pos - 1);
+                token.append(data + pos + 1, token_size);
                if (token.find_first_of('~') != string_t::npos)
                {
                    detail::unescape(token);
@@ -1528,10 +1751,13 @@ class json_pointer
                {
                    case detail::value_t::null:
                    {
-                        if (arrays.contains(data, pos))
+                        // a first child 0 makes the value an array without a look-up
+                        if ((token_size == 1 && data[pos + 1] == '0') || arrays.contains(data, pos))
                        {
                            // some reference token below this position is 0, so the
                            // value is an array
+                            *current = BasicJsonType(detail::value_t::array);
+                            reserve_elements(*current->m_data.m_value.array, 2, detail::priority_tag<1> {});
                            current = &current->operator[](array_index<BasicJsonType>(token));
                        }
                        else
@@ -55,6 +55,66 @@ using output_adapter_t = std::shared_ptr<output_adapter_protocol<CharType>>;
 /// output_vector_sink also appends writes of at least this size with insert()
 constexpr std::size_t max_reserve_size = 256;

+/// @brief copy @a length characters, at most 32, without a call of memcpy()
+///
+/// Short strings are the typical payload of a binary value, and a call of
+/// memcpy() costs more than copying them. A copy of 16 to 32 characters is two
+/// possibly overlapping 16-byte copies, one of 8 to 15 two 8-byte copies, and
+/// so on down to a single character; each is a fixed-size load and store.
+/// Nothing outside of [src, src + length) is read or written.
+template<typename CharType>
+inline void copy_short(CharType* dst, const CharType* src, const std::size_t length, std::true_type /*is_byte*/) noexcept
+{
+    JSON_ASSERT(length <= 32);
+    if (length >= 16)
+    {
+        std::array<char, 16> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 16> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 16);
+        std::memcpy(last.data(), src + (length - 16), 16);
+        std::memcpy(dst, first.data(), 16);
+        std::memcpy(dst + (length - 16), last.data(), 16);
+    }
+    else if (length >= 8)
+    {
+        std::array<char, 8> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 8> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 8);
+        std::memcpy(last.data(), src + (length - 8), 8);
+        std::memcpy(dst, first.data(), 8);
+        std::memcpy(dst + (length - 8), last.data(), 8);
+    }
+    else if (length >= 4)
+    {
+        std::array<char, 4> first; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::array<char, 4> last; // NOLINT(cppcoreguidelines-pro-type-member-init,hicpp-member-init): written before read
+        std::memcpy(first.data(), src, 4);
+        std::memcpy(last.data(), src + (length - 4), 4);
+        std::memcpy(dst, first.data(), 4);
+        std::memcpy(dst + (length - 4), last.data(), 4);
+    }
+    else if (length != 0)
+    {
+        // 1 to 3: the first, the middle, and the last character
+        const CharType a = src[0];
+        const CharType b = src[length >> 1];
+        const CharType c = src[length - 1];
+        dst[0] = a;
+        dst[length >> 1] = b;
+        dst[length - 1] = c;
+    }
+}
+
+/// @brief copy @a length characters, of a character type wider than a byte
+template<typename CharType>
+inline void copy_short(CharType* dst, const CharType* src, const std::size_t length, std::false_type /*is_byte*/) noexcept
+{
+    std::memcpy(dst, src, length * sizeof(CharType));
+}
+
+/// the number of characters an empty vector is grown to by output_vector_sink
+constexpr std::size_t min_vector_sink_capacity = 64;
+
 /// @brief non-virtual output sink writing into a std::vector
 ///
 /// This sink is not part of the virtual output_adapter_protocol hierarchy: it is
@@ -78,7 +138,7 @@ class output_vector_sink
 {
  public:
    explicit output_vector_sink(std::vector<CharType, AllocatorType>& vec) noexcept
-        : v(&vec), position(vec.size())
+        : v(&vec), cursor(vec.data() + vec.size()), limit(cursor)
    {}

    output_vector_sink(const output_vector_sink&) = delete;
@@ -86,7 +146,7 @@ class output_vector_sink
    output_vector_sink& operator=(output_vector_sink&&) = delete;

    output_vector_sink(output_vector_sink&& other) noexcept
-        : v(other.v), position(other.position)
+        : v(other.v), cursor(other.cursor), limit(other.limit)
    {
        other.v = nullptr;
    }
@@ -95,7 +155,7 @@ class output_vector_sink
    {
        if (v != nullptr)
        {
-            v->resize(position);
+            v->resize(position());
        }
    }

@@ -104,11 +164,11 @@ class output_vector_sink
    JSON_HEDLEY_ALWAYS_INLINE
    void write_character(CharType c)
    {
-        if (JSON_HEDLEY_UNLIKELY(position == v->size()))
+        if (JSON_HEDLEY_UNLIKELY(cursor == limit))
        {
            grow(1);
        }
-        (*v)[position++] = c;
+        *cursor++ = c;
    }

    // no JSON_HEDLEY_NON_NULL here: binary_writer legitimately passes a null
@@ -117,7 +177,7 @@ class output_vector_sink
    JSON_HEDLEY_ALWAYS_INLINE
    void write_characters(const CharType* s, std::size_t length)
    {
-        if (JSON_HEDLEY_UNLIKELY(v->size() - position < length))
+        if (JSON_HEDLEY_UNLIKELY(space() < length))
        {
            if (length >= max_reserve_size)
            {
@@ -126,10 +186,15 @@ class output_vector_sink
            }
            grow(length);
        }
-        if (length != 0)
+        if (length <= 32)
        {
-            std::memcpy(v->data() + position, s, length * sizeof(CharType));
-            position += length;
+            copy_short(cursor, s, length, std::integral_constant<bool, sizeof(CharType) == 1>());
+            cursor += length;
+        }
+        else
+        {
+            std::memcpy(cursor, s, length * sizeof(CharType));
+            cursor += length;
        }
    }

@@ -144,21 +209,37 @@ class output_vector_sink
    CharType* reserve(const std::size_t length)
    {
        JSON_ASSERT(length <= max_reserve_size);
-        if (JSON_HEDLEY_UNLIKELY(v->size() - position < length))
+        if (JSON_HEDLEY_UNLIKELY(space() < length))
        {
            grow(length);
        }
-        return v->data() + position;
+        return cursor;
    }

    /// keep the first @a length characters written to the space from reserve()
    JSON_HEDLEY_ALWAYS_INLINE
    void commit(const std::size_t length) noexcept
    {
-        position += length;
+        cursor += length;
    }

+    /// nothing to pass on: the characters are in the vector (see output_adapter_sink)
+    void flush() noexcept
+    {}
+
  private:
+    /// number of characters of *v written so far
+    std::size_t position() const noexcept
+    {
+        return static_cast<std::size_t>(cursor - v->data());
+    }
+
+    /// number of characters that fit the vector behind the write position
+    std::size_t space() const noexcept
+    {
+        return static_cast<std::size_t>(limit - cursor);
+    }
+
    /// make room for at least @a length more characters after the position;
    /// kept out of line so that the write functions stay small enough for
    /// GCC to inline them into the writer
@@ -166,12 +247,15 @@ class output_vector_sink
    void grow(const std::size_t length)
    {
        // use the capacity the vector already has, else at least double it;
-        // never less than 64 characters, so that a small value does not
+        // never less than min_vector_sink_capacity, so that a small value does not
        // reallocate for every few bytes
-        const std::size_t needed = position + length;
-        v->resize((std::max)(std::size_t{64}, needed <= v->capacity()
+        const std::size_t written = position();
+        const std::size_t needed = written + length;
+        v->resize((std::max)(min_vector_sink_capacity, needed <= v->capacity()
                             ? v->capacity()
                             : (std::max)(needed, v->capacity() * 2)));
+        cursor = v->data() + written;
+        limit = v->data() + v->size();
    }

    /// append @a length (at least max_reserve_size) characters that do not fit
@@ -181,20 +265,24 @@ class output_vector_sink
    JSON_HEDLEY_NEVER_INLINE
    void append(const CharType* s, const std::size_t length)
    {
-        v->resize(position);
-        if (v->capacity() - position < length)
+        const std::size_t written = position();
+        v->resize(written);
+        if (v->capacity() - written < length)
        {
            // grow geometrically, with room for the writes that follow, so
            // that the next grow() need not reallocate right away
-            v->reserve((std::max)(v->capacity() * 2, position + length + max_reserve_size));
+            v->reserve((std::max)(v->capacity() * 2, written + length + max_reserve_size));
        }
        v->insert(v->end(), s, s + length);
-        position = v->size();
+        cursor = v->data() + v->size();
+        limit = cursor;
    }

    std::vector<CharType, AllocatorType>* v;
-    /// number of characters of *v written so far
-    std::size_t position;
+    /// the write position: in *v, at the number of characters written so far
+    CharType* cursor;
+    /// the end of *v
+    CharType* limit;
 };

 /// output adapter for byte vectors
@@ -277,9 +365,17 @@ class output_string_adapter : public output_adapter_protocol<CharType>
 ///
 /// Wraps the polymorphic output_adapter_t so the same binary_writer template can
 /// also target arbitrary adapters (output streams, strings, user-provided
-/// adapters) via the `output_adapter`-based overloads. Each write still goes
-/// through one virtual call, exactly as before; only the concrete sinks above
-/// avoid it.
+/// adapters) via the `output_adapter`-based overloads.
+///
+/// A write through the adapter is a virtual call, and for a stream a call with
+/// a sentry on top, which costs more than encoding the few bytes of most values.
+/// The sink therefore collects what is written in a buffer and passes it on in
+/// one piece when it is full, a write is too large for it, or @ref flush is
+/// called. Everything written reaches the adapter in the same order and with
+/// the same bytes; the destructor passes on what is left, so that a write that
+/// throws leaves the adapter with exactly the bytes written before it. Only
+/// @ref flush can report an error of the adapter (the destructor must not
+/// throw), so it has to be called once the value is written.
 template<typename CharType>
 class output_adapter_sink
 {
@@ -290,36 +386,121 @@ class output_adapter_sink
        JSON_ASSERT(oa);
    }

+    output_adapter_sink(const output_adapter_sink&) = delete;
+    output_adapter_sink& operator=(const output_adapter_sink&) = delete;
+    output_adapter_sink& operator=(output_adapter_sink&&) = delete;
+
+    output_adapter_sink(output_adapter_sink&& other) noexcept
+        : oa(std::move(other.oa)), used(other.used)
+    {
+        std::memcpy(buffer.data(), other.buffer.data(), other.used * sizeof(CharType));
+        other.used = 0;
+    }
+
+    ~output_adapter_sink()
+    {
+        // a normal return has flushed already, so this is for a write that threw
+        if (used != 0)
+        {
+            JSON_TRY
+            {
+                flush();
+            }
+            JSON_CATCH(...)
+            {
+                // nothing can be done about an adapter that throws here
+            }
+        }
+    }
+
+    JSON_HEDLEY_ALWAYS_INLINE
    void write_character(CharType c)
    {
-        oa->write_character(c);
+        if (JSON_HEDLEY_UNLIKELY(used == buffer_size))
+        {
+            flush();
+        }
+        buffer[used++] = c;
    }

    // no JSON_HEDLEY_NON_NULL: forwards (null, 0) for empty payloads, exactly as
    // the type-erased path already did before this sink existed
+    JSON_HEDLEY_ALWAYS_INLINE
    void write_characters(const CharType* s, std::size_t length)
    {
-        oa->write_characters(s, length);
+        if (JSON_HEDLEY_UNLIKELY(buffer_size - used < length))
+        {
+            write_large(s, length);
+            return;
+        }
+        if (length <= 32)
+        {
+            copy_short(buffer.data() + used, s, length, std::integral_constant<bool, sizeof(CharType) == 1>());
+            used += length;
+        }
+        else
+        {
+            std::memcpy(buffer.data() + used, s, length * sizeof(CharType));
+            used += length;
+        }
    }

    /// space for writing up to @a length characters (see
    /// output_vector_sink::reserve)
-    CharType* reserve(const std::size_t length) noexcept
+    JSON_HEDLEY_ALWAYS_INLINE
+    CharType* reserve(const std::size_t length)
    {
        JSON_ASSERT(length <= max_reserve_size);
-        static_cast<void>(length);
-        return reserved.data();
+        if (JSON_HEDLEY_UNLIKELY(buffer_size - used < length))
+        {
+            flush();
+        }
+        return buffer.data() + used;
    }

-    /// write the first @a length characters written to the space from reserve()
-    void commit(const std::size_t length)
+    /// keep the first @a length characters written to the space from reserve()
+    JSON_HEDLEY_ALWAYS_INLINE
+    void commit(const std::size_t length) noexcept
    {
-        oa->write_characters(reserved.data(), length);
+        used += length;
+    }
+
+    /// pass everything written so far on to the adapter
+    void flush()
+    {
+        if (used != 0)
+        {
+            // reset first: the characters count as written even if the adapter throws
+            const std::size_t length = used;
+            used = 0;
+            oa->write_characters(buffer.data(), length);
+        }
    }

  private:
+    /// a write that does not fit the space left in the buffer
+    JSON_HEDLEY_NEVER_INLINE
+    void write_large(const CharType* s, const std::size_t length)
+    {
+        flush();
+        if (length >= buffer_size)
+        {
+            oa->write_characters(s, length);
+        }
+        else
+        {
+            std::memcpy(buffer.data(), s, length * sizeof(CharType));
+            used = length;
+        }
+    }
+
+    /// large enough for reserve() at any time
+    static constexpr std::size_t buffer_size = 4 * max_reserve_size;
+
    output_adapter_t<CharType> oa;
-    std::array<CharType, max_reserve_size> reserved{};
+    /// number of characters of @ref buffer not yet passed on
+    std::size_t used = 0;
+    std::array<CharType, buffer_size> buffer{};
 };

 /// @brief whether std::basic_string<CharType> has a non-deprecated std::char_traits
@@ -19,6 +19,7 @@
 #define INCLUDE_NLOHMANN_JSON_HPP_

 #include <algorithm> // all_of, find, for_each, none_of
+#include <array> // array
 #include <cmath> // isnan
 #include <cstddef> // nullptr_t, ptrdiff_t, size_t
 #include <cstdint> // uint8_t
@@ -6321,7 +6322,7 @@ public:
            const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
        std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
        vector_writer(result, error_handler).write_cbor(j);
        return result;
    }
@@ -6331,7 +6332,9 @@ public:
    static void to_cbor(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                        const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
-        binary_writer<std::uint8_t>(o, error_handler).write_cbor(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_cbor(j);
+        writer.flush();
    }

    /// @brief create a CBOR serialization of a given JSON value
@@ -6339,7 +6342,9 @@ public:
    static void to_cbor(const basic_json& j, detail::output_adapter<char> o,
                        const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
-        binary_writer<char>(o, error_handler).write_cbor(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_cbor(j);
+        writer.flush();
    }

    /// @brief create a MessagePack serialization of a given JSON value
@@ -6348,7 +6353,7 @@ public:
            const error_handler_t error_handler = error_handler_t::keep)
    {
        std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
        vector_writer(result, error_handler).write_msgpack(j);
        return result;
    }
@@ -6358,7 +6363,9 @@ public:
    static void to_msgpack(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                           const error_handler_t error_handler = error_handler_t::keep)
    {
-        binary_writer<std::uint8_t>(o, error_handler).write_msgpack(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_msgpack(j);
+        writer.flush();
    }

    /// @brief create a MessagePack serialization of a given JSON value
@@ -6366,7 +6373,9 @@ public:
    static void to_msgpack(const basic_json& j, detail::output_adapter<char> o,
                           const error_handler_t error_handler = error_handler_t::keep)
    {
-        binary_writer<char>(o, error_handler).write_msgpack(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_msgpack(j);
+        writer.flush();
    }

    /// @brief create a UBJSON serialization of a given JSON value
@@ -6377,7 +6386,7 @@ public:
            const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
        std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
        vector_writer(result, error_handler).write_ubjson(j, use_size, use_type);
        return result;
    }
@@ -6388,7 +6397,9 @@ public:
                          const bool use_size = false, const bool use_type = false,
                          const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
-        binary_writer<std::uint8_t>(o, error_handler).write_ubjson(j, use_size, use_type);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type);
+        writer.flush();
    }

    /// @brief create a UBJSON serialization of a given JSON value
@@ -6397,7 +6408,9 @@ public:
                          const bool use_size = false, const bool use_type = false,
                          const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
-        binary_writer<char>(o, error_handler).write_ubjson(j, use_size, use_type);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type);
+        writer.flush();
    }

    /// @brief create a BJData serialization of a given JSON value
@@ -6409,7 +6422,7 @@ public:
            const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
        std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
        vector_writer(result, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
        return result;
    }
@@ -6421,7 +6434,9 @@ public:
                          const bjdata_version_t version = bjdata_version_t::draft2,
                          const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
-        binary_writer<std::uint8_t>(o, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type, true, true, version);
+        writer.flush();
    }

    /// @brief create a BJData serialization of a given JSON value
@@ -6431,7 +6446,9 @@ public:
                          const bjdata_version_t version = bjdata_version_t::draft2,
                          const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
-        binary_writer<char>(o, error_handler).write_ubjson(j, use_size, use_type, true, true, version);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_ubjson(j, use_size, use_type, true, true, version);
+        writer.flush();
    }

    /// @brief create a BSON serialization of a given JSON value
@@ -6440,7 +6457,7 @@ public:
            const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
        std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
        vector_writer(result, error_handler).write_bson(j);
        return result;
    }
@@ -6450,7 +6467,9 @@ public:
    static void to_bson(const basic_json& j, detail::output_adapter<std::uint8_t> o,
                        const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
-        binary_writer<std::uint8_t>(o, error_handler).write_bson(j);
+        binary_writer<std::uint8_t> writer(o, error_handler);
+        writer.write_bson(j);
+        writer.flush();
    }

    /// @brief create a BSON serialization of a given JSON value
@@ -6458,7 +6477,9 @@ public:
    static void to_bson(const basic_json& j, detail::output_adapter<char> o,
                        const error_handler_t error_handler = detail::binary_writer_default_error_handler())
    {
-        binary_writer<char>(o, error_handler).write_bson(j);
+        binary_writer<char> writer(o, error_handler);
+        writer.write_bson(j);
+        writer.flush();
    }

    /// @brief create a BON8 serialization of a given JSON value
@@ -6466,7 +6487,7 @@ public:
    static std::vector<std::uint8_t> to_bon8(const basic_json& j)
    {
        std::vector<std::uint8_t> result;
-        result.reserve(detail::binary_reserve_hint(j));
+        result.reserve(detail::binary_initial_capacity(j));
        vector_writer(result).write_bon8(j);
        return result;
    }
@@ -6475,14 +6496,18 @@ public:
    /// @sa https://json.nlohmann.me/api/basic_json/to_bon8/
    static void to_bon8(const basic_json& j, detail::output_adapter<std::uint8_t> o)
    {
-        binary_writer<std::uint8_t>(o).write_bon8(j);
+        binary_writer<std::uint8_t> writer(o);
+        writer.write_bon8(j);
+        writer.flush();
    }

    /// @brief create a BON8 serialization of a given JSON value
    /// @sa https://json.nlohmann.me/api/basic_json/to_bon8/
    static void to_bon8(const basic_json& j, detail::output_adapter<char> o)
    {
-        binary_writer<char>(o).write_bon8(j);
+        binary_writer<char> writer(o);
+        writer.write_bon8(j);
+        writer.flush();
    }

    /// @brief create a JSON value from an input in CBOR format
@@ -6878,6 +6903,72 @@ public:
    /// @name JSON Patch functions
    /// @{

+  private:
+    /// the members of an operation of a JSON patch that are of interest
+    struct patch_members
+    {
+        basic_json* op = nullptr;
+        basic_json* path = nullptr;
+        basic_json* value = nullptr;
+        basic_json* from = nullptr;
+    };
+
+    /// @brief find the members "op", "path", "value", and "from" of the object @a val in one pass
+    static patch_members find_patch_members(const basic_json& val, std::true_type /*key_type is string_t*/)
+    {
+        patch_members members;
+        for (auto& member : *val.m_data.m_value.object)
+        {
+            const string_t& key = member.first;
+            switch (key.size())
+            {
+                case 2:
+                    if (key == "op")
+                    {
+                        members.op = &member.second;
+                    }
+                    break;
+                case 4:
+                    if (key == "path")
+                    {
+                        members.path = &member.second;
+                    }
+                    else if (key == "from")
+                    {
+                        members.from = &member.second;
+                    }
+                    break;
+                case 5:
+                    if (key == "value")
+                    {
+                        members.value = &member.second;
+                    }
+                    break;
+                default:
+                    break;
+            }
+        }
+        return members;
+    }
+
+    /// @brief find the members by lookup for key types other than string_t
+    static patch_members find_patch_members(const basic_json& val, std::false_type /*key_type is not string_t*/)
+    {
+        const auto find_member = [&val](const char* name) -> basic_json *
+        {
+            const auto it = val.m_data.m_value.object->find(string_t(name));
+            return it != val.m_data.m_value.object->end() ? &it->second : nullptr;
+        };
+
+        patch_members members;
+        members.op = find_member("op");
+        members.path = find_member("path");
+        members.value = find_member("value");
+        members.from = find_member("from");
+        return members;
+    }
+
+  public:
    /// @brief applies a JSON patch in-place without copying the object
    /// @sa https://json.nlohmann.me/api/basic_json/patch/
    void patch_inplace(const basic_json& json_patch)
@@ -7051,18 +7142,18 @@ public:
        // a diff creates them: apply the operation and return true; return
        // false, with nothing changed, for all others
        string_t token;
-        const string_t value_key = "value";
-        const auto try_plain = [&result, &token, &value_key](patch_operations operation, const string_t& path, const basic_json & val)
+        typename json_pointer::template plain_steps<basic_json> steps;
+        const auto try_plain = [&result, &token, &steps](patch_operations operation, const string_t& path, const patch_members & members, const basic_json & val)
        {
-            if (path.empty() || path.find_first_of('~') != string_t::npos)
+            if (path.empty())
            {
                return false;
            }

            // a missing "value" is reported below
-            const auto value_it = val.m_data.m_value.object->find(value_key);
-            const basic_json& value = value_it != val.m_data.m_value.object->end() ? value_it->second : val;
-            if (operation != patch_operations::remove && value_it == val.m_data.m_value.object->end())
+            const basic_json* const value_ptr = members.value;
+            const basic_json& value = value_ptr != nullptr ? *value_ptr : val;
+            if (operation != patch_operations::remove && value_ptr == nullptr)
            {
                return false;
            }
@@ -7071,7 +7162,7 @@ public:
            {
                case patch_operations::replace:
                {
-                    basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token);
+                    basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token, steps);
                    if (target == nullptr)
                    {
                        return false;
@@ -7082,7 +7173,7 @@ public:

                case patch_operations::test:
                {
-                    const basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token);
+                    const basic_json* const target = json_pointer::resolve_plain(&result, path, path.size(), token, steps);
                    // a failing test is reported below
                    return target != nullptr && *target == value;
                }
@@ -7102,13 +7193,17 @@ public:
                    }
                    --slash;

-                    basic_json* const parent = json_pointer::resolve_plain(&result, path, slash, token);
+                    basic_json* const parent = json_pointer::resolve_plain(&result, path, slash, token, steps);
                    if (parent == nullptr)
                    {
                        return false;
                    }
                    token.clear();
                    token.append(path.data() + slash + 1, path.size() - slash - 1);
+                    if (token.find_first_of('~') != string_t::npos)
+                    {
+                        return false;
+                    }

                    if (operation == patch_operations::add)
                    {
@@ -7164,31 +7259,32 @@ public:
            }
        };

+        // the name of the first member to check
+        const string_t op_member = "op";
+
        // iterate and apply the operations
        for (const auto& val : json_patch)
        {
            // wrapper to get a value for an operation
-            const auto get_value = [&val](const string_t& op,
-                                          const string_t& member,
+            const auto get_value = [&val](basic_json * found,
+                                          const string_t& op,
+                                          const char* member,
                                          bool string_type) -> basic_json &
            {
-                // find value
-                auto it = val.m_data.m_value.object->find(member);
-
                // check if the desired value is present
-                if (JSON_HEDLEY_UNLIKELY(it == val.m_data.m_value.object->end()))
+                if (JSON_HEDLEY_UNLIKELY(found == nullptr))
                {
-                    detail::throw_patch_missing_member(op, member, false, val);
+                    detail::throw_patch_missing_member(op, string_t(member), false, val);
                }

                // check if the result is of type string
-                if (JSON_HEDLEY_UNLIKELY(string_type && !it->second.is_string()))
+                if (JSON_HEDLEY_UNLIKELY(string_type && !found->is_string()))
                {
-                    detail::throw_patch_missing_member(op, member, true, val);
+                    detail::throw_patch_missing_member(op, string_t(member), true, val);
                }

                // no error: return value
-                return it->second;
+                return *found;
            };

            // type check: every element of the array must be an object
@@ -7197,26 +7293,30 @@ public:
                detail::throw_patch_not_an_array(val);
            }

+            const patch_members members = find_patch_members(val, std::is_same<typename object_t::key_type, string_t> {});
+
            // collect mandatory members
-            const string_t& op = *get_value("op", "op", true).m_data.m_value.string;
-            const string_t& path = *get_value(op, "path", true).m_data.m_value.string;
+            const string_t& op = *get_value(members.op, op_member, "op", true).m_data.m_value.string;
+            const string_t& path = *get_value(members.path, op, "path", true).m_data.m_value.string;
            const patch_operations operation = get_op(op);

            // the usual operations on a plain pointer are applied without
            // building the pointer's reference tokens; anything else (and
            // every error) is handled below
-            if (try_plain(operation, path, val))
+            if (try_plain(operation, path, members, val))
            {
                continue;
            }

+            // the general operations may change any container
+            steps.reset();
            json_pointer ptr(path);

            switch (operation)
            {
                case patch_operations::add:
                {
-                    operation_add(ptr, get_value("add", "value", false));
+                    operation_add(ptr, get_value(members.value, "add", "value", false));
                    break;
                }

@@ -7229,13 +7329,13 @@ public:
                case patch_operations::replace:
                {
                    // the "path" location must exist - use at()
-                    result.at(ptr) = get_value("replace", "value", false);
+                    result.at(ptr) = get_value(members.value, "replace", "value", false);
                    break;
                }

                case patch_operations::move:
                {
-                    const string_t& from_path = *get_value("move", "from", true).m_data.m_value.string;
+                    const string_t& from_path = *get_value(members.from, "move", "from", true).m_data.m_value.string;
                    json_pointer from_ptr(from_path);

                    if (JSON_HEDLEY_UNLIKELY(is_proper_prefix(from_ptr, ptr)))
@@ -7258,7 +7358,7 @@ public:

                case patch_operations::copy:
                {
-                    const string_t& from_path = *get_value("copy", "from", true).m_data.m_value.string;
+                    const string_t& from_path = *get_value(members.from, "copy", "from", true).m_data.m_value.string;
                    const json_pointer from_ptr(from_path);

                    // The copy is functionally identical to an "add"
@@ -7276,7 +7376,7 @@ public:
                    {
                        // check if "value" matches the one at "path"
                        // the "path" location must exist - use at()
-                        success = (result.at(ptr) == get_value("test", "value", false));
+                        success = (result.at(ptr) == get_value(members.value, "test", "value", false));
                    }
                    JSON_INTERNAL_CATCH (out_of_range&)
                    {
@@ -7368,13 +7468,18 @@ public:
    /// @brief append an operation @a op for @a path to @a result, with @a value if it is not null
    static void diff_operation(basic_json& result, const char* op, const string_t& path, const basic_json* value)
    {
+        // the keys are distinct and in the order of std::map; the members are
+        // added without looking them up
        basic_json entry(value_t::object);
-        entry["op"] = op;
-        entry["path"] = path;
+        auto& members = *entry.m_data.m_value.object;
+        auto hint = members.end();
+        json_pointer::add_flattened(members, hint, string_t("op"), string_t(op), detail::priority_tag<2> {});
+        json_pointer::add_flattened(members, hint, string_t("path"), path, detail::priority_tag<2> {});
        if (value != nullptr)
        {
-            entry["value"] = *value;
+            json_pointer::add_flattened(members, hint, string_t("value"), *value, detail::priority_tag<2> {});
        }
+        entry.set_parents();
        result.push_back(std::move(entry));
    }

@@ -7396,6 +7501,62 @@ public:
        diff_operation(result, "add", path, &value);
    }

+    /// @brief the values in the target of the members of a source object, for the
+    ///        objects of up to 16 members without allocating
+    ///
+    /// reserve() must be called with the number of values that are added.
+    class diff_target_values
+    {
+      public:
+        diff_target_values() noexcept
+            : inline_items() // NOLINT(readability-redundant-member-init)
+        {}
+        diff_target_values(const diff_target_values&) = delete;
+        diff_target_values& operator=(const diff_target_values&) = delete;
+        diff_target_values(diff_target_values&&) = delete;
+        diff_target_values& operator=(diff_target_values&&) = delete;
+        ~diff_target_values() = default;
+
+        void reserve(const std::size_t n)
+        {
+            if (n > inline_size)
+            {
+                large.resize(n);
+                items = large.data();
+            }
+        }
+
+        void push_back(const basic_json* value)
+        {
+            items[count++] = value;
+        }
+
+        std::size_t size() const noexcept
+        {
+            return count;
+        }
+
+        const basic_json* operator[](const std::size_t i) const noexcept
+        {
+            return items[i];
+        }
+
+      private:
+        static constexpr std::size_t inline_size = 16;
+        /// the values of a small object, left uninitialized: they are written
+        /// before they are read (the user-provided constructor keeps the
+        /// member initialization, which -Weffc++ asks for, from zeroing them)
+        struct uninitialized_items : std::array<const basic_json*, inline_size>
+        {
+            uninitialized_items() noexcept {} // NOLINT(modernize-use-equals-default,hicpp-use-equals-default,cppcoreguidelines-pro-type-member-init,hicpp-member-init): see above
+        };
+
+        std::size_t count = 0;
+        std::vector<const basic_json*> large{}; // NOLINT(readability-redundant-member-init)
+        uninitialized_items inline_items;
+        const basic_json** items = inline_items.data();
+    };
+
    /// @brief append "/" and the decimal representation of @a index to @a path
    static void diff_append_index(string_t& path, std::size_t index)
    {
@@ -7476,8 +7637,9 @@ public:
    operations that remove every member of @a source and add every member of
    @a target to @a result, and returns false.
    */
+    template<typename TargetValues>
    static bool diff_object_keys(basic_json& result, const basic_json& source, const basic_json& target,
-                                 const string_t& path, std::vector<const basic_json*>& target_values,
+                                 const string_t& path, TargetValues& target_values,
                                 std::vector<const_iterator>& added)
    {
        // first pass: one lookup per source key
@@ -7691,7 +7853,7 @@ public:

            case value_t::object:
            {
-                std::vector<const basic_json*> target_values;
+                diff_target_values target_values;
                std::vector<const_iterator> added;
                if (diff_object_keys(result, source, target, path, target_values, added))
                {
@@ -492,3 +492,278 @@ TEST_CASE("binary_reserve_hint never over-reserves")
        CHECK(nlohmann::detail::binary_reserve_hint(json({{"a", 1}, {"b", 2}})) == 5);
    }
 }
+
+namespace
+{
+
+// numbers around every boundary of the integer encodings, and floats of each
+// kind (exact in float, needing double, special values)
+std::vector<json> boundary_numbers()
+{
+    std::vector<json> numbers;
+    for (int k = 0; k < 64; ++k)
+    {
+        const std::uint64_t p = std::uint64_t{1} << k;
+        for (const std::uint64_t d :
+                {
+                    p - 1, p, p + 1
+                })
+        {
+            numbers.emplace_back(d);
+            if (d <= static_cast<std::uint64_t>((std::numeric_limits<std::int64_t>::max)()))
+            {
+                numbers.emplace_back(static_cast<std::int64_t>(d));
+                numbers.emplace_back(-static_cast<std::int64_t>(d));
+                numbers.emplace_back(-static_cast<std::int64_t>(d) - 1);
+            }
+        }
+    }
+    numbers.emplace_back((std::numeric_limits<std::int64_t>::min)());
+    numbers.emplace_back((std::numeric_limits<std::uint64_t>::max)());
+    for (const double f :
+            {
+                0.0, -0.0, 1.0, -1.0, 0.5, 0.1, 3.25, 1e300, -1e-300, 3.4028234663852886e38, 1e39,
+                std::numeric_limits<double>::infinity(), -std::numeric_limits<double>::infinity(),
+                std::numeric_limits<double>::quiet_NaN()
+            })
+    {
+        numbers.emplace_back(f);
+    }
+    return numbers;
+}
+
+// the concatenation of the encodings of the elements of @a numbers[first, first + count)
+template<typename Single>
+std::vector<std::uint8_t> concat_singles(const std::vector<json>& numbers, const std::size_t first, const std::size_t count, Single single)
+{
+    std::vector<std::uint8_t> bytes;
+    for (std::size_t i = first; i < first + count; ++i)
+    {
+        const auto one = single(numbers[i]);
+        bytes.insert(bytes.end(), one.begin(), one.end());
+    }
+    return bytes;
+}
+
+} // namespace
+
+// Runs of numbers in arrays are encoded straight into space reserved from the
+// sink, by code that is separate from the encoding of a single value. Both have
+// to agree on every number: an array is its head and the encodings of its
+// elements, one after the other.
+TEST_CASE("runs of numbers in arrays are encoded as the numbers one by one")
+{
+    const std::vector<json> numbers = boundary_numbers();
+
+    // arrays of every length that a one-byte head covers, so that runs are as
+    // long as, and longer and shorter than, what fits one reservation
+    for (const std::size_t count :
+            {
+                1u, 2u, 3u, 5u, 14u, 15u, 23u
+            })
+    {
+        for (std::size_t first = 0; first + count <= numbers.size(); first += 7)
+        {
+            const json array(std::vector<json>(numbers.begin() + static_cast<std::ptrdiff_t>(first),
+                                               numbers.begin() + static_cast<std::ptrdiff_t>(first + count)));
+            CAPTURE(array.dump(-1, ' ', false, json::error_handler_t::replace))
+
+            std::vector<std::uint8_t> expected{static_cast<std::uint8_t>(0x80 + count)};
+            auto elements = concat_singles(numbers, first, count, [](const json & j)
+            {
+                return json::to_cbor(j);
+            });
+            expected.insert(expected.end(), elements.begin(), elements.end());
+            CHECK(json::to_cbor(array) == expected);
+
+            expected.assign(1, static_cast<std::uint8_t>(count <= 15 ? 0x90 + count : 0xDC));
+            if (count > 15)
+            {
+                expected.push_back(0);
+                expected.push_back(static_cast<std::uint8_t>(count));
+            }
+            elements = concat_singles(numbers, first, count, [](const json & j)
+            {
+                return json::to_msgpack(j);
+            });
+            expected.insert(expected.end(), elements.begin(), elements.end());
+            CHECK(json::to_msgpack(array) == expected);
+
+            for (const bool bjdata :
+                    {
+                        false, true
+                    })
+            {
+                expected.assign(1, static_cast<std::uint8_t>('['));
+                elements = concat_singles(numbers, first, count, [bjdata](const json & j)
+                {
+                    return bjdata ? json::to_bjdata(j) : json::to_ubjson(j);
+                });
+                expected.insert(expected.end(), elements.begin(), elements.end());
+                expected.push_back(static_cast<std::uint8_t>(']'));
+                CHECK((bjdata ? json::to_bjdata(array) : json::to_ubjson(array)) == expected);
+            }
+
+            // BON8 has no integers above the int64 range
+            bool representable = true;
+            for (std::size_t i = first; i < first + count; ++i)
+            {
+                representable = representable && bon8_representable(numbers[i]);
+            }
+            if (representable)
+            {
+                expected.assign(1, static_cast<std::uint8_t>(count <= 4 ? 0x80 + count : 0x85));
+                elements = concat_singles(numbers, first, count, [](const json & j)
+                {
+                    return json::to_bon8(j);
+                });
+                expected.insert(expected.end(), elements.begin(), elements.end());
+                if (count > 4)
+                {
+                    expected.push_back(0xFE);
+                }
+                CHECK(json::to_bon8(array) == expected);
+            }
+            else
+            {
+                CHECK_THROWS_AS(json::to_bon8(array), json::out_of_range&);
+            }
+        }
+    }
+}
+
+// An array that is written in more than one reservation (more numbers than fit
+// one), and arrays of an index with more digits, round trip for every format.
+TEST_CASE("large arrays and many digits of indices round trip")
+{
+    json array = json::array();
+    for (int i = 0; i < 12000; ++i)
+    {
+        switch (i % 4)
+        {
+            case 0:
+                array.push_back(i);
+                break;
+            case 1:
+                array.push_back(-i);
+                break;
+            case 2:
+                array.push_back(i * 0.25);
+                break;
+            default:
+                array.push_back(i % 8 == 3);
+                break;
+        }
+    }
+    json document = json::object();
+    document["numbers"] = array;
+    document["nested"] = json::array({array, "text", array});
+
+    CHECK(json::from_cbor(json::to_cbor(document)) == document);
+    CHECK(json::from_msgpack(json::to_msgpack(document)) == document);
+    CHECK(json::from_ubjson(json::to_ubjson(document)) == document);
+    CHECK(json::from_ubjson(json::to_ubjson(document, true, true)) == document);
+    CHECK(json::from_bjdata(json::to_bjdata(document)) == document);
+    CHECK(json::from_bson(json::to_bson(document)) == document);
+    CHECK(json::from_bon8(json::to_bon8(document)) == document);
+}
+
+// Strings and keys are written with their head in one reservation up to a
+// length; around that length, and with every kind of head, they must read back.
+TEST_CASE("strings of every head length round trip")
+{
+    json document = json::object();
+    json array = json::array();
+    for (const std::size_t length :
+            {
+                0u, 1u, 2u, 3u, 4u, 7u, 8u, 15u, 16u, 17u, 23u, 24u, 25u, 31u, 32u, 33u, 63u, 127u, 128u, 129u, 200u, 245u, 246u, 247u, 248u, 249u, 250u, 251u, 252u, 253u, 254u, 255u, 256u, 257u, 300u, 1000u, 65535u, 65536u, 70000u
+            })
+    {
+        const std::string text(length, 'x');
+        array.push_back(text);
+        document[std::string(length, 'k')] = text;
+    }
+    document["array"] = array;
+
+    CHECK(json::from_cbor(json::to_cbor(document)) == document);
+    CHECK(json::from_msgpack(json::to_msgpack(document)) == document);
+    CHECK(json::from_ubjson(json::to_ubjson(document)) == document);
+    CHECK(json::from_ubjson(json::to_ubjson(document, true, true)) == document);
+    CHECK(json::from_bjdata(json::to_bjdata(document)) == document);
+    CHECK(json::from_bson(json::to_bson(document)) == document);
+    CHECK(json::from_bon8(json::to_bon8(document)) == document);
+}
+
+#if !defined(JSON_NOEXCEPTION)
+// What a write that throws leaves behind in an output adapter is what was written
+// before it, whatever the adapter and however the adapter sink buffers. (Without
+// exceptions, the writes in CHECK_THROWS_AS do not run.)
+TEST_CASE("a write that throws leaves the same bytes in every kind of output")
+{
+    json array = json::array();
+    for (int i = 0; i < 700; ++i)
+    {
+        array.push_back(i * 1000);
+        array.push_back("text " + std::to_string(i));
+    }
+    array.push_back("bad \xFF byte");
+    json document = json::object();
+    document["first"] = array;
+    document["second"] = 1;
+
+    const auto strict = json::error_handler_t::strict;
+
+    SECTION("cbor")
+    {
+        std::vector<std::uint8_t> as_vector;
+        std::string as_string;
+        std::ostringstream as_stream;
+        CHECK_THROWS_AS(json::to_cbor(document, as_vector, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_cbor(document, as_string, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_cbor(document, as_stream, strict), json::type_error&);
+        CHECK(!as_vector.empty());
+        CHECK(std::string(reinterpret_cast<const char*>(as_vector.data()), as_vector.size()) == as_string); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+        CHECK(as_stream.str() == as_string);
+    }
+
+    SECTION("msgpack")
+    {
+        std::vector<std::uint8_t> as_vector;
+        std::string as_string;
+        std::ostringstream as_stream;
+        CHECK_THROWS_AS(json::to_msgpack(document, as_vector, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_msgpack(document, as_string, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_msgpack(document, as_stream, strict), json::type_error&);
+        CHECK(!as_vector.empty());
+        CHECK(std::string(reinterpret_cast<const char*>(as_vector.data()), as_vector.size()) == as_string); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+        CHECK(as_stream.str() == as_string);
+    }
+
+    SECTION("ubjson")
+    {
+        std::vector<std::uint8_t> as_vector;
+        std::string as_string;
+        std::ostringstream as_stream;
+        CHECK_THROWS_AS(json::to_ubjson(document, as_vector, false, false, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_ubjson(document, as_string, false, false, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_ubjson(document, as_stream, false, false, strict), json::type_error&);
+        CHECK(!as_vector.empty());
+        CHECK(std::string(reinterpret_cast<const char*>(as_vector.data()), as_vector.size()) == as_string); // NOLINT(cppcoreguidelines-pro-type-reinterpret-cast)
+        CHECK(as_stream.str() == as_string);
+    }
+
+    SECTION("bson")
+    {
+        // the size pass finds the string before anything is written
+        std::vector<std::uint8_t> as_vector;
+        std::string as_string;
+        std::ostringstream as_stream;
+        CHECK_THROWS_AS(json::to_bson(document, as_vector, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_bson(document, as_string, strict), json::type_error&);
+        CHECK_THROWS_AS(json::to_bson(document, as_stream, strict), json::type_error&);
+        CHECK(as_vector.empty());
+        CHECK(as_string.empty());
+        CHECK(as_stream.str().empty());
+    }
+}
+#endif
@@ -285,6 +285,58 @@ TEST_CASE("custom number types: integers beyond 64 bits")
        CHECK(wide_json::to_bjdata(wide_json::array({uint_big, uint_big}), true, true) == expected_bjdata_array);
    }

+    SECTION("integers in arrays and objects")
+    {
+        // runs of numbers and scalar object members are encoded by loops of
+        // their own for the number types up to 64 bits; a wider number type
+        // takes the generic path, which checks the range of every integer
+        const wide_json run = wide_json::array({make_int(1), make_int(-3), make_uint(7), wide_json(1.5), int_big, make_int(5)});
+        const wide_json run_negative = wide_json::array({make_int(1), make_int(-3), int_big_negative});
+        const wide_json run_unsigned = wide_json::array({make_uint(1), make_uint(7), uint_big});
+        const wide_json members = wide_json::object({{"a", make_int(1)}, {"b", wide_json(1.5)}, {"c", int_big}});
+
+        const std::vector<wide_json> failing = {run, run_negative, run_unsigned, members, wrap(run), wrap(members)};
+        for (const auto& value : failing)
+        {
+            CHECK_THROWS_AS(_ = wide_json::to_cbor(value), out_of_range&);
+            CHECK_THROWS_AS(_ = wide_json::to_msgpack(value), out_of_range&);
+            CHECK_THROWS_AS(_ = wide_json::to_bson(wrap(value)), out_of_range&);
+            CHECK_THROWS_AS(_ = wide_json::to_bon8(value), out_of_range&);
+        }
+
+#if !defined(JSON_NOEXCEPTION)
+        // a failing write leaves no more than the elements before the failing one
+        // (without exceptions, the writes in CHECK_THROWS_AS do not run)
+        std::vector<std::uint8_t> out;
+        CHECK_THROWS_AS(wide_json::to_cbor(run, out), out_of_range&);
+        // (the head of the array, 1, -3, 7, and 1.5 as a single-precision float)
+        CHECK(out == std::vector<std::uint8_t>({0x86, 0x01, 0x22, 0x07, 0xFA, 0x3F, 0xC0, 0x00, 0x00}));
+        out.clear();
+        CHECK_THROWS_AS(wide_json::to_bon8(run, out), out_of_range&);
+        CHECK(!out.empty());
+#endif
+
+        // the same values that fit are written exactly as with 64-bit number types
+        const wide_json fits = wide_json::array({make_int(1), make_int(-3), make_uint(7), wide_json(1.5), make_int(min_int64), make_int(max_int64), make_int(200), make_int(-70000)});
+        const nlohmann::json fits64 = nlohmann::json::array({1, -3, 7, 1.5, (std::numeric_limits<std::int64_t>::min)(), (std::numeric_limits<std::int64_t>::max)(), 200, -70000});
+        CHECK(wide_json::to_cbor(fits) == nlohmann::json::to_cbor(fits64));
+        CHECK(wide_json::to_msgpack(fits) == nlohmann::json::to_msgpack(fits64));
+        CHECK(wide_json::to_ubjson(fits) == nlohmann::json::to_ubjson(fits64));
+        CHECK(wide_json::to_ubjson(fits, true, true) == nlohmann::json::to_ubjson(fits64, true, true));
+        CHECK(wide_json::to_bjdata(fits) == nlohmann::json::to_bjdata(fits64));
+        CHECK(wide_json::to_bjdata(fits, true, true) == nlohmann::json::to_bjdata(fits64, true, true));
+        CHECK(wide_json::to_bon8(fits) == nlohmann::json::to_bon8(fits64));
+        CHECK(wide_json::to_bson(wrap(fits)) == nlohmann::json::to_bson(nlohmann::json::object({{"v", fits64}})));
+
+        const wide_json fit_members = wide_json::object({{"a", make_int(1)}, {"b", make_int(-300)}, {"c", make_uint(max_uint64)}, {"d", wide_json(2.5)}});
+        const nlohmann::json fit_members64 = nlohmann::json::object({{"a", 1}, {"b", -300}, {"c", (std::numeric_limits<std::uint64_t>::max)()}, {"d", 2.5}});
+        CHECK(wide_json::to_cbor(fit_members) == nlohmann::json::to_cbor(fit_members64));
+        CHECK(wide_json::to_msgpack(fit_members) == nlohmann::json::to_msgpack(fit_members64));
+        CHECK(wide_json::to_ubjson(fit_members) == nlohmann::json::to_ubjson(fit_members64));
+        CHECK(wide_json::to_bjdata(fit_members) == nlohmann::json::to_bjdata(fit_members64));
+        CHECK(wide_json::to_bson(fit_members) == nlohmann::json::to_bson(fit_members64));
+    }
+
    SECTION("BJData ND-array")
    {
        // an ND-array element beyond 64 bits does not fit any dtype, so the
@@ -1580,3 +1580,94 @@ TEST_CASE("invalid UTF-8 at every position is handled like the reference")
        }
    }
 }
+
+TEST_CASE("short strings are written like the reference, as values and as keys")
+{
+    // strings of up to 16 characters (32 with vector instructions) that need no
+    // escaping are written as a whole (together with the quotes and, for a key,
+    // the colon); everything else, including a string that is merely short,
+    // takes the general path.
+    // Every length around the edges of the size classes of the fast path is
+    // combined with each kind of byte that must send a string to the general
+    // path, at each position, and compared with the reference for values and,
+    // via the value, for keys.
+    const std::vector<std::string> specials =
+    {
+        "\"", "\\", "\n", "\x1f", "\x7f", "\xc3\xa9", "\xe2\x82\xac", "\xf0\x9f\x98\x80", "\xff", "\x80", "\xc3"
+    };
+
+    for (std::size_t length = 0; length <= 36; ++length)
+    {
+        const std::string plain(length, 'k');
+        check_dump_against_reference(plain, false);
+
+        for (const auto& special : specials)
+        {
+            for (std::size_t pos = 0; pos <= length; ++pos)
+            {
+                std::string s = plain;
+                s.insert(pos, special);
+                check_dump_against_reference(s, false);
+
+                for (const bool ensure_ascii :
+                        {
+                            false, true
+                        })
+                {
+                    // a key and a value are written alike, in compact and
+                    // in pretty-printed form
+                    const json value = s;
+                    const json object = json::object({{s, 1}});
+                    for (const auto error_handler :
+                            {
+                                json::error_handler_t::replace, json::error_handler_t::keep
+                            })
+                    {
+                        const std::string quoted = value.dump(-1, ' ', ensure_ascii, error_handler);
+                        CHECK(object.dump(-1, ' ', ensure_ascii, error_handler) == "{" + quoted + ":1}");
+                        CHECK(object.dump(2, ' ', ensure_ascii, error_handler) == "{\n  " + quoted + ": 1\n}");
+                    }
+                }
+            }
+        }
+    }
+}
+
+TEST_CASE("runs of multi-byte characters are escaped like the reference under ensure_ascii")
+{
+    // a run of well-formed multi-byte characters is escaped straight into the
+    // write buffer; it has to flush that mid-run, stop at an ASCII character or
+    // an ill-formed sequence and leave those to the general path, and pick up
+    // behind what the general path has pending
+    const std::vector<std::string> characters =
+    {
+        "\xc2\x80", "\xdf\xbf", "\xe0\xa0\x80", "\xe2\x82\xac", "\xed\x9f\xbf", "\xef\xbf\xbf",
+        "\xf0\x90\x80\x80", "\xf0\x9f\x98\x80", "\xf4\x8f\xbf\xbf"
+    };
+    const std::vector<std::string> interruptions =
+    {
+        "", "a", "\n", "\"", "\x7f", "\xff", "\x80", "\xc3", "\xe2\x82", "\xf0\x9f\x98", "\xed\xa0\x80"
+    };
+
+    for (std::size_t shift = 0; shift <= 13; ++shift)
+    {
+        for (const auto& interruption : interruptions)
+        {
+            for (const std::size_t count :
+                    {
+                        std::size_t{1}, std::size_t{2}, std::size_t{85}, std::size_t{200}
+                    })
+            {
+                std::string s(shift, 'x');
+                for (std::size_t i = 0; i < count; ++i)
+                {
+                    s += characters[(i + shift) % characters.size()];
+                }
+                s += interruption;
+                s += characters[shift % characters.size()];
+                s += characters[(shift + 3) % characters.size()];
+                check_dump_against_reference(s, shift < 2);
+            }
+        }
+    }
+}
@@ -48,14 +48,16 @@ namespace
 class chunked_streambuf : public std::streambuf
 {
  public:
-    chunked_streambuf(std::string s, std::size_t chunk_size)
-        : m_data(std::move(s)), m_chunk_size(chunk_size)
+    // The first @a unbuffered characters are served without a get area, the
+    // rest in get areas of @a chunk_size.
+    chunked_streambuf(std::string s, std::size_t chunk_size, std::size_t unbuffered = 0)
+        : m_data(std::move(s)), m_chunk_size(chunk_size), m_unbuffered(unbuffered)
    {}

  protected:
    int_type underflow() override
    {
-        if (m_chunk_size == 0)
+        if (m_chunk_size == 0 || m_pos < m_unbuffered)
        {
            return m_pos < m_data.size() ? traits_type::to_int_type(m_data[m_pos]) : traits_type::eof();
        }
@@ -77,7 +79,7 @@ class chunked_streambuf : public std::streambuf

    int_type uflow() override
    {
-        if (m_chunk_size == 0)
+        if (m_chunk_size == 0 || m_pos < m_unbuffered)
        {
            return m_pos < m_data.size() ? traits_type::to_int_type(m_data[m_pos++]) : traits_type::eof();
        }
@@ -88,6 +90,7 @@ class chunked_streambuf : public std::streambuf
    std::string m_data;
    std::size_t m_pos = 0;
    std::size_t m_chunk_size;
+    std::size_t m_unbuffered;
 };

 // how a test reads from the stream
@@ -101,9 +104,9 @@ enum class read_mode
 // Read @a input from a chunked_streambuf and describe everything observable:
 // the values or the exception, the stream state after each value, and the
 // characters left in the stream.
-std::string observe(const std::string& input, std::size_t chunk_size, read_mode mode)
+std::string observe(const std::string& input, std::size_t chunk_size, read_mode mode, std::size_t unbuffered = 0)
 {
-    chunked_streambuf buf(input, chunk_size);
+    chunked_streambuf buf(input, chunk_size, unbuffered);
    std::istream is(&buf);
    std::string result;
    for (int i = 0; i < 4; ++i)
@@ -212,6 +215,31 @@ TEST_CASE("parsing from a std::istream scans the get area in bulk")
        }
    }

+    SECTION("results do not depend on a get area that appears late")
+    {
+        // the first characters come without a get area, then the streambuf
+        // starts to use one: the token that is read at that point is complete
+        for (const auto& input : corpus())
+        {
+            const std::string expected = observe(input, 0, read_mode::extract);
+            for (const std::size_t unbuffered :
+                    {
+                        1u, 2u, 3u, 5u, 8u, 13u, 21u
+                    })
+            {
+                for (const std::size_t chunk_size :
+                        {
+                            1u, 2u, 3u, 7u, 64u
+                        })
+                {
+                    CAPTURE(input)
+                    CAPTURE(unbuffered)
+                    CAPTURE(chunk_size)
+                    CHECK(observe(input, chunk_size, read_mode::extract, unbuffered) == expected);
+                }
+            }
+        }
+    }
 #endif // !defined(JSON_NOEXCEPTION)

    SECTION("values are the same as when parsing from a string")
@@ -0,0 +1,4 @@
+# astyle is deliberately pinned: newer versions reformat the code, and this version
+# defines the formatting that `make amalgamate` produces and check_amalgamation.yml enforces.
+# Dependabot is told to ignore it in .github/dependabot.yml.
+astyle==3.4.13
@@ -1,4 +1,62 @@
-# astyle is deliberately pinned: newer versions reformat the code, and this version
-# defines the formatting that `make amalgamate` produces and check_amalgamation.yml enforces.
-# Dependabot is told to ignore it in .github/dependabot.yml.
-astyle==3.4.13
+#
+# This file is autogenerated by pip-compile with Python 3.14
+# by the following command:
+#
+#    pip-compile --generate-hashes --no-emit-index-url --output-file=tools/astyle/requirements.txt --strip-extras tools/astyle/requirements.in
+#
+astyle==3.4.13 \
+    --hash=sha256:0456c5bb1bed18b30f7d9e9c7d6499684a62cc808949173d52bf6def655f48b9 \
+    --hash=sha256:061b320ea1d265ac01babdd82b056c7b55913c726c770cff26191af78884c5ab \
+    --hash=sha256:0ad050135504994bee50233db59bedf95f1343977118e27f495eaf2f401ecfe5 \
+    --hash=sha256:162cb29abc5ddd5d03ab48b15fd8603baf04c1ba5913e1ef475e09818744216b \
+    --hash=sha256:181d7b6da8eef2cc4dc757dc3381f0d1f2932d6ceb2f6ce5bea432de698e9a8f \
+    --hash=sha256:191abde742ff5ae788a489722a669f7bc0b18cefc1020b617bc0e84361b71a58 \
+    --hash=sha256:1bf59aa3006546d795443d1ff69682f989fe95e4c71850c67f8251eab183fee3 \
+    --hash=sha256:31c86eb8aae517b9cf1debd325ea5f6ac653d52787d5c5c9d56197afc9bc97e3 \
+    --hash=sha256:35b979a62d1499b2daf241517ddb0aee3ab4b82dc82aa3f8604a2db1e9cc4e66 \
+    --hash=sha256:36a9f14f88c6bc3c4240f977a3f3fd3d0eac84373bc10004258712e6b3cb08f2 \
+    --hash=sha256:46ad33bf795e68f4b18bf947cc7276e26579ab6df89574617c7797b0ed2cf473 \
+    --hash=sha256:4803a16330b67b98c994b87b984af00d7dcaff0d4af69d17ab0fd705141725fe \
+    --hash=sha256:481b6e195dfa56380bb19cd134dc698cd7f5732d7d5cd781c13b5c15112b8fc2 \
+    --hash=sha256:48f0218622e807360d50d9d419e527d8e4c0f7f49c51c6fac3d4c6f343a0573e \
+    --hash=sha256:4a3179783910ea386125d88a16e86cc6cfee41c513d0d8e692bdabd17577ce93 \
+    --hash=sha256:508d5debe7874a5dd1b491bbae0c48ebc94aa3736545718208bb33c648962e62 \
+    --hash=sha256:55b4335f20b717f6ec57a237d29246e3c7196bd014e509f2aa376aca36ea9967 \
+    --hash=sha256:56714bb2e2a0fe3911474b66e563aac508d97a9f48d5e55c8e182b1675db0e05 \
+    --hash=sha256:599ab574cd53b7105877d5e1b6ab0c933b771e4b2343b0bc54aa06af64653449 \
+    --hash=sha256:5b183422a1edbce971758b2b4a2b180127d9e8c46119183d410c77ec1d42a516 \
+    --hash=sha256:5c6de6933f7818b4380b309704c45da9330d906bbe479bb76023261e63545900 \
+    --hash=sha256:5f7efd14d9aaee69a4a33e60722a8a09d0b7cb834aed698ce0793cecec4e7452 \
+    --hash=sha256:66ecd8a2a778bc78bdb2df4f5f371f100c6f180ae859e7d414f374eb2636b531 \
+    --hash=sha256:6fa1c231c82a0ea6ccfa8e19276dec4753d9e084229d606f9522260fe37b955f \
+    --hash=sha256:76d53ee9e07f8ad32050363c8402923095020856c44015968b4dd9ea5b7a18a1 \
+    --hash=sha256:81f0173912da9cf687cf2ed1361026fd957fcfd5b9673803c50265db63427d7f \
+    --hash=sha256:81f4345adbbb463a2d5ce314a1ca689a0aa05a6240c86fc0225e92474e3f44b4 \
+    --hash=sha256:840e1d0436a1a3793ebe14c95fefe797f77bd78d6194a7c0bcdde705cb583403 \
+    --hash=sha256:8d6a69f9b316b3941cd43f16eb109f738bb171963408b5f135521ef70ad364f1 \
+    --hash=sha256:8e2879d770e3fbf53ff62728f4d62a6edcacd1ed3d62b36c012706d778048340 \
+    --hash=sha256:900e442ed002e5e05c31628fac817db62f6d28e1cc2426040c69a10d39f92651 \
+    --hash=sha256:920c8516f420da68765ced84692e550d0f34932663219c48db9b0c79544002d3 \
+    --hash=sha256:9417573994a9d264f873cd6122f9e0ea7044d628d9a8b9e84dbbc85b0e7c2880 \
+    --hash=sha256:95eaf11aadd770bccd833c8ffe222ea6055fe5c2c4b3974cb43f3366913b62a4 \
+    --hash=sha256:9b7471f4a5f174604cdfdf4f9091ba319edb52471e07672ea71745157d121fe7 \
+    --hash=sha256:9e016b099f809c8d4894fa6d36ef925973f52ebc50136306b1b4b9ce65094063 \
+    --hash=sha256:9e04dbd51f564316a4879448258d9b9e68ade9b062351ea6fc7356d0f42471a0 \
+    --hash=sha256:a002a3a1ac187ba1b441f678aca16b4ecea47a1f3d2462c3ec9c59e71e6083b7 \
+    --hash=sha256:a177039c38f6f6a43f46bcd3571756be385c0c47462e6a06eb5950397d0bdcbe \
+    --hash=sha256:a8963c261b980a0c12fc1523f86eab3edc67940c114ea544b21d2b44e9196d23 \
+    --hash=sha256:a97f8ef2a054ec067e3af65bc5b145c51b6c2a5f2efe997a72a9d958c3ab6d1a \
+    --hash=sha256:aa8ba72c1349b7e45d1ce01d69daecfac680713624d27d83ebf2a9d1b4984f65 \
+    --hash=sha256:abba627fb930a5398414ba398169f52e3865855501b5ee5430bb53d0974074c6 \
+    --hash=sha256:ac8db98b392b609f36cd7d6e7f18a247376723b4397578a6f450a445a9dacb3d \
+    --hash=sha256:c9d4b1a0a7fbc35b96273695143c791df9c1a4dcab0815ae6d9db8483314430f \
+    --hash=sha256:cf247a3bca29f43398af4e20c7c922b7f5130a22f306f068ad038b94e72854d6 \
+    --hash=sha256:d3da892b9ec2932a2cb49941aae79447f6baafe5023ee31680c9cabcc471b451 \
+    --hash=sha256:db8c4555b66067aba253c2f32dc4ff6ee92882dd3e95a00ebb3bbf145efe067d \
+    --hash=sha256:dde61d5acb6008ebefcecc37625b7f2776df2f368f92d47714e5f50276347c9f \
+    --hash=sha256:dede5398c41b3a00ad0a9f77c57c05444640f8b28500b9bb4624ab6f2d0a90b7 \
+    --hash=sha256:e3cc45784dbac7ade37967e24cbd6b5b89517f3df8fbed276dffffdcf289f04d \
+    --hash=sha256:eb4d38f4a6be78cd28ce44397a1244c08af15c1a00293c40634c2104b44c943c \
+    --hash=sha256:f9e6cd9ee60a8c42eab85cb57e0c82b1c947261b71f5091bbb9f61e864c64e07 \
+    --hash=sha256:fb1ddc816b483c83de15db98847a5d2dd3ba9df9f4a937ad47aabadeeaeed58c
+    # via -r tools/astyle/requirements.in
@@ -0,0 +1 @@
+jinja2==3.1.6
@@ -1 +1,163 @@
-jinja2==3.1.6
+#
+# This file is autogenerated by pip-compile with Python 3.14
+# by the following command:
+#
+#    pip-compile --generate-hashes --no-emit-index-url --output-file=tools/generate_natvis/requirements.txt --strip-extras tools/generate_natvis/requirements.in
+#
+jinja2==3.1.6 \
+    --hash=sha256:0137fb05990d35f1275a587e9aee6d56da821fc83491a0fb838183be43f66d6d \
+    --hash=sha256:85ece4451f492d0c13c5dd7c13a64681a86afae63a5f347908daf103ce6d2f67
+    # via -r tools/generate_natvis/requirements.in
+markupsafe==3.0.4 \
+    --hash=sha256:007e1ffd9bf65bb6ee96df7b258fc632a4868dd5566037986c64781f35a36e98 \
+    --hash=sha256:02fa4acbc6a3fc5c693c34d4dd8c1130b7fe99cc915181b0ddd6f72aeb296002 \
+    --hash=sha256:03470d1a8268e692ecf79ecd565593e59d44219377a7ead61f1f1b94c1f7ff6b \
+    --hash=sha256:04e7902ba80ee4bac1d50a549606527a1dcf0476cd81403db41099d3b60ec653 \
+    --hash=sha256:051417f74bcaaefa316276e0ff723f541616ca51043d070da00249d9bddd3e3c \
+    --hash=sha256:05295589e619b9bed252a86b532b8e27350abc372d18ba89b59375325e91ec1e \
+    --hash=sha256:06de8ef6331f6e822c28d577dc8bf43fe398800477c49498f38fc38b67ff33fc \
+    --hash=sha256:0764a13d34cae40db7bbf3a09b7e9b491bf4603e20b263a7a9d6b8e324975d0a \
+    --hash=sha256:077293e425f28ec737dbcad442a71752e28f8ae27cde3d68acd1fb212091cd92 \
+    --hash=sha256:0930db9bdc62d22944e10b066448bb65dc9abe9112880c7cab8da54db4284d5f \
+    --hash=sha256:0cee7cb0f9a1b6892ea482237d9403b3d1b4603aee057d0ff01f0fac2d019a97 \
+    --hash=sha256:0d9c47709875fdb321452056622e930c52afbc07a7d780762fbb8b4d91ce6fa4 \
+    --hash=sha256:11935df9bf455ed0c04eb87bcd720f02b1fe5e02128a9430f23aed6f93336fc7 \
+    --hash=sha256:12a606a492de952afcb43b59a14aaaaad120e708d3663dd0fdf2d738d427a691 \
+    --hash=sha256:14bd2d845d62ab678eaf81da89d7b621b51756c72346745c1a594c09d49207a2 \
+    --hash=sha256:15ba9e28640feef770374b116a6f019c21f52404aeabe516aa7f800587b98cfc \
+    --hash=sha256:18a801868a884f216e784d7d14db2a4077143ce7610440aee2ce8f734e7cfcde \
+    --hash=sha256:1c0df495a977d10460a94941799c72d5b5ab03d3858d949b55b5a66c8f371c99 \
+    --hash=sha256:1caa2fa5a6184fb233153b35f654e6687bd555476f6170f29d8ee9be1a8b0af9 \
+    --hash=sha256:1e1451fab512d1bcc3dc26988ec1edb0b82c2db909132872cd9356070a6b63df \
+    --hash=sha256:1f1f9477e174582b0a1b583d60b66e1f2cf5d3fe12cee985e4aedf44766600e5 \
+    --hash=sha256:2628d3a8cb648ecebb3c5d6b0a1052d400e4d8b7ac0fb786be8d285b50040d17 \
+    --hash=sha256:26e9867520db70d37f7fb421a7f0d8adb40171011fb84ce869afa1a83370dfa8 \
+    --hash=sha256:2a6ef68ae94aed8721934072b27a3b654ea2100b97e4ab864cf1489c90926fbc \
+    --hash=sha256:2b2b1e18af909b448bb3cf9e3433366f7a8726271fc214e8b10e0f62a78c724b \
+    --hash=sha256:2cb3dd71fc6be918ad4264346a8ed69485f9b7ed7bf35495d8e22807cd6b8bea \
+    --hash=sha256:2d1b7d9308288661f56672b1b157d75fc536714d3638487bbea17b6318a78248 \
+    --hash=sha256:2dad610540cb2e6272855c178f08ae9a1c7ac258a7fb71660553a5f104b42741 \
+    --hash=sha256:2e5a7cd7fdd14fcb1ae5d7d8bf23d24fbd1daefd1fbca2580132e1ea75f098b5 \
+    --hash=sha256:2e9ad7dd851bf45fab9f75cbff4cb493fee9979e8d8c7c9c3ee119022518edd6 \
+    --hash=sha256:340cbb1957ba99929cbf19a75626d36ba1ae21d1730b287d1cf7f824a20c4fc7 \
+    --hash=sha256:34bdde374c5932765d7dc685c4a1d191a3207852d67e8e0a9eb6ea85156181f1 \
+    --hash=sha256:353bd63081912ab8cfa6a0c7d185934cdf8426f04c618bba6bc4b394f2069b67 \
+    --hash=sha256:387d8cd30e69b3f0a72877b9ae717033396404e19095b17fe89753a981fda44f \
+    --hash=sha256:3882fb412298575bae3b9c46868251f15cc69307359f87bb1b382e53d6e5a2c9 \
+    --hash=sha256:38fc55594dab834470b6733dead2ee9e3f657fb0608c769dcafa0ba5ab52f45c \
+    --hash=sha256:396ec4e65cc889f69786b3b89478b471cee5a3bcf468b9d9bb03e1a30fb291fc \
+    --hash=sha256:39dbacefc411633db5b4378b066a9aca70a3d7e2922c9e578d825f844026eeba \
+    --hash=sha256:3a93d9616ddecfb393727a0041a562cf0b15a244e20f2bd25efc7949be4c4f17 \
+    --hash=sha256:3d23795802fc8bd72534836d64489bbf0f67c088959091bdb22e10735a5107bf \
+    --hash=sha256:434139499bb20b502ed3baa1f169e618f924a97e7a777fea1a49446d80106cf6 \
+    --hash=sha256:436e3ffc6310d3c41878c601db29098102fe5d8a467c49da4a4125254e0980f2 \
+    --hash=sha256:489505b03f692c3f376394e49194fa7a7f9e8558d6e293a7056a0032b0c38163 \
+    --hash=sha256:4a540e2d3192792fc84eced57bef37851ccb2b41f73291bb17408eea77bcd278 \
+    --hash=sha256:4a7cdc2a420ca01058182da4253329764d4bfa055564d1eced90e6ba1e8b1d3d \
+    --hash=sha256:4bced6e2a6dba6a28f7dd3c6ce14df1b2dd495923f16ea484cad03decd463b2b \
+    --hash=sha256:4cf3468d5ec187ffffcaca8e61929a37448f215dafc1386a12c750a72fe53634 \
+    --hash=sha256:4e2c4809c14559aa7ef426f27fb35afbb38104c349a903bf8f3600456764bb38 \
+    --hash=sha256:4ed644d75aa94a2baf7ec3a96eaa160ea58c742eb9d27c6506053c5c40fc84ed \
+    --hash=sha256:4f6e0852a0283b1b1fd776eeb7b766a5f440b3e2bd31ab51af3b400585f3965c \
+    --hash=sha256:5066b244f576f91afc8ee3ba029a89f99d39c79b1853fe9d39bea9f0afbec148 \
+    --hash=sha256:5086f9975abb1ab531ee6afca1761e4b59a19b446f3f6522ed776963228cfe5a \
+    --hash=sha256:50b5bedc9ed8a94fc8857a42ef4f84a81ea88f8d4f05dc8705fb23ee6d8dcca7 \
+    --hash=sha256:52704c5d36eb6dda8866493decd61111fff86244c9b1ad225ca01b9e91e5970f \
+    --hash=sha256:55ffd6ce583d97dc71dc92e930324c8c0d25aea7e3ade6ae54ef77cedb096811 \
+    --hash=sha256:569d65055d367e3dcdf30c3f41119467b73d9ee9faf332bdf40402644f5ac08e \
+    --hash=sha256:57f9947a7e57a081c1e3e0a2dd0d2dcf290a4531450e6f611e30084c222a7295 \
+    --hash=sha256:5989cb26b2e1efc6a42216a9f6b5ee495ce5ace2e5b352a9af489976b32d1ee2 \
+    --hash=sha256:5c22873ad1f0532ba40fa1727f3c0fc1bbbaab6d373d4cbe3f0dc74b2e2521c7 \
+    --hash=sha256:5e8b3d0b18fd623afa12ecb2ce8d8becef69f9b5440c6330c7972200e0bb84b0 \
+    --hash=sha256:61631e08084be9e21a8967ec3139c7616ed7c5e9368e05c86d1b39562c8a57b6 \
+    --hash=sha256:64511c54db4e4987aef4c41923235927428729e8174c5dba488429be70a998ed \
+    --hash=sha256:6669c1bf34080161ce49c589cc512ef24d4c704ac9d2b2d3667f519c60418378 \
+    --hash=sha256:672d207103e6b16ca098611b0f9efad6bc00afd47c03d6ef62186495ca677dc0 \
+    --hash=sha256:6768d67d1bce64270e0fdc2e69309d68b9b18ae56ddf6c711d168e9d051c2cac \
+    --hash=sha256:6a45c3d514f2436064db00d7fc8778d888f0236ebfed649b53d13a59e69ad51b \
+    --hash=sha256:6bd9e1788e15bfcf6a9082de42e30387e7b85d211ab21e57a939bb8cfaaf8d96 \
+    --hash=sha256:6d2a9efe686f9de00d0d1ea32a4a5a86d558a2277501bd78d964214eab625e59 \
+    --hash=sha256:6da83a088f8ef93b2d483a8232a4dbf4d69d3d8496b568a03c56becac43e1808 \
+    --hash=sha256:7018d4af1cd272e847aa5917983ab5e83e4f6579f9dbfecd4a79c0ca80b144c2 \
+    --hash=sha256:71f88e749ea29f67f21f3b36433c1dc54c7729ed2a6d9e2da2e0d9e0d7b224eb \
+    --hash=sha256:737c9c3981998eba27f11786f84fddcbabc74068b72a4a1f454ea02094b57b65 \
+    --hash=sha256:73e77980c7207854f00fc4e71fb1626868d5740ab4012623d55c7a99ad122a72 \
+    --hash=sha256:799c39bdf5e2f1292fedd3009f7b3c9e760f10b2420cb9638d56920840ff6db8 \
+    --hash=sha256:7a83aa6e4805df46fed18e989d3d16f86ef60cb50bbc8d9ce3a6be89165fbf6e \
+    --hash=sha256:7d3391b2188d18737cb2fa147028b1096236eaa7e156446c650a489fa2cadc91 \
+    --hash=sha256:7e1636da3d8dfc220b6dd10264db5f2b165e4888c4518594898fbe381049af8a \
+    --hash=sha256:805c8b84534fa10891890f0e4be39f3a99e94615d93e8836bf9fa1fdca2feeb2 \
+    --hash=sha256:811d02d5122171c1941357efd8f9bf4ffe907b7f0a1a4e729a880e4be3f46e3e \
+    --hash=sha256:8138eb83940ec7299024d92d4dee45f601b9e6c5ffde9d25f4e35e326203c707 \
+    --hash=sha256:83b3944fea42a8400edf92fd1770fb8d0d4f7de651353bd2d8525a92dba69a21 \
+    --hash=sha256:849dd2bb0e5e4ab2b71c7191726a4a8d5aa8a610daa584728cbee0b710ddc4ef \
+    --hash=sha256:8698d70a8081ee8c090dbb394768b5789a1da8b131b5499f89d071dd3cfaf6be \
+    --hash=sha256:8781a792a070cf2bd1b86d3aa943894115faaba6e88122a7bf32d62072742453 \
+    --hash=sha256:88d59b473bfb03259722600839af9bbd7fa13a2eb514beefeedb95997882f69a \
+    --hash=sha256:8909c2f1c6dd65e054ac4b573a91c8384d1492281e55d82d159d653f7a13adf6 \
+    --hash=sha256:8965520ac587c94a4ac48b729be3d8b8de00af39699b17585dfb599babe77977 \
+    --hash=sha256:8b5d563170ff8ba3181caa967c99a3c804d1dedb702c7cb93a6a7c32247da978 \
+    --hash=sha256:8e124f974786f831d6043728e38296969d3579db8896fe004682f5758e613581 \
+    --hash=sha256:8f0fac8b13d14bb06c68195f849371924ae53dd7b1c00fed24650f704383b692 \
+    --hash=sha256:9240187afb63d2f9ddc3e032c670356fe941f6e20662ea168a5dc3f1f317e1b3 \
+    --hash=sha256:925f929d6b59a8b3f8b8c6ac363cd0af7eecc81efb3071770b3c6717c450a369 \
+    --hash=sha256:9348cbb300d224fe3b89793262cb093504d4ae927004468463f745188a193e4a \
+    --hash=sha256:9388003072b95f2f1e3fd908604194d653ba21330d811961a78b7da1a77e9e36 \
+    --hash=sha256:9438a2648b2195980cb2dd8e53ed7b8df91319e2d0b70ae61a9e1d1bc8d3bec9 \
+    --hash=sha256:94e4c421742086aeee4c32a506eec8859d7634aad943f7e6aacf70f813478768 \
+    --hash=sha256:94f5407f7bc64fa6463906b896f9904beeeb7dd8dc116ee8e9056c8714ff9916 \
+    --hash=sha256:971a3bbb75d97ae4e2e8f7d4834236f86f85f0c85e04ab2e191db1123b04f80b \
+    --hash=sha256:9e227f3dbe6bde7491cf0a9965d00b88c6b1a4a95d11480ddf88bb96d397c19f \
+    --hash=sha256:9e25feb9e330b63edb0278a0acdf85e50d0cb0fbf49c3084abbe4e24ae195346 \
+    --hash=sha256:9f098115c247e11d138ab83a28fa0323c77015007ea2df73ba5fd714dfefd67c \
+    --hash=sha256:a18f38cafc329bac5e3c2b96c765b4c96d3d103421ed22ab7988c1e3fce27464 \
+    --hash=sha256:a4bbd2d87dd233b9fc5812160c3d0ffbe42edc22a26ce0469f58479ede633fe9 \
+    --hash=sha256:a5fcffb37e602b0b3c1638a97746b9b96125caa9bcf6fa41d337a9261de231ee \
+    --hash=sha256:a8e9f292fcda89b324f2f5c91d13f1424a153e40fc2756f38ee23b15835ff300 \
+    --hash=sha256:a9f54054101545a9a9cccefddf54316aa6e4491611fcbef9e91b3b6bebec04f6 \
+    --hash=sha256:aa2c838cc024642cc04c6854232f32b43e5e22833dd11119c1766c7873b8370d \
+    --hash=sha256:ac0c7c9f1609b0c4c114feb1d7a3409564c7fb77e360bed9e97e5d25dfeaf868 \
+    --hash=sha256:add96447a86d205ab616665d53b2950ee81083757f56e6ea833c8b2917646b46 \
+    --hash=sha256:ae9dcb8fbe244cb82f8a6458b455b927a03685e383d9bacf1ea5ce180b96dc97 \
+    --hash=sha256:b4a635a0487774f841cb1fb62e907e7195cc95bc761e053184b8acc3ceb20733 \
+    --hash=sha256:b4d12837e0203bbace818ff4a7461afdcd78bcd782351cea148139180d7bcffe \
+    --hash=sha256:b61687d0828e72bf5cda24a2690188f37170bd31c9359ac97e4e66569f120a16 \
+    --hash=sha256:b807e598953730f82e4eae3bd30f6a122cf6b31c398c6b504c0e04c13c170429 \
+    --hash=sha256:b8cd1f918b26fd7b1832ece557cc18f2d8747309ff8b3f0ef9d4250c5ad67a39 \
+    --hash=sha256:b91cc9d336957239ff200f30097e6fea2dc6d6fb3c81e853eaa09eac904fd894 \
+    --hash=sha256:bd3ce56ae2cbae3ba82b683bc425cd7e48d2ed8b10f3e818186b6f5646d9271c \
+    --hash=sha256:be6cb0c799abb0e2ba3e618e6d28ddddf7e485f6c2ce938dfa237daf3905072c \
+    --hash=sha256:befb4158af32106b9a93db8d6d1d1cbbd418c0d5aca0cabb7b1780abf0c89169 \
+    --hash=sha256:bf053da3c97a4bc5ecfbb218cdd2983febd91c617be8367d139882aa11e490aa \
+    --hash=sha256:c02e8f18bdedba082cef725942ac823b9b60656db07f7e265cb31618dfd00d77 \
+    --hash=sha256:c1bc67752d5f21013cfe430df4062441714eab79f65a6a05e01505957e9c35fe \
+    --hash=sha256:c61750fadcd119d0825bcb7d7d675dd264dcc89cc05292aab5be68ebdbb374ad \
+    --hash=sha256:c90d5b3d4e944e065a301d741b3c1d784f6bd1f503aa68b4967e32b2ba313d85 \
+    --hash=sha256:c9a7f43c0b202b334cc9184af09bb8f21d3a209e038efaf106936fb69e6b026e \
+    --hash=sha256:cb96e6e088d6cf71c1ea977510948320234824cf226e32f6f6e044f7a9c82b34 \
+    --hash=sha256:cf63c214fe879a65e69a386f915e36104fc84254ab141240f8854602d8e0be2a \
+    --hash=sha256:d1aca03ede943eb80ab3d63bb082c84b7aab85ea83bd0fd0c200260945fb49d9 \
+    --hash=sha256:d2e56fd3b00222722abfb3f5f0759ddbae4b90811b5ad4343c64030ad1bde70c \
+    --hash=sha256:d5f93ebbeb8032d47e349328ec8662d973d9b05a70b3c35df1f91fe419b84749 \
+    --hash=sha256:d882a373d8093c2941e01291b7ced96e9cbe4781da9a7751ca7e6c70385e5214 \
+    --hash=sha256:d920abdfa61279ba1a2ef9484aab07bf03331f8c08a10120fa332353d06e6932 \
+    --hash=sha256:da2af0d7aebfc2074080d72efa6ab8317c62481ef1f896f65d9999c1c01f4494 \
+    --hash=sha256:dd8ea6ebee7aedbf7c749fa80521d9ccf1ba473e0d1e14805caafbaad281c889 \
+    --hash=sha256:de8b364c423ef0a4bad9069657d617f9a5d2b2062457a89b1fa16ee199c399c1 \
+    --hash=sha256:df1ae86ff54725a01fa1a0510b914ca53a161b7050be74f6204e24aded5971d0 \
+    --hash=sha256:dff05cb7016dff1e9fd68f4122c127b65dfc59de5306cfb7ad92f956f230bee2 \
+    --hash=sha256:e1a622f13970d81f95d0c72f9dc090dce9085fccfa4c9f2174377ee32bd15786 \
+    --hash=sha256:e49fb0d1ce92cfa0cb198cc5b1b11cdf9d0638658e2a2db2687e39db7c87fc78 \
+    --hash=sha256:e5c802729725bd07e2bc3ab7b76dc7e0bbfc53129d8f1eb1c002c24cf774717e \
+    --hash=sha256:e841068dc0be4cb6dfb5c890eb88cbdcff2f4a332393c7ec94e8e618bd32c1a8 \
+    --hash=sha256:e916035e3e9930cbdfdd10abf48861340221857f45509565898e012263f7b289 \
+    --hash=sha256:eba154571c16e032112afac0dc2dfe9e63c2ceb7aedd07bb7eecf2ce26d4dd4c \
+    --hash=sha256:f03460ff076f70ab595bb45a0205ccea1971443575b6920c52e755dec2b3fbfe \
+    --hash=sha256:f0ec3b750b59375eab5b0fb2b9254810c00a3375be6d789899f1055a1d556237 \
+    --hash=sha256:f291bcf42ae98eb5107edb162c3c998b4a89648fd8e99ed4cbd12705292788cd \
+    --hash=sha256:f61efe1d2fe0de16158a5fe1d1cf3c14bdb6aecd54d8938fd26512c525c1f624 \
+    --hash=sha256:f68edfc67aabac33708941f26f22a7b8e9f81429bc0cf249fcf7d66b23af8d19 \
+    --hash=sha256:fa95848c929b6a75f6848d3c9793e59db365ee436776e57db835cdbfa79ba977 \
+    --hash=sha256:fd9f8797427910198f95bced71ddfed61130d7e349213bfb8466c9c99e2c46a8 \
+    --hash=sha256:fdb4ca07ab75ffadab4a8b135ad59cdbb3156b99310f3d565370da74a15d6bd3
+    # via jinja2
Author	SHA1	Message	Date
Niels Lohmann	d1235cecc9	Pin CI pip installs by hash and update KaTeX/DOMPurify The pip installs in check_amalgamation.yml, ubuntu.yml (Meson) and semgrep.yml now use --require-hashes with requirements files compiled by pip-compile --generate-hashes (Python 3.14, as on ubuntu-26.04). The astyle and generate_natvis requirements move to requirements.in, and Meson and Semgrep get theirs in .github/requirements, which Dependabot now watches. check_amalgamation.yml installs the natvis requirements from the PR tree, as it already does for astyle. The Mermaid checker overrides KaTeX to 0.18.11 (GHSA-238p-pmpm-9mq7; Mermaid 11 still requires ^0.16.47) and updates DOMPurify to 3.4.17 (GHSA-cg2g-xh64-v4q7, GHSA-mrv2-3pqr-j5jr, GHSA-c6gr-qvv4-5v3f). Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 07:20:44 +02:00
dependabot[bot]	e28506dd86	Bump jsdom from 30.1.2 to 30.1.3 in /docs/mkdocs/scripts/mermaid (#5898 ) Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 06:54:23 +02:00
dependabot[bot]	03a0595c1e	Bump actions/upload-artifact from 7.0.1 to 7.0.2 (#5897 ) Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 06:54:07 +02:00
dependabot[bot]	4e72af9d54	Bump step-security/harden-runner from 2.21.1 to 2.22.1 (#5896 ) Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 06:53:50 +02:00
dependabot[bot]	0f5d07310c	Bump the codeql-action group with 4 updates (#5895 ) Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 06:53:32 +02:00
Niels Lohmann	0b924f5375	Split and print JSON pointers in one pass (#5887 ) Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 06:53:15 +02:00
Niels Lohmann	6fa14d8f95	Write binary formats through reserved sink space and number runs (#5885 ) Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 06:52:58 +02:00
Niels Lohmann	dec0e9b749	Write strings, scalars and indentation in dump() with fewer calls (#5883 ) Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 06:52:41 +02:00
Niels Lohmann	86d5c56c54	Speed up flatten(), unflatten(), diff() and patch() further (#5886 ) Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 06:48:12 +02:00
Niels Lohmann	aa0b371f7c	Faster std::istream input: inline get-area path, binary reads, unbuffered streambufs (#5884 ) Signed-off-by: Niels Lohmann <mail@nlohmann.me>	2026-10-23 06:48:00 +02:00