mirror of
https://github.com/nlohmann/json.git
synced 2026-11-08 16:11:15 +00:00
2739c0b0afed840c92b9cba13ebdc61b322dcf9c
5265
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Niels Lohmann
|
35802e78d6 |
Remove dead Makefile targets; document macro_builder; tidy serve_header (#5735)
* Remove dead doctest help entry and pretty_format target from Makefile The top-level Makefile still carried three leftovers: - The help text listed a "doctest" target that was removed in #4560, so "make doctest" fails with "No rule to make target". The example check now runs as "make check_output -C docs". - "pretty_format" ran clang-format on all sources, but .clang-format was deleted in #4573, so the target reformatted everything in the default LLVM style, against the Artistic Style formatting that "make pretty" applies and CI enforces. - "clean" removed benchmarks/files/numbers/*.json, a directory that no longer exists since the benchmarks moved to tests/benchmarks (#3462). Only maintainer tooling changes; the library is not affected. Part of #5717 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Document tools/macro_builder and tidy up serve_header.py tools/macro_builder generates the NLOHMANN_JSON_EXPAND, NLOHMANN_JSON_GET_MACRO and NLOHMANN_JSON_PASTE* macros in macro_scope.hpp, but nothing referred to it. Add a README that explains what it generates, how to run it and where the output goes, and which dependent tables (NLOHMANN_JSON_DOUBLE_PASTE, NLOHMANN_JSON_TYPE_BODY) are maintained by hand. Point to it from a comment above NLOHMANN_JSON_EXPAND. The generator itself is unchanged; following the README reproduces the header byte for byte. In serve_header.py, drop the LGTM suppression (LGTM.com shut down in 2022), replace the """.""" placeholder docstrings with real ones, and import socket and ssl at module level. DualStackServer.server_bind uses socket, which was only imported under __main__; when the module was imported instead, the NameError was swallowed and IPV6_V6ONLY was not cleared. The header change is a comment only; behavior, API and ABI are unchanged. Part of #5717 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Hash every release_files artifact, not a hardcoded subset The `release` target signed and copied json_fwd.hpp into release_files alongside json.hpp, but the shasum line that writes hashes.txt only listed json.hpp, include.zip and json.tar.xz. Users could not verify the published json_fwd.hpp against hashes.txt. Hash every file in release_files except the .asc signatures instead of naming files by hand, so a newly shipped header (such as the json_literals.hpp that #5610 adds to this target) cannot be missed again. Only affects the generated hashes.txt release artifact; the library itself is unaffected. Overlaps #5610, which touches the same lines to add json_literals.hpp to the release target. #5717 item 1 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Remove the broken fuzz_testing* Makefile targets fuzz_testing and fuzz_testing_{bon8,bson,cbor,msgpack,ubjson} seeded fuzz-testing/testcases from tests/data, which was removed in |
||
|
Niels Lohmann
|
791cd88dfc |
Fix raw-TeX formulas and other documentation infrastructure debt (#5736)
* Fix math formulas rendering as raw TeX in the published docs The privacy plugin self-hosts MathJax 2.7.0 but drops its ?config=TeX-MML-AM_CHTML query string, so the rehosted script loads no input jax and the 10 formulas across 6 pages render as raw TeX to readers. Remove pymdownx.arithmatex and the MathJax extra_javascript entry, and rewrite the formulas in plain HTML (<sup>, <i>) instead. This also drops a nine-year-old third-party script from every page. Part of #5718 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix publish_documentation triggers and persist unneeded git credentials publish_documentation.yml only triggered on docs/mkdocs/** pushes, but the site also embeds .github/CODE_OF_CONDUCT.md, CONTRIBUTING.md, SECURITY.md, cmake/{clang,gcc}_flags.cmake, .clang-tidy, tools/astyle/.astylerc and tests/fmt_formatter/project/main.cpp via pymdownx.snippets, so changes to those files never republished the site. Extend the path filter to cover them, and switch runs-on from the long-pinned ubuntu-22.04 to ubuntu-latest to match ci_test_documentation. Also add persist-credentials: false to the checkouts in ci_icpx and ci_nvhpc (ubuntu.yml) and msvc-vs2026/msvc-arm64 (windows.yml), none of which pushes with git, matching every other checkout in these workflows. Part of #5718 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix example build: broken debug echo, deprecations hidden for everything docs/Makefile's debug echo used a space instead of a comma in $(call cxx_standard ...), so it always printed an empty standard. Every example was also compiled with -Wno-deprecated-declarations, which would silently hide an accidental deprecated-API call in any of them. Factor the duplicated compile flags into EXAMPLE_CPPFLAGS/ EXAMPLE_WARNFLAGS, build with -Werror=deprecated-declarations by default, and only allow the three examples that intentionally document deprecated API (the DEPRECATED_EXAMPLES list) to suppress it. Part of #5718 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix check_structure.py NOLINT parsing and an off-by-one report line `line.strip("<!-- NOLINT")` followed by `.strip(" -->")` strips any of the characters in those sets from both ends, not a literal prefix; it only happened to work for "Examples". A NOLINT'd section name starting with N, O, L, I or T (e.g. "Notes", "Template parameters", "Iterator invalidation", "Literals") was silently mangled, so the suppression did not apply and the checker could report a spurious missing/misordered section. Parse the comment with a regex instead. The same fragile strip() pattern was used for heading text; replace it with a plain prefix slice. Also fix the admonition_title report, which used the 0-based line index while every other report in the file uses lineno+1. Part of #5718 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix docset README's non-existent make target and stale fallback URL The README told readers to run `make nlohmann_json.docset`, but the Makefile's targets are `all`, `JSON_for_Modern_C++.docset` and `install_docset_zeal`; the documented command has failed with "No rule to make target" since #2967 (2021). Point the README at the real target and folder name. Info.plist's DashDocSetFallbackURL also still pointed at the old nlohmann.github.io/json/ URL instead of the canonical https://json.nlohmann.me/ from mkdocs.yml's site_url. Leave list_missing_pages/list_removed_paths alone: they may become redundant once #5638's check_docset() lands, which is a follow-up. Part of #5718 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Remove two leftover Doxygen-era .link files from the examples directory parse__iterator_pair.link and parse__pointers.link each held only a Wandbox "online" permalink from the old Doxygen docs. #3071 deleted every other .link file in 2021; these two came in through a parallel PR (#3100) and were never referenced by any page, script or config. Part of #5718 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix MacPorts CMake example to include its own snippet files The MacPorts "Example: CMake" block included integration/homebrew/example.cpp and integration/homebrew/CMakeLists.txt instead of the MacPorts files right next to it, a copy-paste slip from the Homebrew section. Nothing referenced integration/macports/CMakeLists.txt as a result. The page rendered correctly only because the homebrew, macports and vcpkg/CMakeLists.txt snippets are byte-identical, so a future edit to the MacPorts files would not have shown up on the page. Part of #5718 item 6 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Do not persist git credentials in publish_documentation's checkout The checkout step in publish_documentation.yml left the default persist-credentials: true, so GITHUB_TOKEN stayed writable in .git/config for the rest of the job (zizmor's artipacked finding). The Deploy documentation step authenticates through its own github_token input to peaceiris/actions-gh-pages and does not push with the checked-out credentials, so persist-credentials: false is safe here, matching every other checkout in the workflow set. Overlaps #5638, which edits this same checkout step (adds fetch-depth: 0); expect a rebase conflict there. Part of #5718 item 2 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Replace list_missing_pages/list_removed_paths with a comm(1)-based diff The docset Makefile's list_missing_pages ran one sqlite3 query per mkdocs page, and list_removed_paths nested a loop over all mkdocs pages inside a loop over all docset index paths (O(n*m) shell iteration). Issue #5718 item 5 suggested removing or reducing these targets once #5638's check_docset() lands, but that PR is still open and covers only API pages and macros, not the full page set these targets check. Replace the loops with two sorted path lists (DOCSET_PAGE_PATHS from mkdocs' markdown sources, DOCSET_INDEX_PATHS from the built docset index) compared with a single comm(1) call each, verified to produce output identical to the old loops against the current docSet.dsidx. The sed expression used '#' as its delimiter, which GNU Make reads as a comment character even inside a variable assignment, truncating the line and orphaning the closing paren of $(shell ...) ("unterminated call to function 'shell': missing ')'"). Use '@' as the delimiter instead. Part of #5718 item 5 Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
7d7055ec50 |
Fix stack overflow converting deep values between specializations (#5723)
* Fix stack overflow converting deep values between specializations Constructing a basic_json from another specialization (json to ordered_json or back, also via get<ordered_json>()) converted every container with its range constructor, which calls the converting constructor for each element. The call stack therefore grew with every nesting level, and a value nested some 30,000 levels deep overflowed it. The conversion now bounds its descent the way the copy constructor does since #5387: the first 128 levels are converted exactly as before, and below that convert_iteratively() finishes the value with an explicit stack. It builds each container bottom-up from its converted elements with the container's range constructor, so member order and keys that become equal are handled as before, and it gives a value its type only once its container exists, so an exception leaves nothing behind that cannot be destroyed. Parents (JSON_DIAGNOSTICS) and positions (JSON_DIAGNOSTIC_POSITIONS) are set for every value. Converting a null value no longer resets its positions: the constructor assigned null to a value that already was null, which swapped in the positions of the temporary. Fixes #5650. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Explain why converting null keeps positions and why next is a reference Review feedback on #5723 (gregmarr): clarify in comments that the converting constructor has already copied the positions of val, which the null case keeps like every other case, and that next must be a reference into pending so that ++next advances the stored iterator. Comments only; no code change. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Refer to recursion_depth_limit() in the convert_structured() docs The comment still named nesting_depth_limit, which #5637 removed on develop in favor of detail::recursion_depth_limit(). Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Advance the pending iterator through pending.back() and shorten the null comment Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
3b6ae43c53 |
Add JSON_DISABLE_TUPLE_REFERENCE_CONVERSION to fix std::tuple conversions (#5598)
* Add JSON_DISABLE_TUPLE_REFERENCE_CONVERSION to fix std::tuple conversions basic_json can be constructed from std::tuple<json&>, which it turns into a one-element array. Because of this, std::tuple picks its converting constructor that converts the whole source tuple instead of the element-wise one. As a result, std::tuple<const json&> built from std::forward_as_tuple(j) binds to a temporary (a compile error with libc++, a dangling reference with other standard libraries), and std::tuple<json> built the same way holds [j] instead of a copy of j. The new opt-in macro JSON_DISABLE_TUPLE_REFERENCE_CONVERSION (CMake option JSON_DisableTupleReferenceConversion) removes the conversion from a one-element tuple holding a reference to the same basic_json type, so std::tuple converts element-wise. It is off by default, so existing behavior is unchanged. It does not change any function body and therefore is not part of the ABI tag. Fixes #2226 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Convert one-element tuples to arrays on every compiler to_json for std::tuple assigns a braced list, j = { std::get<Idx>(t)... }. With a single element that is itself a basic_json, Apple clang 15 and 16 treat j = {x} as a copy of x, so std::tuple<json>{true} became true instead of [true]. The macOS jobs (Xcode 15.1, 16.1) failed the new checks in unit-disable-tuple-reference-conversion and unit-regression2. The one-element overload that already handles JSON_BRACE_INIT_COPY_SEMANTICS builds the array (or object, for a [string, value] element) explicitly, the same way the initializer-list constructor does. Use it unconditionally. The output is unchanged on compilers that already wrapped the element. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Skip json reference tuple tests on clang < 4 and GCC < 5 ci_test_compilers_gcc_old (4.8) and ci_test_compilers_clang (3.4) could not compile the new tuple tests. Creating a std::tuple of basic_json references, e.g. std::forward_as_tuple(j), makes these compilers instantiate basic_json's conversion operator for libstdc++'s internal tuple bases, which fails hard. This happens with and without JSON_DISABLE_TUPLE_REFERENCE_CONVERSION, so it is a limitation of these compilers, not of the new option. Tested with the CI images: clang 3.4 to 3.9 and GCC 4.8 and 4.9 fail, clang 4, 5, and 6 and GCC 5 and 6 compile all cases. Skip only the checks that create such tuples; the is_constructible checks still run. Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
e96e2982a5 |
Fix CI: useless cast in the growth of the view's node index
GCC -Werror=useless-cast (ci_test_gcc on Linux x86-64) rejected static_cast<std::size_t>(guess + (guess / 4) + 64): the sum is a std::uint64_t prvalue, the same type as std::size_t there, while the cast is needed where std::size_t is 32 bits wide. Cast a named variable instead, which GCC does not report. The build stopped at an earlier error before, so the previous CI run did not show this one. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
0a64e6c99b |
Fix CI: MSVC C4127 in the view builder and the single-header test build
- msvc (Win32, /W4 /WX) reported C4127 (conditional expression is constant) for `TrailingCommas && cur() == ']'` and the like when the option is off. Route the template arguments through a static enabled() function, as json.hpp's nesting_depth_exhausted() does. - ci_test_single_header compiled unit-json_view_builder.cpp against single_include/, which does not contain the internal nlohmann/detail/view headers. Build that test only with the multiple headers. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
d8b8c2498f |
Test the C++11 stand-in for std::string_view of json_view
Six members of string_ref (length, begin, end, operator[], operator!=, and operator<<) were not reached before C++17, where string_ref is std::string_view. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
2fb66ba859 |
Remove the eight-digit case of the view's digit parser
Since whole blocks of eight digits are read directly, parse_upto8() only gets fewer than eight digits; its eight-digit case was dead code. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
ab49a7b1bf |
Address the cpplint findings of the view's parser
The exponent of the overflow check is an std::int64_t instead of a long (runtime/int). Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
154f240022 |
Read whole blocks of eight digits of json_view directly
A block of eight digits of a number token lies inside the input (the digits were counted while scanning, or are recorded in the digit layout), so parse_upto19() reads it without the bounds check of the last, partial block. Traversing canada.json: -11% instructions, -6% cycles. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
8d3280de1e |
Classify a NUL inside a string of json_view as a control character
json::parse ends the input at a NUL only between values (where it does at all); inside a string, a NUL is a control character that must be escaped. The view reported it as a missing closing quote. (Only the error code differed: the exception comes from the library parser.) Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
118af5015e |
Test the overflow check of json_view at the largest double
Numbers of more than 19 digits at the boundary of the largest double are decided by the exact comparison with the midpoint in the overflow check. The check uses the floating-point type of the document: with float, the view rejects what parse() rejects (1e39, 3.4028236e38, the midpoint between the largest float and 2^128), and double documents are not affected. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
9b0b4ff26d |
Address the clang-tidy findings of the view's parser
- tables as std::array; the frames of the first 64 levels stay a C array (not initialized on purpose, NOLINT) - \u escapes are decoded with the library's hex_codepoint() instead of a second table - the parse failure is private, with an accessor; the special member functions of the builder are all declared - no nested conditional operators; explicit parentheses; a repeated branch body merged; auto for casts - the test's C arrays, fixed seed, and escaped literals are marked, as in the other tests Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
694bfd1b4e |
Keep the arguments of the view's throw helpers used without exceptions
With JSON_NOEXCEPTION, NLOHMANN_VIEW_THROW(e) was std::abort() alone, so the parameters of the functions that build the exceptions were unused, a warning that the builds with -Werror turn into an error. The exception is now evaluated before std::abort(); the program ends anyway. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
ad82821c89 |
Add the one-pass parser of json_view (internal)
The builder parses JSON text in one pass into the node index: strings and numbers stay in the source (escaped strings are decoded into an arena), integers are converted while their digits are in cache, and floats keep their digit layout for a later conversion. It accepts exactly what json::parse accepts, for every combination of comments and trailing commas, with and without a terminating NUL, and with JSON_STRICT_NUL_HANDLING. Parse state lives in a local cursor whose address never escapes, so that it stays in registers; out-of-line helpers (errors, regrowth, escapes, comments) are members of the builder and get the positions they need. The value dispatch is expanded once for array elements and once for member values. Literals are compared with memcmp and words read in a fixed byte order, so nothing depends on the platform's byte order. Error messages come with the public classes. Tests (unit-json_view_builder.cpp): accept/reject and values against json::parse for handwritten, generated, and damaged documents under all option combinations, from std::string and from exact-size buffers (no read past the input under AddressSanitizer), deep nesting up to 100,000 levels, NUL/BOM/whitespace cases, and the test-suite files. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
1bf0b1b6c2 |
Add the node index and scanning primitives of json_view (internal)
Internal parts of the zero-copy view (#5295), under detail/view and not included by json.hpp, so that users of json.hpp compile nothing of it: - macro_scope.hpp/macro_unscope.hpp: the few macros the view needs, under its own prefix (json.hpp undefines its own at its end); the throw macro honors JSON_NOEXCEPTION and JSON_THROW_USER like JSON_THROW - string_ref.hpp: std::string_view from C++17 on, else a small stand-in - node.hpp: the 16-byte node of the index; its kinds are value_t values (checked by a static_assert) - document_data.hpp: the storage of a parsed document (node array, decode arena, owned input) - scan.hpp: string and digit scanning with unrolled checks at fixed offsets (after yyjson) and the library's SWAR and UTF-8 checks, independent of the byte order Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
b88e5f9107 |
Keep the behavior-changing configuration readable after json.hpp
json.hpp undefines JSON_STRICT_NUL_HANDLING and JSON_USE_LEGACY_DISCARDED_VALUE_COMPARISON at its end. Code that builds on the library after it, such as the planned json_view.hpp, reads them from detail::abi_config instead. The constants live in the ABI namespace, which already encodes both settings, so they always match the basic_json in use. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
d23803fd32 |
Decode \u escapes with a table in the lexer
get_codepoint() read the four hex digits of a \u escape with four calls to get(), each classified by a chain of range comparisons. For contiguous input, get_codepoint_bulk() now decodes them with one lookup per byte (hex_codepoint() in string_scan.hpp, after yyjson's read_hex_u16): a 256-entry table maps a byte to its value, or 0xFF for anything else, and an invalid digit shows in the OR of the four values. It then skips the four bytes and updates the position counters as four get() calls would. If a digit is invalid or fewer than four bytes are left, it changes nothing and the existing loop runs, so errors are reported with the same message and position as before. json::parse, best of 5 runs in separate processes (M1 Max): the escaped twitter.json (every non-ASCII character as \u) -13.6%, all other files within 0.3%. Tests compare the contiguous and the streaming path (value or exception message) for valid escapes, surrogate pairs, truncated and invalid digits at every position, and 3,000 seeded random escapes. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
f1014c938a |
Fix CI: useless casts to std::size_t in the string-scan tests
GCC -Werror=useless-cast rejected static_cast<std::size_t>(next() % n): on 64-bit Linux std::uint64_t and std::size_t are the same type, while the cast is needed where std::size_t is 32 bits wide. Draw the sizes from a 32-bit value instead, which converts to std::size_t implicitly on every platform. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
e91fdad877 |
Find the stop byte of a string run without a byte loop
find_string_special() and find_ascii_copyable_run() test eight bytes at a time, but located the stopping byte inside a word with a byte loop. The lowest flagged byte of the SWAR tests is always a true hit (the borrows of the subtractions can only flag bytes above one), so its index is now the trailing-zero count of the mask; words are read in little-endian order on every platform, so this does not depend on the byte order. scalar_string_bulk_run() validates a run of multi-byte UTF-8 sequences one after another instead of searching for the next special byte in between, which helps text in non-Latin scripts. The kernels serve the lexer's contiguous fast path, the serializer, and the binary formats. New tests compare all three with byte-by-byte reference scans on 100,000 generated buffers at three alignments; the portable fallback of count_trailing_zeros() was checked against the builtin. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
9c71689715 |
Convert long doubles under a multi-byte decimal point completely
The strtold fallback, which is left only for long double formats that are not binary64 (x87, binary128), substituted the first byte of the locale's decimal point for '.'. Under a locale whose decimal point is longer than one byte, such as fa_IR.UTF-8 or ar_EG.UTF-8 (U+066B), strtold stopped there and the value was truncated at the decimal point. A longer decimal point is now put into a copy of the token. The test "locale with a multi-byte decimal point" now compares the long double values with those of the "C" locale; with x87 long doubles it failed before. Fixes #5660. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
44ec53c77b |
Convert float and double with the library's own correctly rounded parser
float, double, and long double where it is IEEE-754 binary64 (MSVC, Apple arm64) are now converted by the library itself, correctly rounded and independent of the locale and of the C and C++ libraries: - The token is split into sign, significand w (at most 19 digits), and decimal exponent q, using the positions of the decimal point and the exponent that the scanners already recorded, so no character is classified again. - Clinger's fast path where w and 10^|q| are exact. - Eisel-Lemire otherwise, now templated for binary32 and binary64. - For tokens with more than 19 digits whose w and w + 1 round differently, an exact big-integer comparison with the midpoint between the two candidates (the digit comparison of fast_float, simplified). This replaces the separate token walks of Clinger's fast path and of Eisel-Lemire, the significant-digit gate that avoided the former, and, for float and double, std::from_chars and the locale-aware strtod. std::from_chars and strtold remain only for other long double formats (x87, binary128, double-double) and for types that are not IEEE-754. Values are bit-identical to before wherever the previous conversion was correctly rounded; tokens converted in a locale with a multi-byte decimal point are now also exact. Overflow still gives out_of_range.406, underflow a signed zero. convert_float() is the entry point for other parsers of JSON text: it converts like the lexer, without allocation for binary32/binary64. Tests: exact-bit tests for double and float (ties, subnormal and overflow boundaries, huge exponents, more digits than any midpoint), Eisel-Lemire for binary32, the round trips of 200,000 doubles and 100,000 floats without declines, 508 generated hard cases with the expected bits of both formats (float_hard_cases.hpp) through the converter and both scanners, and JSON-level overflow/underflow checks for double and float. The locale tests now check the values in a locale with a multi-byte decimal point. Docs: the statements that parsing uses strtod/strtof/strtold; the fast_float credit now names the digit comparison. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
6a073dbae4 |
Give operator>> a strong exception-safety guarantee (#5695)
operator>> parsed directly into its basic_json& target, so a parse error left the target holding whatever was parsed before the error instead of its previous value. With JSON_DIAGNOSTICS=1, that partial value also violated the class invariant, because the parent pointers of an array or object's elements are only set when the container is closed, which a failed parse never reaches; copying such a value then aborted in assert_invariant(). Fix it the way basic_json::parse() already handles this: parse into a temporary and move it into the target only once parsing succeeds, so the target is left unchanged if an exception is thrown. Fixes #5652. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
fdcc569eee |
Use only documented StringType members in json_pointer (#5692)
contains(const json_pointer&) and operator/=(std::size_t) (and hence
operator/(std::size_t)) used string_t operations that the StringType
template parameter documentation explicitly does not require:
comparing string_t with a const char* literal, c_str(), and
constructibility from std::string. This made both functions fail to
compile for a conforming custom StringType, even though the
documentation's own reference StringType satisfies the requirements.
Fix contains() to compare individual chars ('0'..'9') instead of
comparing string_t with const char* literals, and to call data()
(documented to be null-terminated) instead of c_str(). Fix
operator/=(std::size_t) to build the array-index token via the
existing detail::to_string<StringType> helper (ADL int_to_string() or
assignment from std::to_string()) instead of via std::to_string()
directly, matching how diff(), items(), and std::hash already convert
a std::size_t to a StringType.
Add regression tests to tests/src/unit-alt-string.cpp: contains() for
present/missing keys and indices, "-", a leading zero, and a
non-numeric token on an array, plus json_pointer::operator/(std::size_t).
Fixes #5666.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
|
||
|
Niels Lohmann
|
6fc0d501f3 |
Move the user-defined string literals to <nlohmann/json_literals.hpp> and add JSON_NO_AUTOMATIC_UDLS (#5610)
* Add JSON_NO_UDLS to leave out the user-defined string literals The bodies of operator""_json and operator""_json_pointer call the parser, so every translation unit including the library instantiates it, even if it never parses anything. Defining JSON_NO_UDLS leaves the literals out entirely, which saves 15-35% compile time for such translation units (#5294). Nothing changes if the macro is not defined. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Mention JSON_NO_UDLS in the list of exported module symbols Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Move the user-defined string literals to <nlohmann/json_literals.hpp> Following the review in #5294, the literals now live in their own header instead of being removed entirely: <nlohmann/json.hpp> includes it at the end unless JSON_NO_AUTOMATIC_UDLS (renamed from JSON_NO_UDLS) is defined, so a project can opt out globally and include the header only where the literals are used. The header only uses public and standard macros, because the library's internal macros are undefined at the end of json.hpp and the amalgamation inlines macro_scope.hpp only once. For the same reason, the library no longer defines and undefines JSON_USE_GLOBAL_UDLS, so a user's definition is still visible to the header. The single-header copy is identical to the multi-header one, as it only includes <nlohmann/json.hpp>. The module always exports the literals. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix CI: include cycle, GCC 4.8 literal operator spacing, and global UDLs off in the JSON_NO_AUTOMATIC_UDLS test - Suppress clang-tidy misc-header-include-cycle on the intentional mutual include of json.hpp and json_literals.hpp. - Use operator"" _json with a space for GCC 4.8 in the test's detection aliases, as the header does. - Only test the global literal operators when JSON_USE_GLOBAL_UDLS is on. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Declare the literal operators through a local macro The GCC 4.8 spacing condition was repeated for both operator definitions and the global using-declarations. NLOHMANN_JSON_LITERAL_OPERATOR(suffix) now selects operator""##suffix or operator"" suffix in one place and is undefined at the end of json_literals.hpp. Suggested by gregmarr in review. Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
67435c9c7e |
Give a deep copy its type only after its container exists (#5721)
When copying a value nested deeper than 128 levels, and an allocation fails while an inner array or object is being copied, the partially built copy ended up with an element typed array/object but holding a null pointer. That element was already a fully constructed member of its parent's container, so destroying the parent during stack unwinding dereferenced the null pointer (release builds) or failed assert_invariant() (debug builds), instead of letting std::bad_alloc reach the caller. copy_iteratively() set a pending worklist element's type right after popping it, before the next loop iteration created its container in copy_array_level()/copy_object_level(). Move that type assignment into those two functions, right after the container is successfully created, and drop the premature one in copy_iteratively(), so a half-built element stays a null value - as copy_shallow()'s comment already promised - until it can safely hold one. Add a regression test to tests/src/unit-allocator.cpp that copies a value nested 130 levels deep (both arrays and objects, with a std::map- and an ordered_map-backed object_t) and fails every allocation of the copy in turn: each attempt must throw std::bad_alloc without crashing, and the source must stay unchanged. Fixes #5640. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
2ea6d8c127 |
Require the found key to equal the looked-up key when comparing objects (#5720)
For an object type whose comparator treats unequal keys as equivalent (for example a std::map with a case-insensitive comparator), compare_iteratively() looked up a mismatched left key in the right object with find(), which uses the object's own comparator, and accepted whatever entry it found without checking that the keys are actually equal. A case-insensitive comparator then found "KEY" for "key", so two objects nested past the recursion bound (or at every depth with JSON_NO_THREAD_LOCAL) could compare equal even though the object type's own operator== - and basic_json itself, below the bound - consider them different. Accept the found entry only if its key equals (not just compares equivalent to) the looked-up key. Fixes #5655. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
66877675b1 |
Check the iterator range for binary values in basic_json(first, last) (#5719)
basic_json(first, last) treated value_t::binary like the structured types (array, object) in the range check, so it always copied the whole binary value regardless of the iterators, even for an empty range such as (b.end(), b.end()). The other primitive types (number, boolean, string) already reject such a range with invalid_iterator.204, and erase(first, last) already does the same for binary values, so this made the constructor inconsistent with both. Move case value_t::binary into the group of checked primitive types. Also update the two matching passages in basic_json.md that describe overload 7, and add a version-history note. Fixes #5670. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
7fd6895788 |
Hide a discarded container's content from the parser callback (#5706)
When a parser callback rejects an object's or array's start event,
json_sax_dom_callback_parser kept calling it for everything inside
that container anyway: nested keys, values, and the start/end events
of containers below it. This contradicts parser_callback_t's own
documentation, which promises that discarding a container at its
start event also hides its content from the callback.
The same code path also kept a full copy of every key inside such a
discarded container in key_stack until the whole parse finished,
because the early return for values that are not stored skipped the
matching pop. Filtering out a large subtree is the main reason to use
a callback, so this made peak memory during the parse scale with the
size of the very subtree the callback was trying to skip.
Fix start_object(), start_array(), and key() so that a container
whose own start event was discarded, or that is nested inside one, is
never handed to the callback, and no longer pushes onto the key
stacks. A container whose start event was accepted but whose key was
rejected still gets its content reported, as documented ("the
callback is still called for the associated value, but its return
value has no further effect"); only its own bookkeeping is skipped
since it will not be stored.
Fixes #5643.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
|
||
|
Niels Lohmann
|
6ae17630a4 |
Reject integral keys for contains(), find(), and count() at compile time (#5705)
j.contains(0), j.find(0), and j.count(0) used to compile: the literal 0 is a null pointer constant, so it converts to a null const char*, and the overloads taking const typename object_t::key_type& accepted it by constructing a std::string from that null pointer, which is undefined behavior (a crash with both libc++ and libstdc++). value(0, default_value) had the same problem in C++11, where the object comparator is not transparent. Add deleted overloads for integral arguments to contains(), find() (const and non-const), count(), and value() so that these calls are compile errors in every supported language mode instead of crashing. Calls with string, string_view, json_pointer, and size-typed element access (at(), operator[](), erase()) are unaffected. Fixes #5657. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
5bd766aa50 |
Move to_bson's binary subtype check into calc_bson_sizes (#5703)
to_bson() rejected a binary value's subtype above 255 (out_of_range.415) in write_bson_binary(), which only has the binary_t, not the basic_json value that holds it, so the exception was created with no JSON_DIAGNOSTICS context even though the equivalent to_msgpack() check names the value's path. The check also ran after the document size, all preceding elements, and this element's header and length had already reached the output adapter, so a caller-provided std::vector or std::string ended up holding a truncated document. calc_bson_sizes() already walks every value before anything is written, to size embedded documents and arrays and to reject invalid keys (out_of_range.409) up front. The subtype check now runs there instead, in calc_bson_binary_size(), which is given the basic_json value so the exception can use it as context. The now-redundant check in write_bson_binary() is removed, since calc_bson_sizes() always throws first if any binary value in the document has an oversized subtype. Fixes #5675. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
4bb1b14b06 |
Trim the compiler-appended NUL from wide/UTF string literals too (#5702)
With JSON_STRICT_NUL_HANDLING defined to 1, parsing a wide, UTF-16, UTF-32, or (C++20) UTF-8 string literal (e.g. json::parse(L"[1]")) failed with parse_error.101 at the terminating NUL of the literal, and accept() returned false. The array overload of input_adapter() only dropped the compiler-added trailing '\0' for arrays of char, so for wchar_t, char16_t, char32_t, and char8_t arrays that terminator was passed to the parser as data, which the macro then rejected. Broaden the trimming to every character type that a string literal can use (char, wchar_t, char16_t, char32_t, and, since C++20, char8_t). Arrays of any other element type (unsigned char, std::uint8_t, ...), as used for CBOR/MessagePack, are unaffected: a trailing zero byte there is still read as genuine data. Fixes #5658. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels LohmannandClaude Sonnet 5
|
bfe0f32d71 |
Fix std::terminate and null pointer access in input_stream_adapter (#5699)
Parsing from a std::istream crashed in two unusual but valid stream states, both in input_stream_adapter: - With eofbit in the stream's exceptions() mask, get_character() sets eofbit via is->clear(), which throws std::ios_base::failure. While that exception unwinds, ~input_stream_adapter() called clear() again to reset eofbit, which is still set and still in the exception mask, so it throws a second time out of the (implicitly noexcept) destructor and std::terminate() is called. The destructor now only calls clear() if a bit other than eofbit remains set, so the first exception can propagate normally. - For an std::istream without a stream buffer (rdbuf() == nullptr, e.g. std::istream(nullptr)), the constructor stored the null pointer without checking it, and get_character() dereferenced it. input_adapter(std::istream&) now throws parse_error.101 for such a stream, the same as it already does for a null FILE* or char*. Added regression tests to unit-deserialization.cpp and, for the JSON_PRECISE_STREAM_POSITION variant of get_character(), to unit-precise-stream-position.cpp; both crashed before this fix. Documented the two exceptions in parse.md and operator_gtgt.md. Fixes #5646. Signed-off-by: Niels Lohmann <mail@nlohmann.me> Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com> |
||
|
Niels Lohmann
|
44a88d85be |
Fix NLOHMANN_JSON_SERIALIZE_ENUM_STRICT's from_json message (#5698)
from_json built its out_of_range.410 message with "..." + j.dump(). If the unmatched value is (or contains) a string with invalid UTF-8, that dump() itself throws type_error.316, so the caller got type_error.316 instead of the documented out_of_range.410; such strings can reach get<Enum>() unvalidated, e.g. from from_cbor()/from_msgpack(). With a custom string_t, j.dump() returns that type, and "const char*" + string_t does not compile unless the type happens to provide operator+, so the macro failed to compile for such types. Build the message with detail::concat(), which appends any type exposing data()/size() and always yields a std::string, and dump with error_handler_t::replace so building the message itself cannot throw. Added regression tests: an invalid-UTF-8 case in the existing strict-enum test in unit-conversions.cpp, and a strict-enum use with alt_string (the custom string_t from unit-alt-string.cpp) to cover the compile failure. Fixes #5667. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
5b11a0282c |
Exchange the CustomBaseClass subobject in basic_json::swap() (#5697)
basic_json::swap() (and the friend swap() and the pre-C++20 std::swap overload that forward to it) only exchanged m_data.m_type/m_data.m_value, leaving each value's json_base_class_t subobject in place. This is inconsistent with the copy and move constructors and copy assignment, which all carry the base class along with the value, so after a.swap(b) any metadata stored in a CustomBaseClass ended up attached to the wrong value. Algorithms that mix swap() with moves, such as std::sort, scrambled the metadata across the whole container. Fix the member swap() to also exchange the json_base_class_t subobject and extend the noexcept specifications of swap() and the friend swap() accordingly. Fixes #5653. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
bfea6f36d3 |
Fix to_msgpack() reading the inactive number union member (#5694)
* Fix to_msgpack() reading the inactive number union member basic_json stores number_integer and number_unsigned in a union, and number_unsigned_t only has to be at least as wide as number_integer_t (with the default types, both are 64-bit and have the same representation). When number_integer_t is narrower, write_msgpack() read the wrong union member in two places: - The number_unsigned case wrote number_integer's bits instead of number_unsigned's, silently writing the wrong value whenever it did not fit in number_integer_t. - The number_integer case (non-negative branch) picked the encoded width by comparing number_unsigned's bits, which is undefined behavior, though the value written was still number_integer's, so at worst a too-wide encoding was chosen. Read the active member in both cases, like the other binary writers (CBOR, UBJSON, BJData, BSON, BON8) already do. Fixes #5644. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Cast number_integer to number_unsigned_t only once in to_msgpack() Addresses review comment by @gregmarr. Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
e444a66276 |
Copy values before inserting an initializer list into an array (#5693)
* Copy values before inserting an initializer list into an array
insert(pos, {...}) inserted wrong values when the initializer list
contained const references to elements of the array being inserted
into. json_ref stores only a pointer for a const lvalue, so the
initializer_list_t range passed straight to the array's range insert
aliased the array's own storage; std::vector::insert(pos, first, last)
may move or shift elements before copying from that range, so the
source elements were already stale by the time they were read
(different wrong results on libc++ and libstdc++).
Copy the referenced values into a temporary array_t first, then move
that temporary into place, so the source range never aliases the
array being modified.
Fixes #5656.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Use the reserve_array helper in the initializer_list insert fix
The previous commit called array_t::reserve() directly on the
temporary buffer used to copy an ilist's values before inserting.
std::deque, a documented ArrayType (tests/src/unit-custom-array-type.cpp),
has no reserve(), so insert(pos, initializer_list) no longer compiled
for it. Use the existing detail::reserve_array() SFINAE helper (already
used by the SAX DOM parser) instead, which leaves array types without
reserve() untouched.
Added a regression check that deque_json::insert(pos, {...}) compiles
and handles the aliasing case from #5656.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
---------
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
|
||
|
Niels Lohmann
|
1151826508 |
Take diff()'s fast path unless the object type reorders members (#5691)
* Take diff()'s fast path unless the object type reorders members For every object type except an insertion-ordered one like ordered_map, diff() no longer produced a member-by-member patch when target had a key that sorts before a key the two objects share: it fell through to the slow path, which removes every member of source and re-adds every member of target, instead of just adding the new key. #5465 added an order check to require the fast path to also reproduce target's member order, needed because ordered_map's patch()-driven "add" appends a new member at the end. The check compared the common keys' order between source and target and also required that every added key come after every common key in target's order ("new_keys_form_suffix"). The comment above it argued this check is always true for std::map, and that reasoning is correct for the order of the common keys themselves, but not for new_keys_form_suffix: a std::map iterates in sorted key order, so a new key that sorts before an existing common key is enumerated between common keys, making new_keys_form_suffix false even though std::map's own key order does not need reordering at all - it places every member itself, regardless of insertion history, so a member-by-member diff already reproduces target's iteration order. Only require the order check for an object type that keeps insertion order, using the same detail::is_ordered_map trait the library already uses to recognize such an object type in set_parent(). Every other object type - std::map in key order, a hash map in an order its operator== ignores - always takes the fast path. Added a regression test to unit-json_patch.cpp: the issue's example now yields a single "add" op for json, while ordered_json still takes the slow path to reproduce target's member order. Fixes #5639. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Only track the target key order in diff() for insertion-ordered objects common_keys_target_order and new_keys_form_suffix are only read when object_t keeps its members in insertion order; skip building them otherwise. Addresses review comment by @gregmarr. Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
6218aa212b |
Throw std::length_error for operator[](SIZE_MAX) instead of corrupting the array (#5687)
For idx == SIZE_MAX, the non-const array operator[] computed the new size as idx + 1, which wraps to 0. resize(0) then emptied the array, and the subsequent operator[](idx) on the now-empty vector wrote one element before its buffer. Every other too-large index (e.g. SIZE_MAX - 1) already went through resize(), which throws std::length_error and leaves the array unchanged; SIZE_MAX was the one value for which the overflow bypassed that safety net. Add a guard that throws std::length_error before computing idx + 1 when idx is the largest representable size_type value, so the array is left unchanged, matching the exception vector::resize() already throws for smaller (but still too large) indices. Fixes #5647. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
42f89e3130 |
Let to_json(std::optional<T>) propagate exceptions from T's to_json (#5684)
* Let to_json(std::optional<T>) propagate exceptions from T's to_json The overload was marked noexcept even though its body assigns *opt to the JSON value, which calls T's to_json (or allocates for std::string, std::vector, or json). Any exception from there -- a user-defined to_json reporting an error, or std::bad_alloc -- called std::terminate() instead of propagating. The noexcept also made basic_json's converting constructor noexcept(true) for std::optional<T>, so json j = opt; could not report the error either. Fixes #5642. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix CI: make to_json(std::optional<T>) conditionally noexcept GCC's -Wnoexcept (an error in ci_test_gcc) fired at to_json_fn's noexcept(noexcept(to_json(j, val))): after dropping the unconditional noexcept, to_json(std::optional<int>) had no exception specification although GCC could prove its body cannot throw. It also made json(std::optional<int>) lose its noexcept. Declare the overload noexcept exactly when assigning the contained value to the JSON value is (std::is_nothrow_assignable<BasicJsonType&, const T&>), which is what the body does. std::optional<int> is noexcept again; a T whose to_json may throw still propagates the exception. Static assertions in the test check both cases. The test's throwing to_json triggered -Wmissing-prototypes and -Wmissing-noreturn (clang) and -Wmissing-declarations and -Wsuggest-attribute=noreturn (GCC). Move the type and its to_json into an anonymous namespace, mark the function [[noreturn]], and compile them only without JSON_NOEXCEPTION, like the test that uses them. Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
6d7845d207 |
Fix deprecated json_pointer/string operator== warning in value() (#5683)
value(KeyType&&, default) is constrained on is_comparable_with_object_key, which passes KeyType as a reference. is_comparable's dispatch on is_json_pointer_of<A, B> only matches a json_pointer as a plain type or a plain reference, so a const-qualified reference (as produced when KeyType is deduced from a json_pointer argument) fell through to is_comparable_no_json_pointer, which instantiates the deprecated json_pointer/string comparison operators. This made ordered_json's transparent comparator (and any transparent comparator on a custom string type) warn under -Wdeprecated-declarations when calling value(json_pointer, default), even though no such comparison is ever performed. at() was already fixed for this in #5289, which does not use is_comparable_with_object_key. Strip references and cv-qualifiers with uncvref_t before the is_json_pointer_of dispatch, so any reference-to-json_pointer is recognized regardless of qualifiers. Fixes #5664. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
678fd3017b |
Fix element path for map/unordered_map JSON_DIAGNOSTICS errors (#5681)
When converting a JSON array to std::map or std::unordered_map with a non-string key, each element must itself be a [key, value] array. If an element is not an array, from_json() threw type_error 302 with the outer array's value (&j) as the exception context, so with JSON_DIAGNOSTICS enabled the message pointed at the whole array instead of the offending element (e.g. "(/outer/m)" instead of "(/outer/m/2)"), even though the message text already described the element's type. Both from_json() overloads now pass the element (&p) as the context, so the reported JSON Pointer matches the type named in the message, the same way std::vector<std::vector<T>> and similar conversions already do. Fixes #5668. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
fc4c9c3446 |
Fix clear() to also reset the subtype of a binary value (#5680)
* Fix clear() to also reset the subtype of a binary value clear() on a binary value cleared the bytes but left the subtype untouched, so the result was not equal to a default-constructed binary value even though the documentation says clear() has the same effect as *this = basic_json(type()). The fix calls byte_container_with_subtype::clear_subtype() alongside the existing clear() call. Extended the "filled binary" clear() test in unit-modifiers.cpp with a case that uses a subtype, since the existing cases only covered binary values without one. Fixes #5669. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix the table alignment in clear.md Addresses review comment by @gregmarr. Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
68beba727c |
Fix from_json() for enums with underlying type bool (#5679)
get_arithmetic_value() rejects boolean_t, so the default from_json()
for enums failed to compile for an enum whose underlying type is
bool (e.g. enum class Flag : bool { off, on }), even though the
matching to_json() serializes such enums as an unsigned number.
Read the underlying value through number_unsigned_t in that case,
matching what to_json() writes, then cast back to the underlying
type before constructing the enum.
Fixes #5671.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
|
||
|
Niels Lohmann
|
0050d0f7a4 |
Share one nesting depth limit between all bounded descents (#5637)
Copying and comparing stopped their descent at basic_json::nesting_depth_limit(), while serializing, hashing and merging used detail::recursion_depth_limit(). Both were 128, but nothing kept them equal. The thread-local count now tests against detail::recursion_depth_limit() as well, and a static_assert keeps the limit small enough for the byte that holds the count. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
6a8a7735ed |
Do not throw in contains() for an empty array reference token (#5614)
* Do not throw in contains() for an empty array reference token json_pointer::contains() rejected malformed array indices, but an empty reference token (e.g. "/a/" where "a" is an array, or "/" on an array) passed every check and reached array_index(), which throws out_of_range.404. contains() must not throw (cf. #5395), so it now returns false for an empty token. at() still throws out_of_range.404. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix CI: bind j_nested_const by reference in the json_pointer test clang-tidy (ci_clang_tidy) flagged the new test with performance-unnecessary-copy-initialization: the local copy j_nested_const of j_nested is never modified. Bind it as a const reference instead; it still exercises the const overloads of at() and contains(). Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
65af260117 |
Move instead of deep-copy ordered_json values when an object grows (#5609)
* Move instead of deep-copy ordered_json values when an object grows ordered_map keeps its elements in a std::vector<std::pair<const Key, T>>. With a std::string key, that pair is not nothrow move constructible (the const key has to be copied), so std::vector copies every element when it reallocates. For ordered_json, this deep-copies every member value an object already holds, including whole nested subtrees, on each growth step. Grow the storage in ordered_map instead, copying the keys and moving the values. This happens in two phases, so the strong exception guarantee is kept without try/catch. The first phase may throw, but only touches a temporary buffer: it copies the keys, value-initializes the values, and constructs the new element. The second phase moves the values (noexcept) and swaps the buffers. Because the new element is constructed before any value is moved, arguments that refer to elements of the container stay valid, as with std::vector. Types that cannot take this path keep the std::vector behavior. Parsing into ordered_json (ParseStringOrdered, Apple M1 Max, clang -O3): twitter 3.20 -> 1.70 ms, citm_catalog 7.73 -> 3.67 ms, jeopardy 219 -> 177 ms, canada unchanged. The number of allocations for twitter and citm_catalog drops by two thirds. Also add ParseStringOrdered rows to the benchmarks, and document the growth behavior and the exception safety of ordered_map. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix CI: skip std::pair noexcept assumptions on EDG-based compilers ci_icpc and ci_nvhpc failed to compile unit-ordered_map.cpp: the static assertion that std::pair<const std::string, ordered_json> is not nothrow move-constructible fails there. The EDG front end (Intel icpc 2021.10, NVIDIA nvc++ 25.5) considers the defaulted move constructor of std::pair<const Key, T> noexcept even if copying Key can throw. With these compilers, std::vector already moves such elements itself when it grows, and ordered_map correctly leaves growing to it. The same misjudgement makes std::vector call std::terminate when a key copy throws during growth, so the exception-safety test with throwing_key would abort on these compilers as well. Skip the static assertion and the exception-safety section when __EDG__ is defined. Verified with icpc 2021.10 (-std=gnu++11) and nvc++ 25.5 (C++11 and C++17) on Compiler Explorer: unit-ordered_map and unit-disabled_exceptions build and pass. Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
d268eaa693 |
Convert floats with Eisel-Lemire when std::from_chars is unavailable (#5617)
* Convert floats with Eisel-Lemire when std::from_chars is unavailable Float tokens that Clinger's fast path cannot convert (e.g. the 17-digit coordinates of canada.json) went to strtod unless std::from_chars was available. It is not used in C++11/14, and not with libc++, which does not define __cpp_lib_to_chars. The Eisel-Lemire algorithm (after fast_float's compute_float) now converts them with integer arithmetic, correctly rounded for any token with at most 19 significant digits. Longer tokens are truncated; the result is used if w and w + 1 round alike, else strtod decides as before. Overflow still yields infinity (out_of_range.406). The table of powers of five (fast_float's) lives in pow5_table.hpp; a unit test recomputes every entry with big-integer arithmetic. Further tests: known values generated with Python (whose float() is correctly rounded), 200,000 round trips through to_chars, and the 128-bit multiplication and leading-zero count against big-integer references (both with and without a 128-bit type). Checked against strtod on 6.5 million tokens, among them 60,000 exact halfway cases: no difference. json::parse on canada.json: -8.6% (C++11), -7.6% (C++17, Apple clang); other files unchanged. Compile time of a TU including json.hpp: +0.7%. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix CI: unused parse result and find() == npos in the Eisel-Lemire tests GCC (-Werror=unused-result) rejected CHECK_THROWS_WITH_AS(json::parse(...)) because parse() is [[nodiscard]]; assign the result to a dummy json as the other tests do. clang-tidy flagged longer.find('.') == npos with abseil-string-find-str-contains; store the position in a variable first. Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
5b9ffae8bb |
Move the float conversion chain out of the lexer (#5616)
lexer::convert_number() converted float tokens with std::from_chars (when available), Clinger's fast path, and the locale-aware strtod fallback, all as lexer members. They are now free functions in number_parse.hpp: - convert_float_fast(): std::from_chars, then Clinger's fast path, skipped when the mantissa has too many significant digits - convert_float_locale_aware(): strtof/strtod/strtold with the decimal point of the current locale, retried when the locale changed (#5198) so that other code converting JSON number tokens gets the same values. No change in behavior; the lexer no longer includes <clocale> and <cstdlib>. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |
||
|
Niels Lohmann
|
04f6ddd227 |
Keep external headers as #include when amalgamating (#5615)
A header that builds on json.hpp (such as the planned json_view.hpp) must not inline json.hpp: its single-header version would contain a second copy of the library, and that copy would change with every library change. The optional config key "external" lists include paths that are kept as #include directives. Only the first directive per path is kept; repeated ones are commented out, as the tool already does for inlined headers. config_json_view.json uses it for json_view.hpp; the existing configs do not set it, and json.hpp and json_fwd.hpp regenerate byte-identically. Signed-off-by: Niels Lohmann <mail@nlohmann.me> |