Repository navigation
Fix value(json_pointer, default) aborting under JSON_NOEXCEPTION - #5700
Merged
Merged
Conversation
With exceptions disabled (JSON_NOEXCEPTION or -fno-exceptions),
value(const json_pointer&, default) called std::abort() for array
reference tokens that array_index() rejects with out_of_range.404/410:
indices too large to fit size_type, the empty token ("/"), and tokens
like "/1a". With exceptions enabled, the same tokens correctly yielded
the default value, because get_checked_or_null() relied on
JSON_TRY/JSON_INTERNAL_CATCH (detail::out_of_range&) to turn the
exception into nullptr; under JSON_NOEXCEPTION, JSON_THROW aborts
before that catch is ever reached.
get_checked_or_null() now detects those out-of-range tokens itself,
the same way contains(json_pointer) already does (#5495), and only
calls array_index() for tokens that must still raise parse_error.106
or parse_error.109 (e.g. "/01", "/+1"), matching the documented
behavior of value().
Added regression tests to tests/src/unit-disabled_exceptions.cpp
(built with JSON_NOEXCEPTION and -fno-exceptions) and the matching
checks to tests/src/unit-element_access2.cpp for normal exception
mode.
Fixes #5672.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
12 tasks done
nlohmann
added
the
review needed
nlohmann
added a commit
that referenced
this pull request
Sep 30, 2026
array_index(), contains() and get_checked_or_null() each re-implemented the RFC 6901 array-index rules and the size_type range check: array_index() does the canonical parse and throws; contains() (which must not throw, #5395) re-validates every digit by hand and runs its own strtoull/ERANGE check before calling array_index() anyway, parsing every array token twice; get_checked_or_null() wraps array_index() in JSON_TRY/ JSON_INTERNAL_CATCH (detail::out_of_range&) to turn an unrepresentable index into "not found". Add a single private, noexcept parse_array_index(s, idx) returning an array_index_status (ok / leading_zero / not_a_number / unresolved / exceeds_size_type). array_index() becomes a thin wrapper mapping each status to the existing parse_error.106/109 or out_of_range.404/410; contains() and get_checked_or_null() switch on the status directly. This removes contains()'s digit-validation loop and its second strtoull call, and get_checked_or_null()'s JSON_TRY/JSON_INTERNAL_CATCH. Bugfix as a consequence: get_checked_or_null()'s JSON_TRY/ JSON_INTERNAL_CATCH was dead code under JSON_NOEXCEPTION (JSON_TRY expands to "if(true)" and the catch to "if(false)", so JSON_THROW's std::abort() ran unconditionally), meaning value() and contains() would abort instead of returning the default/false for an out-of-range-sized or oversized array index when exceptions are disabled (#5672). Switching on parse_array_index()'s return value instead of relying on an actual throw/catch fixes this: get_checked_or_null() now returns nullptr for array_index_status::unresolved/exceeds_size_type in every build configuration, and still calls JSON_THROW (aborting under JSON_NOEXCEPTION, as before) only for a malformed index (leading_zero/not_a_number), matching its documented @throw list. All existing error ids, messages and diagnostic paths are unchanged; a few reference tokens that used to fail contains()'s manual per-character validation (e.g. "1a") now fail via array_index_status::unresolved instead, with no observable difference since contains() only returns bool. Adds regression tests to unit-element_access2.cpp's "access on array type" section covering value() with an index that exceeds size_type and one with a trailing non-digit, both of which must yield the default value rather than abort/throw. Public API: no change. Overlaps #5700, #5614 and #5692, which touch the contains() and get_checked_or_null() array hunks; this change replaces those hunks with calls into the new shared parser, so a rebase will need to re-apply their token-handling changes (e.g. the empty-token case) on top of the switch statements here. Signed-off-by: Niels Lohmann <mail@nlohmann.me> #5708 item 4
Conflicts: - tests/src/unit-disabled_exceptions.cpp: kept both new sections (#5672 value(json_pointer) test and develop's ordered_json growth test) Signed-off-by: Niels Lohmann <mail@nlohmann.me>
gregmarr
approved these changes
Oct 4, 2026
nlohmann
added
🚀 ready to merge
nlohmann
added a commit
that referenced
this pull request
Oct 4, 2026
#5728) * Remove unused is_sax and is_detected_convertible detail::is_sax had no user: the parser and the binary reader only use is_sax_static_asserts, so is_sax was a second, unchecked copy of the SAX event list. is_sax_static_asserts asserted boolean(bool) twice in a row, and detail::is_detected_convertible was never used anywhere. Remove all three and include <cstddef> for size_t instead of <cstdint>. Only names in nlohmann::detail are removed; behavior, public API and ABI are unchanged. The diagnostics for an incomplete SAX handler are the same, apart from the duplicated boolean() message. Part of #5708 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Replace meta/logic.hpp with a disjunction trait meta/logic.hpp added a second set of type-level boolean helpers (cxpr_and, cxpr_or, cxpr_not, ...) next to the existing conjunction and negation in type_traits.hpp. It was used only by one static_assert in from_json_tuple_impl, two of its templates were never used, and it was the only header without the license banner and relied on transitive includes for <type_traits>. Add the missing disjunction next to conjunction and negation, use the three in the static_assert, and delete logic.hpp together with its BUILD.bazel entry. same_sign now uses disjunction as well, which resolves the 2022 TODO waiting for such a trait. The static_assert accepts and rejects the same types as before. Only names in nlohmann::detail change; behavior, public API and ABI are unchanged. Part of #5708 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Remove unused would_call_std_* from NLOHMANN_CAN_CALL_STD_FUNC_IMPL Besides detail::result_of_begin/end, which is_range and iterator_t use, the macro defined a namespace detail2 with a tag type, a catch-all overload and would_call_std_begin/end, plus would_call_std_begin/end structs directly in namespace nlohmann. Nothing has used them since they were added in #3020. Reduce the macro to its detail part. Without the trailing struct the ';' after the two invocations would be an empty declaration that -Wextra-semi flags, so drop it. macro_scope.hpp included meta/detected.hpp only for this macro; all users of detected.hpp include it (or type_traits.hpp) themselves, so remove the include. Behavior and ABI are unchanged. The undocumented, untested and unused names nlohmann::would_call_std_begin, nlohmann::would_call_std_end and namespace nlohmann::detail2 are no longer declared. Part of #5708 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Simplify is_ordered_map to reuse has_capacity is_ordered_map re-detected capacity() with a C++03 sizeof/vararg trick right after has_capacity did the same detection through is_detected. For ordered_map, the old trick took the address of std::vector::capacity, which [namespace.std]/6 makes unspecified. Reuse has_capacity instead, which removes the unspecified-behavior pointer-to-std-member and two NOLINT suppressions. Part of #5708 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Remove duplicate const overload of json_pointer::get_checked The const and non-const get_checked() overloads had byte-identical 50-line bodies, differing only in the signature. The remaining template deduces a const-qualified BasicJsonType for const callers, so at(), the out_of_range::create() calls and the bounds check all still work. Part of #5708 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix tautological clause in iter_impl's iterator category assertion The static_assert meant to check the LegacyBidirectionalIterator named requirement had a first clause comparing std::bidirectional_iterator_tag to itself, which is always true and checks nothing; only array_t::iterator was actually being checked, despite the message claiming object iterators were checked too. Drop the tautological clause, reword the message to describe what is actually checked, and note that object_t may use a forward-only iterator as long as reverse iteration and operator-- are unused. The check is intentionally not extended to object_t::iterator, since that would reject object types with forward-only iterators that compile and work correctly today. Part of #5708 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix misplaced and stale comments in JSON_HAS_RANGES and conversions The JSON_HAS_RANGES feature-detection block had its libc++ comment sitting above the clang+libstdc++ branch it does not describe, leaving the libc++ branch uncommented and the clang+libstdc++ branch without its own rationale. Move each comment to sit under its own branch, and give the clang+libstdc++ branch (added in issue 5161) its own one-line reason referencing that issue instead of reusing the libc++ branch's comment. Also fix a duplicated-word typo ("in large in large cpp files") in from_json.hpp, drop two unanswered 2017 design questions left as comments in type_traits.hpp and from_json.hpp that no longer reflect open questions, and correct NLOHMANN_JSON_SERIALIZE_ENUM_STRICT's @SInCE tag from 3.12.0 to 3.13.0, the release it was actually introduced in. Part of #5708 Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Support any-rank C arrays in from_json, not just rank 1-4 from_json() for C arrays had four hand-unrolled overloads (rank 1-4, added incrementally in #4262), each with its own nested loops. to_json() already handles any rank recursively, so a rank-5+ C array could be serialized but not read back with get_to()/get<>(). Replace the four overloads with one from_json() SFINAE-constrained on get<remove_all_extents<T>::type>() existing, forwarding to a pair of mutually recursive from_json_c_array_element() helpers: one assigns a non-array element via get<T>(), the other loops over a array element and recurses one dimension at a time. Each dimension still goes through at(), so type_error.304/out_of_range.401 stay unchanged; ranks 1-4 keep their existing behavior and semantics. Adds rank-5 round-trip and mismatched-shape tests to unit-conversions.cpp. Public API: additive only (rank 5+ C arrays become readable). Signed-off-by: Niels Lohmann <mail@nlohmann.me> #5708 item 1 * Move templated_json_throw into nlohmann::detail templated_json_throw() was defined in macro_scope.hpp, which is included outside NLOHMANN_JSON_NAMESPACE_BEGIN, so the helper leaked into the global namespace as ::templated_json_throw with no ABI tag. Unqualified lookup in NLOHMANN_JSON_SERIALIZE_ENUM_STRICT could then bind to a same-named function declared in the user's own namespace instead, which fails to compile with Clang ("does not name a template"). Move the helper next to the exception classes in exceptions.hpp, inside nlohmann::detail, and call it qualified as ::nlohmann::detail::templated_json_throw<...>(...) from both macro expansion sites. Rewrite the doc comment to give the real reason for the helper (JSON_THROW may expand to code that discards its argument, e.g. when exceptions are disabled) and fix the "supress" typo. templated_json_throw was never released (added by #5151 after v3.12.0), so it can be moved freely. Adds a regression test that expands NLOHMANN_JSON_SERIALIZE_ENUM_STRICT inside a namespace declaring its own templated_json_throw. Public API: no change (::templated_json_throw was an unreleased, unintentional global-namespace leak with no callers relying on its location). Overlaps #5698, which rewrites the same two macro call lines; the overlapping hunks are small and should be trivial to reconcile on rebase. Signed-off-by: Niels Lohmann <mail@nlohmann.me> #5708 item 2 * Factor the repeated JSON_HAS_RANGES/MinGW guard into one macro The std::ranges view conversion (excluded on MinGW because of its incomplete C++20 ranges support, #4916) was gated by the same #if JSON_HAS_RANGES && !defined(__MINGW32__) condition at seven independent sites in to_json.hpp and type_traits.hpp, with the MinGW rationale duplicated in two of them and missing from the rest. Since the sites come in matching pairs (one enables is_compatible_range_view and a view-based overload, the other adds the exclusion to the plain-array-type overload), a drift between any pair would produce an ambiguous or missing overload on exactly one platform. Add JSON_HAS_RANGE_VIEW_CONVERSION next to JSON_HAS_RANGES in macro_scope.hpp, combining both conditions with the #4916 reasoning in one place, #undef it in macro_unscope.hpp, and use it at all seven sites. This does not fold the MinGW check into JSON_HAS_RANGES itself: JSON_HAS_RANGES is user-overridable and also gates the enable_borrowed_range specialization in iteration_proxy.hpp, which is not excluded on MinGW. No behavior or public API change: JSON_HAS_RANGE_VIEW_CONVERSION expands to exactly the condition that was previously written out at each site. Overlaps #5585, #5600 and #3575, which touch the same to_json.hpp and type_traits.hpp lines; the change here is a mechanical search-and-replace of the guard condition and should rebase cleanly. Signed-off-by: Niels Lohmann <mail@nlohmann.me> #5708 item 11 * De-duplicate from_json.hpp's map and array-fallback bodies Several from_json() overload pairs in from_json.hpp were copies of each other, so a fix has to be applied twice (as #5681 already does): - from_json(..., std::map&) and from_json(..., std::unordered_map&) for non-string keys had identical 16-line bodies: array check, m.clear(), pair check loop, m.emplace(...). Route both through a new from_json_pair_array_to_map(j, m) helper. - The from_json_array_impl priority_tag<1> and priority_tag<0> fallbacks ran the same std::transform/std::inserter loop, differing only in ret.reserve(j.size()). Merge them into one body and, modeled on the existing from_json_object_reserve, add a from_json_array_reserve pair so the reserve() call is only made for ConstructibleArrayType that support it. Error ids (type_error.302), messages, diagnostic paths ((at(0)/at(1)) and behavior for types with/without reserve() are unchanged; only the duplication is removed. Public API: no change. Overlaps #5681, which changes the "&j" to "&p" line in both map bodies; the shared helper here should make that a one-line change instead of two on rebase. Signed-off-by: Niels Lohmann <mail@nlohmann.me> #5708 item 5 * Unify json_pointer's three array-index parsers array_index(), contains() and get_checked_or_null() each re-implemented the RFC 6901 array-index rules and the size_type range check: array_index() does the canonical parse and throws; contains() (which must not throw, #5395) re-validates every digit by hand and runs its own strtoull/ERANGE check before calling array_index() anyway, parsing every array token twice; get_checked_or_null() wraps array_index() in JSON_TRY/ JSON_INTERNAL_CATCH (detail::out_of_range&) to turn an unrepresentable index into "not found". Add a single private, noexcept parse_array_index(s, idx) returning an array_index_status (ok / leading_zero / not_a_number / unresolved / exceeds_size_type). array_index() becomes a thin wrapper mapping each status to the existing parse_error.106/109 or out_of_range.404/410; contains() and get_checked_or_null() switch on the status directly. This removes contains()'s digit-validation loop and its second strtoull call, and get_checked_or_null()'s JSON_TRY/JSON_INTERNAL_CATCH. Bugfix as a consequence: get_checked_or_null()'s JSON_TRY/ JSON_INTERNAL_CATCH was dead code under JSON_NOEXCEPTION (JSON_TRY expands to "if(true)" and the catch to "if(false)", so JSON_THROW's std::abort() ran unconditionally), meaning value() and contains() would abort instead of returning the default/false for an out-of-range-sized or oversized array index when exceptions are disabled (#5672). Switching on parse_array_index()'s return value instead of relying on an actual throw/catch fixes this: get_checked_or_null() now returns nullptr for array_index_status::unresolved/exceeds_size_type in every build configuration, and still calls JSON_THROW (aborting under JSON_NOEXCEPTION, as before) only for a malformed index (leading_zero/not_a_number), matching its documented @throw list. All existing error ids, messages and diagnostic paths are unchanged; a few reference tokens that used to fail contains()'s manual per-character validation (e.g. "1a") now fail via array_index_status::unresolved instead, with no observable difference since contains() only returns bool. Adds regression tests to unit-element_access2.cpp's "access on array type" section covering value() with an index that exceeds size_type and one with a trailing non-digit, both of which must yield the default value rather than abort/throw. Public API: no change. Overlaps #5700, #5614 and #5692, which touch the contains() and get_checked_or_null() array hunks; this change replaces those hunks with calls into the new shared parser, so a rebase will need to re-apply their token-handling changes (e.g. the empty-token case) on top of the switch statements here. Signed-off-by: Niels Lohmann <mail@nlohmann.me> #5708 item 4 * Regenerate single_include after merging develop The merge commit kept develop's single_include/nlohmann/json.hpp because make amalgamate saw it as up to date. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Address review: switch in array_index, drop redundant inline - json_pointer::array_index() dispatches on array_index_status with a switch, matching the other parse_array_index() caller - drop `inline` from the function templates this PR adds or moves in from_json.hpp - reword a comment that described the change rather than the code Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me>
nlohmann
added a commit
that referenced
this pull request
Oct 4, 2026
- discard the results of dump() and from_*() in CHECK_THROWS with utils::ignore_return_value (GCC -Werror=unused-result) - give unit-bson's huge_string_t a default constructor (MSVC C2512, GCC 5, clang 3.5) - unit-disabled_exceptions: use the literals namespace when the global UDLs are off (ci_test_noglobaludls; #5700) - unit-binary_utf8_strict: expect the JSON pointer prefix with JSON_DIAGNOSTICS (#5741) - skip the tests that rely on exceptions under JSON_NOEXCEPTION (#5678, #5732) - clang-tidy and clang -Werror: static test data, CAPTURE(...);, const-correctness, use-after-move alias, unused conversion operator, a missing <iterator> include Signed-off-by: Niels Lohmann <mail@nlohmann.me>
nlohmann
added a commit
that referenced
this pull request
Oct 4, 2026
* Keep the serializer conversion for objects whose keys cannot be converted #5591 added a test converting nlohmann::json into a basic_json whose string type cannot be constructed from std::string. That instantiates convert_iteratively(), whose members.emplace_back(next.key(), ...) needs exactly that key conversion, and broke the build of unit-alt-string. Dispatch on the key's constructibility and leave such conversions to the serializers, as the levels above the nesting bound already do (#3425). Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix the remaining CI failures on develop - unit-wstring: with a 16-bit wchar_t (Windows), a lone surrogate is reported as the ill-formed byte 0xFF since #5704; the std::wstring expectations still had the previous <U+0000>. - ci_single_binaries: json_literals.hpp (#5610) and json.hpp include each other on purpose, and IWYU, not following the cycle, asks to replace json.hpp with json_fwd.hpp. Report its findings without failing the build, as already done for json.hpp. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix the library warnings and noexcept specifications from the merged PRs - binary_reader: rename the error_handler constructor parameter, which shadowed the member (-Wshadow, -Wshadow-field-in-constructor; #5746) - basic_json(copy_construct_tag, ...): declare it noexcept when copying the base class is (GCC 16 -Wnoexcept; #5690) - the scalar-on-left legacy comparison operators: noexcept only when converting the scalar is, like their member counterparts (#5682, #5751) - compare_leaves: use std::is_eq/is_lt/is_gt instead of comparing a std::partial_ordering with 0 (-Wzero-as-null-pointer-constant; #5686) - serializer: silence MSVC C4127 for the EnsureAscii template parameter (#5741, #5746) - clang-tidy: return the sanitized reference in binary_writer, take the key of ordered_map::find_impl by const reference (#5727), and mark the switches over parse_array_index (#5728) - ordered_map: keep <memory> for std::allocator (IWYU) Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Split unit-conversions.cpp so MinGW can link it clang 18 with the MinGW linker failed to link test-conversions_cpp17 ("relocation truncated to fit: IMAGE_REL_AMD64_REL32"). As windows.yml recommends, keep the objects small by splitting the test file. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Fix the tests added by the merged PRs for all CI configurations - discard the results of dump() and from_*() in CHECK_THROWS with utils::ignore_return_value (GCC -Werror=unused-result) - give unit-bson's huge_string_t a default constructor (MSVC C2512, GCC 5, clang 3.5) - unit-disabled_exceptions: use the literals namespace when the global UDLs are off (ci_test_noglobaludls; #5700) - unit-binary_utf8_strict: expect the JSON pointer prefix with JSON_DIAGNOSTICS (#5741) - skip the tests that rely on exceptions under JSON_NOEXCEPTION (#5678, #5732) - clang-tidy and clang -Werror: static test data, CAPTURE(...);, const-correctness, use-after-move alias, unused conversion operator, a missing <iterator> include Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Title the macro examples and add JSON_STRICT_BINARY_UTF8 to the docset The documentation style check requires "Example: ..." titles on pages with several examples (#5741, #5591) and a docset entry for every macro page. Signed-off-by: Niels Lohmann <mail@nlohmann.me> * Regenerate BUILD.bazel and nlohmann_json.natvis Signed-off-by: Niels Lohmann <mail@nlohmann.me> #5746 added detail/output/error_handler.hpp and #5741 the json_abi_sbu8 ABI tag. * Install libidn11 for the CMake 3.5.0 binary in ci_cmake_flags Signed-off-by: Niels Lohmann <mail@nlohmann.me> #5733 moved ci_cmake_options from ubuntu:focal to ubuntu:24.04, which no longer ships libidn.so.11; the CMake 3.5.0 release binary links against it, so every ci_cmake_flags run has failed since. Install focal's libidn11 package for that matrix entry only. * Suppress Infer's false STACK_VARIABLE_ADDRESS_ESCAPE in get_impl get_impl() returns its local by value. A test added by the merged PRs instantiates it with a type Infer misreads, so ci_infer reported the 2021 code for the first time. Signed-off-by: Niels Lohmann <mail@nlohmann.me> --------- Signed-off-by: Niels Lohmann <mail@nlohmann.me>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
With exceptions disabled (
JSON_NOEXCEPTIONor-fno-exceptions),value(const json_pointer&, default_value)called
std::abort()for array reference tokens thatarray_index()rejects without_of_range.404/410:indices too large to fit
size_type, the empty token ("/"), and tokens like"/1a". With exceptions enabled,the same tokens correctly returned
default_value, becauseget_checked_or_null()relied onJSON_TRY/JSON_INTERNAL_CATCH (detail::out_of_range&)to turn the exception intonullptr; underJSON_NOEXCEPTION,JSON_THROWaborts before that catch is ever reached.contains(json_pointer)did not havethis problem, since it already checks these tokens before calling
array_index()(#5495).Changes
get_checked_or_null()(include/nlohmann/detail/json_pointer.hpp) now detects, without exceptions, thetokens that
array_index()would reject without_of_range.404/410and returnsnullptrfor them directly.It calls
array_index()only for tokens that must still raiseparse_error.106orparse_error.109(e.g."/01","/+1"), which matches the documented behavior ofvalue().single_include/nlohmann/json.hppviamake amalgamate.Tests
tests/src/unit-disabled_exceptions.cpp(built withJSON_NOEXCEPTIONand-fno-exceptions) covering"/7","/1a","/","/99999999999999999999999", and"/18446744073709551615". Built against pristinedevelopheaders, this test crashes withSIGABRTon thesecond check (
"/1a"); against the patched headers, it passes.tests/src/unit-element_access2.cppso normal-exception mode stays covered too.nlohmann::jsonandnlohmann::ordered_json(where templated) withclang++ -fsanitize=address,undefined -Wall -Wextra -Werror, C++11 and C++17: all pass, ASan/UBSan clean.Public API
No breaking changes. This only fixes an implementation bug so behavior matches what was already documented for
value(const json_pointer&, default_value); no signatures, exceptions, or return types changed.Fixes #5672
This PR was written by Claude Code on behalf of @nlohmann.
🤖 Generated with Claude Code