[fzlzjerry]

Fixes #1341.

OurReader::decodeNumber() subtracts a Value::UInt last digit from a LargestInt expression. With JSON_NO_INT64, both have the same width, so the usual arithmetic conversions make the subtraction unsigned. For example, -1 becomes a uintValue containing 4294967295; even -0 gets the wrong stored type.

Keep the last digit signed. The existing quotient/remainder calculation still handles the minimum signed value without negating it directly.

Making the affected configuration testable

Current master first fails to compile with JSON_NO_INT64 on GCC/Clang because the private integerToDouble(Json::UInt64) overload refers to a disabled type. The patch guards that overload, along with 64-bit-only test helpers and API checks. The existing no-int64 floating-point assertions are updated to the current 17-digit formatting.

The reader/writer fixtures that contain 64-bit limits still run. The test driver reads the executable's existing --json-config output and selects explicit floating-point expectations for those three inputs when integer storage is limited to 32 bits. All other expectations, and all default-configuration expectations, stay unchanged; no fixture is skipped. Invalid configuration output fails the test run.

Added regression tests cover negative zero, digit boundaries, signed limits, scalar/array/object contexts, unsigned limits, and the existing out-of-range floating-point fallback. A separate Linux CMake job exercises the complete no-int64 CTest suite so the configuration stays covered. It pins checkout to a full commit, limits permissions to read-only contents, and disables persisted checkout credentials.

There are no public-header changes. This does not add sign-only-token validation; that is covered separately by #1703. The verified reproduction is a Linux build with the library and consumer both compiled with JSON_NO_INT64, rather than a reproduction of the original MSVC 2015 environment.

Validation

• Pristine master: the default build passes all 3 CTests; the no-int64 build fails at the unguarded overload. With only that prerequisite guard applied, all 8 negative CharReader consumer cases fail while the 8 legacy Reader controls pass. Both readers pass all 16 cases after the fix.
• GCC 14 and Clang 19 × C++11/17/20 × default/no-int64: all 12 configurations pass all 3 CTests with the repository's normal warning settings. Both integer configurations pass all 133 C++ unit tests.
• Meson 1.5.1, Clang 19, C++11, shared/static libraries and werror=true: all 3 test suites pass in both configurations.
• Full no-int64 CTest run passes with AddressSanitizer, UndefinedBehaviorSanitizer and LeakSanitizer enabled.
• 21 configuration/real-CLI scenarios check configuration errors, expected-file selection, missing variants, and deliberately corrupted expectations.
• C++11 amalgamated builds and installed-library consumers built with Clang C++17/C++23 pass in both configurations. The default Linux dynamic export set is unchanged (1,905 symbols).
• clang-format 18 over src, include and example, actionlint, and git diff --check pass.

An additional warnings-as-errors stress run encounters pre-existing GCC C++20 implicit-this capture and Clang C++17/20 signed-size warnings. Those failures were reproduced on pristine master; the unrelated source lines are unchanged.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge; no actionable issues remain.

Summary

Fixes negative integer parsing with JSON_NO_INT64 by keeping the final subtraction signed.

• Negative JSON integers keep their signed value in 32-bit storage builds.
• The test suite now builds and runs in 32-bit integer-storage mode.

_{Reviews (4) · Last reviewed commit: "Merge branch 'master' into fix/1341-no-i..." · Reviewed by Greptile}

[coveralls]

Coverage Report for CI Build 40354144714

Coverage remained the same at 90.132%

Details

• Coverage remained the same as the base build.
• Patch coverage: 1 of 1 lines across 1 file are fully covered (100%).
• No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	2744
Covered Lines:	2626
Line Coverage:	95.7%
Relevant Branches:	2647
Covered Branches:	2233
Branch Coverage:	84.36%
Branches in Coverage %:	Yes
Coverage Strength:	23818.57 hits per line

---

💛 - Coveralls