COMP: Update jsonxx to latest TransitApp/jsonxx

[thewtex]

Update vendored jsonxx from hjiang/jsonxx v1.0.1 (2021) to TransitApp/jsonxx master (eede690, 2025-11-24). TransitApp's fork is the actively maintained continuation with includes cleanup and bug fixes.

Changes

• jsonxx.h / jsonxx.cc: Updated to latest TransitApp/jsonxx (eede690). Includes cleanup: cassert moved out of header, string_view support added, general code improvements.
• MSVC fix: Replaced GCC case range extension ('0'...'9') with individual case labels for MSVC compatibility.
• clang-format: Applied project clang-format 19.1.7 to vendored files.
• Rebased onto current master (v5.4.6 CI, blowekamp's iterator benchmarks, CMake fixes from ENH: Replace --no-merges with --first-parent for comment #76).

Previous review comments

• jhlegarreta's comment about removing 14 from CMake standard comment — already addressed on master via earlier PRs.

[jhlegarreta]

Thanks for doing this Matt.

[thewtex]

I think we need to move to a different JSON library -- the existing one fails to build with MSVC and is not maintained.

[phcerdan]

I think we need to move to a different JSON library -- the existing one fails to build with MSVC and is not maintained.

This one is really famous: https://github.com/nlohmann/json, it seems great to use, but slow performance (not sure about our requirements).

The fastest is simdjson: https://github.com/simdjson/simdjson/#performance-results

[thewtex]

:Ooo:, simdjson looks nice!

I am also finding glaze:

https://github.com/stephenberry/glaze

Claims to be faster than simdjson, but requires C++20. Has a neat reflection API.

[hjmjohnson]

Pushed fix for MSVC linker errors (LNK2019: unresolved external symbol for Object::Object, Array::Array, Value::Value constructors/destructors).

Root cause

TransitApp/jsonxx declares static const Object EmptyObject = {}; (and Array/Value variants) in the header. Each TU that includes the header creates its own copy, calling constructors defined in jsonxx.cc. On Windows with shared libraries, these constructors aren't __declspec(dllexport)-ed from the DLL, so the header test TU gets LNK2019.

Fix: replaced with inline accessor functions returning a local static — C++17-safe, ODR-correct, no cross-TU linker dependency.

[hjmjohnson]

Windows CI still failing — but this is now an upstream action bug, not a jsonxx issue.

The v5.4.6 branch of ITKRemoteModuleBuildTestPackageAction received the Windows path-shortening fix (#128 → #129), but that fix introduced a shell compatibility bug: the "Download ITK" step uses $ITK_BUILD_ROOT (bash syntax) without shell: bash, so on Windows (where GitHub Actions defaults to pwsh) the variable is undefined and ITK gets cloned to the wrong location.

Fix PRs: ITKRemoteModuleBuildTestPackageAction#130 (v5.4.6) and #131 (main). Once merged, CI here should be re-triggered.

Error trace

CMake Error: The source directory "D:/PerformanceB/ITK" does not exist.
ninja: error: loading 'build.ninja': The system cannot find the file specified.

The junction D:\PerformanceB → D:\a\ITKPerformanceBenchmarking is created correctly, but the Download ITK step's cd "$ITK_BUILD_ROOT" is a no-op in PowerShell (needs $env:ITK_BUILD_ROOT). ITK gets cloned inside the repo checkout dir instead of at the junction root.

[hjmjohnson]

Updating to the new vendored code seems to work accross the platforms.

[thewtex]

💚