mirror of
https://github.com/HDFGroup/hdf5.git
synced 2026-09-25 04:09:44 +03:00
Scot Breitenfeld
7dd5182d91
Verified against the fork's actual GitHub Actions CI (not just local builds) and found two real regressions plus applied formatting the CI bot detected but couldn't push: - test/CMakeTests.cmake: register test_filters_v3.h5 (the Critical #3 golden file) in HDF5_REFERENCE_TEST_FILES so a fresh checkout's build tree actually stages it -- it only passed locally because the file had been generated directly inside an existing local build directory. Was failing H5TEST-tfilter2 on every CI job running the standard suite. - tools/lib/CMakeLists.txt: link the tools library against LINK_LIBS (libm), matching src/CMakeLists.txt's existing pattern. The new hex-float code in h5tools_dump.c is the first tools-library code to need floor()/isfinite(), and FreeBSD/OpenBSD's stricter linkers reject the resulting undefined symbols (Linux tolerated it silently). Was breaking the link step on FreeBSD 14.3/15.0 and OpenBSD 7.9. - Applied the clang-format fixes the CI bot found but couldn't push back (pre-existing fork credentials gap, confirmed unrelated to this branch): tfilter.cpp, h5pDCPLImp.c, H5Znbit.c, tfilter2.c, h5tools_dump.c. Verified via ctest -R H5TEST-tfilter2 on a build tree with the stale golden-file copy removed (reproducing CI's exact test invocation), plus a full local rebuild (0 warnings/errors) and the Fortran/parallel suites. Details and CI failure logs analyzed in REVIEW-6153.md's new "Post-push CI verification" section.