stillwater-sc / universal / 26593562331

84%

master: 84%

Build Type

push

github

Committed by web-flow

Commit Message

fix(cfloat): isnormal() returns false for +-0 (IEEE / std::isnormal) (#1031)

cfloat::isnormal() returned true for +-0 -- the zero-encoding early-out returned
true ("filter out the one special case"), but zero is not a normal number
(std::isnormal(0)==false). Subnormal/inf/nan were already classified correctly;
zero was the only wrong category. One-line fix: the zero early-out returns false.

Caller sweep (no adaptation needed): every cfloat-internal use of isnormal() in
conversion/normalization (cfloat_impl.hpp 455, 512, 2128, 2189, 2276, 2338) sits
in an else-branch after an earlier iszero() check, so zero never reaches the
isnormal() test there. elreal block.is_normalised() guards with is_zero_block()
first; complex isnormal short-circuits `re==0.0 ||`; closure_plot_png excludes
zero explicitly. Verified: cfloat-to-cfloat and double->cfloat conversion tests
pass unchanged on gcc + clang.

Tests:
- classify.cpp: added VerifyIsNormalClassification asserting +-0/inf/nan/subnormal
  are not normal and finite non-zero normals are (the suite was print-only before).
- fractional.cpp: removed the #1027 isnormal()-and-not-iszero workaround now that
  isnormal() is correct for zero.

Resolves #1030

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Coverage Stats

1 of 1 new or added line in 1 file covered. (100.0%)

18 existing lines in 4 files now uncovered.

46924 of 55859 relevant lines covered (84.0%)

5791451.9 hits per line