FEniCS / ffcx / 39130325279

88%

main: 89%

push

github

Suffix float literals when the real type is 'float' (#914)

Floating point literals are 'double' in C unless suffixed, so in a
float32 kernel every table entry and every scalar initialiser is a
narrowing conversion. MSVC reports one diagnostic per occurrence:
104127 C4305 and 165 C4244 warnings over a DOLFINx test run.

Emit the 'f' suffix when the kernel's real type is 'float', which covers
float32 and the components of complex64.

The change is diagnostic only: the decimal digits are unchanged, so each
literal rounds to the same value it did as a 'double' initialiser. The
object file compiled from a float32 Poisson kernel is byte-identical
before and after (clang -O2 -g0).

Fixes #715.

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

3 of 3 new or added lines in 1 file covered. (100.0%)

4791 of 5444 relevant lines covered (88.01%)

0.88 hits per line