[RFC] Turning off trapping-math for C and C++ front-ends by default

[Matthias Kretz]

Andrea Pinski via Gcc [Thursday, 2 July 2026, 01:02:42 CEST]:
> So GCC has always defaulted to -ftrapping-math but that has always
> come with many issues. Especially if we go by Joseph's definition of
> the flag (from https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53805#c4
> and https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54192):

At https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90993#c4 I was also calling 
for changing the default.

I believe we should have a diagnostic on calls to the corresponding fenv 
functions when no-trapping-math is in effect. Sketch:
https://compiler-explorer.com/z/rfM79fYGe

Wrt. what the C++ standard says ... not much actually. At this point our 
working model in WG21/SG6 is that C++ hand-waves "there be floats". 
Implementations then implement ISO 60559 in addition to the C++ standard. 
60559 specifies "attributes" that act on "blocks" which can affect this kind 
of behavior:
- compiler flags implement these attributes
- blocks are TUs.

From my experience of the majority of floating-point code out there (casual 
uses rather than floating-point-expert magic), I very much believe that no-
trapping-math is the correct default.

-- 
──────────────────────────────────────────────────────────────────────────
 Dr. Matthias Kretz                           https://mattkretz.github.io
 GSI Helmholtz Center for Heavy Ion Research               https://gsi.de
 std::simd
──────────────────────────────────────────────────────────────────────────