https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110703
Bug ID: 110703
Summary: Incorrect "-Wfloat-conversion" diagnostic with
_Generic
Product: gcc
Version: 12.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: jepler at unpythonic dot net
Target Milestone: ---
When compiled with `gcc -Wfloat-conversion -c boo.c` the following code results
in a `-Wfloat-conversion` diagnostic across a range of gcc versions including
gcc version 12.2.0 (Debian 12.2.0-14) and godbolt "trunk" as of today
(https://godbolt.org/z/3K7sTTv1x)
```
#include <math.h>
#define tgsin(x) \
_Generic((x), \
double: sin((x)), \
float: sinf((x)) \
)
int f(double d) {
return tgsin(d) > 0;
}
```
The result is
```
$ gcc -Wfloat-conversion -c boo.c
boo.c: In function ‘f’:
boo.c:10:18: warning: conversion from ‘double’ to ‘float’ may change value
[-Wfloat-conversion]
10 | return tgsin(d) > 0;
| ^
boo.c:6:25: note: in definition of macro ‘tgsin’
6 | float: sinf(x) \
| ^
```
As far as I can tell, the diagnostic is spurious, because it applies to the
"float" case of _Generic, which is not the one that is actually chosen. That
is, inspecting the resulting object file `boo.o` there is a call to sin but not
to sinf.
This is prompted by https://github.com/v923z/micropython-ulab/pull/636 and
earlier
https://github.com/micropython/micropython/commit/f31f9a8b70db03cbcbcf39b493f959d0e284962a
-- it at first appeared to be related to -Os size optimization, but that's just
related to how glibc chooses to implement fpclassify, whether via _Generic or
not.
This may be related to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68193