#503204 gcc-4.1: erroneously emits warning on C99-required #pragma STDC FENV_ACCESS

Package:
gcc-defaults
Source:
gcc-defaults
Submitter:
Matthew Vernon
Date:
2010-05-30 17:12:16 UTC
Severity:
important
#503204#5
Date:
2008-10-23 13:55:12 UTC
From:
To:
Hi,

C99 requires that code that tests floating-point state flags (or frobs
other bits of the floating-point environment) calls #pragma STDC
FENV_ACCESS ON:

"If part of an application tests floating-point status flags, sets
floating-point control modes, or runs under non-default mode settings,
but was translated with the state for the FENV_ACCESS pragma off, the
behavior is undefined." (7.6.1.2)

Yet, if you do this in gcc with -Wall turned on, you get a warning
message:

matthew@b144-mcv1-mlt:/tmp$ cat test.c
#include <stdio.h>
#include <fenv.h>

int main(void)
{
  #pragma STDC FENV_ACCESS ON

  if(feclearexcept(FE_ALL_EXCEPT))
    printf("feclearexcept failed\n");
  else printf("feclearexcept succeeded\n");

  return 0;
}
matthew@b144-mcv1-mlt:/tmp$ gcc -std=gnu99 -Wall test.c -lm
test.c: In function ‘main’:
test.c:6: warning: ignoring #pragma STDC FENV_ACCESS
matthew@b144-mcv1-mlt:/tmp$ ./a.out
feclearexcept succeeded
matthew@b144-mcv1-mlt:/tmp$

It's obviously buggy to emit a warning on standards-mandated behaviour!
I know you can turn this warning off with -Wno-unknown-pragmas, but
that's not something you want to be doing in production code.

Regards,

Matthew
#503204#10
Date:
2008-10-23 17:33:15 UTC
From:
To:
gcc does not stat full C99 compilance.

Please show the part of the standard. A compiler may produce warnings
without reason.

Bastian
#503204#15
Date:
2008-10-24 09:49:50 UTC
From:
To:
Hi,
Bastian Blank wrote:

Was the section I quoted in my original bug report not sufficient?
Paragraph 7.6.1.2 of C99 says [again]:

"If part of a program tests floating-point status flags, sets
floating-point control modes, or runs under non-default mode settings,
but was translated with the state for the FENV_ACCESS pragma "off", *the
behaviour is undefined*." [emphasis mine]

These are things that many correct programs will want to do. For
example, SuSv3 has this to say about pow():

"An application wishing to check for error situations should set errno
to zero and call feclearexcept(FE_ALL_EXCEPT) before calling these
functions. On return, if errno is non-zero or fetestexcept(FE_INVALID |
FE_DIVBYZERO | FE_OVERFLOW | FE_UNDERFLOW) is non-zero, an error has
occurred."[1]

Gcc as-is, will encourage (quite strongly, since may people will compile
with -Wall, +/- -Werror) users to write code that calls feclearexcept()
and suchlike without using #pragma STDC FENV_ACCESS ON first. This is
undefined behaviour, according to C99! It seems clear to me that this is
a bug in gcc.

Regards,

Matthew

[1] http://www.opengroup.org/onlinepubs/009695399/functions/pow.html
#503204#20
Date:
2010-05-30 17:09:14 UTC
From:
To:
forwarded 503204 http://gcc.gnu.org/PR20785
tag 503204 + upstream
reassign 503204 gcc-defaults
thanks

reassigning to gcc-defaults, same in 4.5 and trunk.
#503204#31
Date:
2010-05-30 17:09:14 UTC
From:
To:
forwarded 503204 http://gcc.gnu.org/PR20785
tag 503204 + upstream
reassign 503204 gcc-defaults
thanks

reassigning to gcc-defaults, same in 4.5 and trunk.