Re: [patch] ld: optimize vfinfo output slightly

[Mike Frysinger <vapier at gentoo dot org>]

On Saturday 31 March 2012 00:59:32 Ian Lance Taylor wrote:
> Mike Frysinger writes:
> > On Friday 30 March 2012 20:30:48 Alan Modra wrote:
> >> On Fri, Mar 30, 2012 at 01:20:24PM -0400, Mike Frysinger wrote:
> >> > the reason i didn't do that is it causes warnings since fwrite() is
> >> > marked as "warn on unused result"
> >> 
> >> (void) fwrite (...);
> > 
> > that doesn't work either
> > 
> > $ cat test.c
> > #include <stdio.h>
> > main(){(void)fwrite("foo", 1, 1, stderr);}
> > $ gcc -O2 -Wall test.c -c
> > test.c:2:8: warning: ignoring return value of âfwriteâ, declared with
> > attribute warn_unused_result [-Wunused-result]
> > 
> > assigning it to a variable but not checking the variable no longer works
> > with newer gcc and -Wunused-but-set-parameter
> 
> The warn_unused_result attribute in gcc was designed for the realloc
> function, where it is truly an error to ignore the result.  It is
> absolutely not an error to ignore the result of fwrite; after all, it's
> perfectly reasonable to use ferror.  It was a serious misunderstanding
> for anybody to attach the warn_unused_result attribute to fwrite.  This
> misunderstanding was made even more serious by somebody deciding that it
> would be a good idea to have a compiler that defaults to defining
> _FORTIFY_SOURCE.

while i agree with not using wur on fwrite, i obviously disagree with fortify-
by-default being a bad thing.  it has caught plenty of real bugs with little 
overhead.
-mike

Attachment: signature.asc
Description: This is a digitally signed message part.