Re: failure to build due to ignoring fwrite() result

[Eric Blake]

On 08/30/2010 01:52 PM, Jim Meyering wrote:
> fwrite is an exception in that its value really is ignorable,
> if you know what you're doing.  Hence, the warn_unused_result
> attribute on its declaration is the problem, not gcc, and not
> the ignore-value module.

Agreed.  Do we have an open bugzilla against glibc for this yet?

> The solution here could be to make gnulib provide a better declaration of
> fwrite: one without the offending attribute.  Deprecating ignore-value or
> somehow disabling all warn_unused_result attributes would be overkill --
> no reason to throw the baby out with the bathwater.

In fact, I think that gnulib could do something like:

static inline int
rpl_fwrite(...)
{
  return fwrite(...);
}
#define fwrite rpl_fwrite

to work around broken glibc headers.

In fact, I'm leaning towards the idea of doing that workaround only if 
the close-out (or close-in) modules are in use, so that you still get 
the warning if you aren't using the preferred gnulib idiom for making 
ignored fwrites safe.  That is, the gnulib idiom is that you add:

atexit(close_stdout);

to main(), as the guarantee that all possible code paths will safely 
detect fwrite() and fflush() failures via ferror(), so that nowhere else 
in the program needs to bother with the return values of intermediate 
steps on FILE*.

--
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org