Re: [PATCH] do not ignore errors like ENOSPC, EIO when writing to stdout

[Akim Demaille]

Le 30 janv. 2012 à 11:43, Jim Meyering a écrit :

> Hi Akim,
> 
> So are these soft diagnostics important enough that their loss
> (due to failed write-to-stderr) would merit inducing a nonzero
> exit status?  Hey! I know... exit (0.5) :-)

Don't use literals, prefer (EXIT_SUCCESS + EXIT_FAILURE)/2 :).

> But seriously, I suppose a patch (maybe even a new gnulib module --
> I haven't looked) would be welcome solely on principle.  Probably via
> an atexit-callable function very similar to close_stdout.

Actually close_stdout already does it.  You pushed it :)

commit 7193a969abceafecf878f3efdcf996203e4a4daa
Author: Jim Meyering <address@hidden>
Date:   Fri Oct 6 13:59:10 2006 +0000

    [ChangeLog]
    * modules/closeout (Description): Mention stderr too.
    [lib/ChangeLog]
    * closeout.c (close_stdout): Also close stderr.
    * closeout.h: Update comment.

> /* Close standard output.  On error, issue a diagnostic and _exit
>    with status 'exit_failure'.
> 
>    Also close standard error.  On error, _exit with status 'exit_failure'.
> 
>    Since close_stdout is commonly registered via 'atexit', POSIX
>    and the C standard both say that it should not call 'exit',
>    because the behavior is undefined if 'exit' is called more than
>    once.  So it calls '_exit' instead of 'exit'.  If close_stdout
>    is registered via atexit before other functions are registered,
>    the other functions can act before this _exit is invoked.
> 
>    Applications that use close_stdout should flush any streams
>    other than stdout and stderr before exiting, since the call to
>    _exit will bypass other buffer flushing.  Applications should
>    be flushing and closing other streams anyway, to check for I/O
>    errors.  Also, applications should not use tmpfile, since _exit
>    can bypass the removal of these files.
> 
>    It's important to detect such failures and exit nonzero because many
>    tools (most notably 'make' and other build-management systems) depend
>    on being able to detect failure in other tools via their exit status.  */
> 
> void
> close_stdout (void)
> {
>   if (close_stream (stdout) != 0
>       && !(ignore_EPIPE && errno == EPIPE))
>     {
>       char const *write_error = _("write error");
>       if (file_name)
>         error (0, errno, "%s: %s", quotearg_colon (file_name),
>                write_error);
>       else
>         error (0, errno, "%s", write_error);
> 
>       _exit (exit_failure);
>     }
> 
>    if (close_stream (stderr) != 0)
>      _exit (exit_failure);
> }


I confess I am a bit puzzled by the tendency in gnulib to document
in the implementation file instead of the header.  Why is that?