In article <20160324125858.516aa36bc70bddd1b180e...@schemamania.org>,
James K. Lowden <tech-userlevel@netbsd.org> wrote:
>On Thu, 24 Mar 2016 14:29:39 +0000 (UTC)
>mlel...@serpens.de (Michael van Elst) wrote:
>
>> >(I agree that calling strerror(0) is odd and a likely sign of a bug,
>> >but that's separate from complying with standards when compliance
>> >isn't harmful.)
>>
>> Well, if calling strerror(0) is a bug, then usually because it's
>> called after an error condition that didn't set errno.
>
>Has that happened to you? I feel sure that's never happened to me, but
>maybe I'm less adventurous.
>
>IMO it's incorrect to say 0 is an "undefined error". It's defined as
>success by nearly every syscall. The emitted string should reflect
>that, even if [sic] Posix says so. Perhaps,
Actually not; syscalls set errno on error, don't clear errno on success [1].
Thus it is incorrect code to test errno if the syscall succeeded. It shows
nothing, try:
#include <stdio.h>
#include <fcntl.h>
#include <errno.h>
int
main(void)
{
errno = 3;
open("/", O_RDONLY);
printf("%d\n", errno);
return 0;
}
[1] There is at least one syscall stub that clears errno before the syscall
which I find bogus; can you guess which one?
christos
