Re: test-readlink failure on GNU/Linux

[Ludovic Courtès]

Hi Eric,

Eric Blake <address@hidden> skribis:

> On 10/14/2011 04:25 PM, Ludovic Courtès wrote:
>> Hello,
>>
>> ‘test-readlink’ from GNU M4 1.4.16 fails on GNU/Linux with Linux 3.0.4
>> and GNU libc 2.13 with an assertion failure at test-readlink.h:41:
>>
>>      40: ASSERT (func ("", buf, sizeof buf) == -1);
>>      41: ASSERT (errno == ENOENT);
>>
>> It turns out that errno is set to EINVAL instead of ENOENT, which is
>> indeed not what POSIX suggests [0].
>>
>> Thoughts?
>
> Well known report.

Great!

> Gnulib has already been relaxed to work around the Linux violation of
> POSIX,

Are you referring to commit 68483ca35c0dbd347744327b1020b2e4215454f1?

> and the next time m4 is released, it will include the gnulib
> relaxation.  Good luck convincing the kernel folks to fix their
> violation, or the glibc folks of the need to work around the kernel
> error.

Heh.  :-)  The latter is more surprising to me since ISTR the glibc
folks were more careful about providing the right errno for Hurd and
Linux.

Thanks,
Ludo’.