[LTP] [PATCH 4/7] syscalls/fsmount: New tests

[Li Wang]

On Mon, Feb 17, 2020 at 5:58 PM Petr Vorel <pvorel@suse.cz> wrote:

> Hi Li,
>
> > On Mon, Feb 17, 2020 at 4:29 PM Viresh Kumar <viresh.kumar@linaro.org>
> > wrote:
>
> > > ...
> > > > Adding #include "lapi/fcntl.h" in case of ‘AT_FDCWD’ undeclared. It
> seems
> > > > we have to do this for all the tests involves  ‘AT_FDCWD’.
>
> > > my fsmount.h header includes <fcntl.h>, won't that be enough ?
>
>
> > I'm afraid it's not enough.
>
> > After having a closer look, the reason is that the AT_FDCWD is defined
> with
> > condition __USE_ATFILE,  the __USE_ATFILE depends on _ATFILE_SOURCE
> > defined, and _ATFILE_SOURCE needs enable _GNU_SOURCE.
>
> > So another effective way is to add '#define _GNU_SOURCE' in front of your
> > test then that <fcntl.h> will be work for you.
>
> OK, we can add #define _GNU_SOURCE to fsmount01.c, which actually needs it
> + keep <fcntl.h> there and load lapi/fsmount.h later (see below).
> But generally this will be the approach for all uses of <fcntl.h> (and
> probably
> some other headers) for old distros. IMHO this change is caused by:
>
> c941736c92 Remove _BSD_SOURCE and _SVID_SOURCE. (glibc-2.20)
> c688b41960 Add _DEFAULT_SOURCE feature test macro. (glibc-2.19)
>

Thanks for figure out this.

>
> So we can either add it to many places or detect this old glibc and
> compile with
> -D_GNU_SOURCE (but this might break other things).
>

-1 remove old Glibc is a bad choice.


> But that's another story.
>
> > # rpm -qa glibc-headers
> > glibc-headers-2.5-123
>
> > # cat /usr/include/fcntl.h |grep AT_FDCWD -B 2 -A 2
> > #ifdef __USE_ATFILE
> > # define AT_FDCWD -100   /* Special value used to indicate
> >                                                 the *at functions should
> > use the
> >                                                 current working
> directory.
> > */
>
> So is this diff to v7 better?
>

To use "lapi/fcntl.h" in v7 is simpler I guess, sometimes we just need to
get compile pass on old distros but not perform it indeed. And that's the
purpose why we need lapi/*.

-- 
Regards,
Li Wang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linux.it/pipermail/ltp/attachments/20200217/8ae9c1ca/attachment.htm>