Re: Failures in the conform tests removed from arch-independent failure list for release.

["Joseph S. Myers" <joseph at codesourcery dot com>]

On Wed, 3 Sep 2014, Carlos O'Donell wrote:

> FAIL: conform/POSIX/semaphore.h/conform
> 
> Is caused by:
> Testing <semaphore.h>
> ---------------------
>   Checking whether <semaphore.h> is available... FAIL
>     Header <semaphore.h> not available  Compiler message:
>     -----------------------------------------------------------------------
>     /home/carlos/build/glibc/conform/POSIX/semaphore.h/scratch/semaphore.h-test.c:1:23: fatal error: semaphore.h: No such file or directory
>      #include <semaphore.h>
>                            ^
>     compilation terminated.

That sounds like the change to NPTL not being an add-on has caused 
<semaphore.h> not to be in the include paths used by conformtest 
(specifically, caused nptl/ no longer to be in $(+sysdep-includes)), so 
making the test either fail or wrongly find an existing installed 
<semaphore.h>.

The simple fix would be to have a file include/semaphore.h that just does 
#include <nptl/semaphore.h>, as done for other headers to ensure a header 
from one bit of glibc can be found while building another bit.  That's not 
logically clean, however, since there could be other thread library 
implementations on other OSes.  Or you could move the header to 
sysdeps/nptl/ alongside pthread.h.  Really, nothing about this header is 
NPTL-specific at all, so maybe it (and the "headers" setting to install 
it) belongs somewhere such as sysdeps/pthread that would be shared by all 
thread library implementations.

The <cpio.h> and <fmtmsg.h> failures seem like those should also have 
trivial wrappers in include/ (but I don't see any reason for those to be 
new, unlike the <semaphore.h> failures where the change to how NPTL is 
built could well have affected things).

-- 
Joseph S. Myers
joseph@codesourcery.com