Conform failures in glibc?

Conform failures in glibc?

[Steve Ellcey]

I just ran the glibc testsuite and am seeing a number of conform failures
with the following messages in the err file:

/tmp/tmprhwms3a9/test.c:1:10: fatal error: ndbm.h: No such file or directory
    1 | #include <ndbm.h>
      |          ^~~~~~~~
compilation terminated.
In file included from /tmp/tmpyhffvdkg/test.c:1:
/home/sellcey/gcc_newbuild/install/lib/gcc/aarch64-linux-gnu/10.0.0/include/varargs.h:4:2: error: #error "GCC no longer implements <varargs.h>."
    4 | #error "GCC no longer implements <varargs.h>."
      |  ^~~~~
/home/sellcey/gcc_newbuild/install/lib/gcc/aarch64-linux-gnu/10.0.0/include/varargs.h:5:2: error: #error "Revise your code to use <stdarg.h>."
    5 | #error "Revise your code to use <stdarg.h>."
      |  ^~~~~


I am using the latest GCC and the latest GLIBC on an Aarch64 system.  I do
have some local changes in glibc but nothing that should be affecting these
files.  Has anyone else run into this?

Steve Ellcey
sellcey@marvell.com

Re: Conform failures in glibc?

[Florian Weimer]

* Steve Ellcey:

> I just ran the glibc testsuite and am seeing a number of conform failures
> with the following messages in the err file:
>
> /tmp/tmprhwms3a9/test.c:1:10: fatal error: ndbm.h: No such file or directory
>     1 | #include <ndbm.h>
>       |          ^~~~~~~~
> compilation terminated.
> In file included from /tmp/tmpyhffvdkg/test.c:1:
> /home/sellcey/gcc_newbuild/install/lib/gcc/aarch64-linux-gnu/10.0.0/include/varargs.h:4:2: error: #error "GCC no longer implements <varargs.h>."
>     4 | #error "GCC no longer implements <varargs.h>."
>       |  ^~~~~
> /home/sellcey/gcc_newbuild/install/lib/gcc/aarch64-linux-gnu/10.0.0/include/varargs.h:5:2: error: #error "Revise your code to use <stdarg.h>."
>     5 | #error "Revise your code to use <stdarg.h>."
>       |  ^~~~~
>

These should be XFAILed and not really count as failures.

I find these errors confusing as well.  On the one hand, I think want
the conform test expectations to reflect POSIX as closely as possible,
but on the other hand, I do not see value in carrying around such fringe
XFAILs forever.  The <varargs.h> one in particular will never get fixed
at this point.  We know that, so reporting it over and over again is
just noise.

Thanks,
Florian

Re: [EXT] Re: Conform failures in glibc?

[Steve Ellcey]

On Tue, 2019-05-21 at 19:10 +0200, Florian Weimer wrote:
> 
> These should be XFAILed and not really count as failures.
> 
> I find these errors confusing as well.  On the one hand, I think want
> the conform test expectations to reflect POSIX as closely as possible,
> but on the other hand, I do not see value in carrying around such fringe
> XFAILs forever.  The <varargs.h> one in particular will never get fixed
> at this point.  We know that, so reporting it over and over again is
> just noise.
> 
> Thanks,
> Florian

I actually hadn't run into these before.  I have recently changed my 
build setup so maybe that is what caused me to start seeing these.

I see glibc/conform/Makefile has some XFAILS:

# Pre-standard C feature no longer supported by GCC (obsoleted in
# newer POSIX standards).
test-xfail-XPG4/varargs.h/conform = yes
test-xfail-XPG42/varargs.h/conform = yes
test-xfail-UNIX98/varargs.h/conform = yes

# Header not provided by glibc.
test-xfail-XPG42/ndbm.h/conform = yes
test-xfail-UNIX98/ndbm.h/conform = yes
test-xfail-XOPEN2K/ndbm.h/conform = yes
test-xfail-XOPEN2K8/ndbm.h/conform = yes


The failures I am seeing are:

FAIL: conform/POSIX2008/arpa/inet.h/conform
FAIL: conform/POSIX2008/netdb.h/conform
FAIL: conform/POSIX2008/netinet/in.h/conform
FAIL: conform/POSIX2008/sys/socket.h/conform
FAIL: conform/UNIX98/arpa/inet.h/conform
FAIL: conform/UNIX98/netdb.h/conform
FAIL: conform/UNIX98/netinet/in.h/conform
FAIL: conform/UNIX98/sys/socket.h/conform
FAIL: conform/XOPEN2K8/arpa/inet.h/conform
FAIL: conform/XOPEN2K8/netdb.h/conform
FAIL: conform/XOPEN2K8/netinet/in.h/conform
FAIL: conform/XOPEN2K8/sys/socket.h/conform
FAIL: conform/XOPEN2K/arpa/inet.h/conform
FAIL: conform/XOPEN2K/netdb.h/conform
FAIL: conform/XOPEN2K/netinet/in.h/conform
FAIL: conform/XOPEN2K/sys/socket.h/conform
FAIL: conform/XPG42/arpa/inet.h/conform
FAIL: conform/XPG42/netdb.h/conform
FAIL: conform/XPG42/netinet/in.h/conform
FAIL: conform/XPG42/sys/socket.h/conform

Steve Ellcey
sellcey@marvell.com

Re: [EXT] Re: Conform failures in glibc?

[Florian Weimer]

* Steve Ellcey:

> On Tue, 2019-05-21 at 19:10 +0200, Florian Weimer wrote:
>> 
>> These should be XFAILed and not really count as failures.
>> 
>> I find these errors confusing as well.  On the one hand, I think want
>> the conform test expectations to reflect POSIX as closely as possible,
>> but on the other hand, I do not see value in carrying around such fringe
>> XFAILs forever.  The <varargs.h> one in particular will never get fixed
>> at this point.  We know that, so reporting it over and over again is
>> just noise.
>> 
>> Thanks,
>> Florian
>
> I actually hadn't run into these before.  I have recently changed my 
> build setup so maybe that is what caused me to start seeing these.

> FAIL: conform/POSIX2008/arpa/inet.h/conform
> FAIL: conform/POSIX2008/netdb.h/conform
> FAIL: conform/POSIX2008/netinet/in.h/conform

You need to check the .out files.  It could be the Linux 5.1 UAPI
header issue.