Re: Test results for egcs-1.1 prerelease on i586-pc-linux-gnu

[Manfred Metzger <ManfredMetzger at swol dot de>]

Jeffrey A Law wrote:

> ...
>   > FAIL: g++.robertl/eb59.C (test for excess errors)
>   > FAIL: g++.robertl/eb60.C (test for excess errors)
>   > FAIL: g++.robertl/eb62.C (test for excess errors)
>   > FAIL: g++.robertl/eb7.C (test for excess errors)
> This seems rather poor.  Can you look at g++.log and see if there is
> any common thread to these unexpected failures?

ALL g++ failures look like this one:

FAIL: g++.brendan/crash30.C (test for excess errors)
In file included from /usr/src/gnu/egcs/libstdc++/stl/stl_alloc.h:73,
                 from /usr/src/gnu/egcs/libstdc++/stl/alloc.h:21,
                 from /usr/src/gnu/egcs/libstdc++/std/bastring.h:39,
                 from /usr/src/gnu/egcs/libstdc++/string:6,
                 from
/usr/src/gnu/egcs/gcc/testsuite/g++.old-deja/g++.brendan/crash30.C:3:
/usr/include/pthread.h:438: parse error before `*'

In pthread.h, line 438 is:

438: extern int pthread_sigmask __P ((int __how, __const sigset_t *__newmask,
439:                                sigset_t *__oldmask));

Looking at the preprocessor output, I noticed that sigset_t is undefined.
While the test uses -ansi, the test programs can be compiled without -ansi (I only
testet a few).
Without -ansi sigset_t is typedef __sigset_t (in /usr/include/signal.h),
__sigset_t  is defined
in /usr/include/sigset.h. Sorry, I donīt have the time to investigate this further.

Maybe a bug in glibc 2.0.6.

>
>
>
>   > I got these results after installing the new egcs-version because there
>   > is a problem
>   > with the g++ testsuite:
>   > The first time I did īmake checkī (with egcs-1.1 not installed) I got a
>   > failure in every g++
>   > execution test because the dynamic linker  couldnīt find
>   > libstdc++.so.2.9 (I had no earlier
>   > version installed). This means that if there is an older version of
>   > libstdc++.so.2.9 installed,
>   > the the tests are done with that library, which is surely not intended.
> There should not be an older libstdc++.so.2.9 installed since egcs-1.1
> will be the first release of libstdc++-2.9.

Ok, but anyway the tests fail when egcs-1.1 is not installed.
The other problem is, that if I would now, that a libstdc++.so.2.9 is installed,
test for example egcs-1.1.x, the tests are done with the libstdc++ from the
egcs-1.1-prerelease, not with the one which I would get with īmake installī.

Manni