Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Bradley Lucier]

Just a note that there were several thousand g++ and libstdc++ test 
failures on 3.4 mainline from yesterday afternoon on sparc and sparc64; 
full test results are at

http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00119.html

Brad

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Eric Botcazou]

> Just a note that there were several thousand g++ and libstdc++ test
> failures on 3.4 mainline from yesterday afternoon on sparc and sparc64;
> full test results are at
>
> http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00119.html

Everything was as usual just hours before:
http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00117.html
http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00116.html
http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00115.html

-- 
Eric Botcazou

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Benjamin Kosnik]

Can you tell me if this stil happens with current CVS sources? Also,
does libstdc++ build?

thanks,
benjamin

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Bradley Lucier]

On Friday, October 3, 2003, at 09:53  AM, Eric Botcazou wrote:

>> Just a note that there were several thousand g++ and libstdc++ test
>> failures on 3.4 mainline from yesterday afternoon on sparc and 
>> sparc64;
>> full test results are at
>>
>> http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00119.html
>
> Everything was as usual just hours before:
> http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00117.html
> http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00116.html
> http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00115.html

You updated a few hours later (I should have given the update time in 
GMT); I'll try it again now.  Expect esults in about 14 1/2 hours :-(.

Thanks.

Brad

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Andreas Tobler]

Benjamin Kosnik wrote:

> Can you tell me if this stil happens with current CVS sources? Also,
> does libstdc++ build?

I started a build this morning on darwin, linuxppc, sol9 and hppa-linux. 
Darwin, linuxppc are ok. Sol9 and hppa suffer these failures. Now I 
synced libstdc++ again on sol9 and build. I think results will be 
available in a few minutes. But I'd say it will be the same. And yes, 
libstdc++ will build.

Affected are g++, libstdc++ and I think also libgcj?

Andreas,

BTW, Benjamin, for further details #gcc irc.oftc.net

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Gabriel Dos Reis]

Andreas Tobler <toa@pop.agri.ch> writes:

| BTW, Benjamin, for further details #gcc irc.oftc.net

I would be highly desirable if discussions on such details about
GCC and its components happened on GCC lists so that they can be
archived for future references.

-- Gaby

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Andreas Tobler]

Gabriel Dos Reis wrote:

> Andreas Tobler <toa@pop.agri.ch> writes:
> 
> | BTW, Benjamin, for further details #gcc irc.oftc.net
> 
> I would be highly desirable if discussions on such details about
> GCC and its components happened on GCC lists so that they can be
> archived for future references.

No worry, I certainly will post my investigations on the LISTS. At the 
moment I binary search the offending patch which introduced the 
failures. As it seems it is a libstdc++-v3 issue and not a front end 
thingy. It takes a few minutes until I have it.

Sorry for mentioning irc. Just thought for faster tracking the failure 
some online hints would help.

Andreas

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Zack Weinberg]

Andreas Tobler <toa@pop.agri.ch> writes:

>
> Sorry for mentioning irc. Just thought for faster tracking the failure
> some online hints would help.

I don't think anyone should apologize for mentioning irc.  The channel
has been a valuable resource for quick communication between
developers, and its use should be encouraged.

However, perhaps it would be a good idea to have logs of channel
traffic posted somewhere.  Other free-software projects that use
IRC for developer discussion do this.

zw

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Gabriel Dos Reis]

"Zack Weinberg" <zack@codesourcery.com> writes:

| Andreas Tobler <toa@pop.agri.ch> writes:
| 
| >
| > Sorry for mentioning irc. Just thought for faster tracking the failure
| > some online hints would help.
| 
| I don't think anyone should apologize for mentioning irc.  The channel
| has been a valuable resource for quick communication between
| developers, and its use should be encouraged.

FWIW, nobody has been discouraging use of IRC for communication
between developpers. My remark was very specific.

-- Gaby

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Bradley Lucier]

On Friday, October 3, 2003, at 10:40  AM, Benjamin Kosnik wrote:

> Can you tell me if this stil happens with current CVS sources?

Yes

>  Also, does libstdc++ build?

Yes.  Final results in a few hours.

Brad

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Eric Botcazou]

> Can you tell me if this stil happens with current CVS sources? Also,
> does libstdc++ build?

This is still present as of today (Thu Oct  9 04:48:45 UTC 2003) on Solaris 
2.6, 2.7, 2.8, 2.9 but not 2.5.1.

Here is for example a GDB session for:
FAIL: g++.dg/expr/enum1.C execution test

Program received signal SIGSEGV, Segmentation fault.
__atomic_add(int volatile*, int) (__mem=0x0, __val=1)
    at 
/opt/build/eric/gcc/sparc-sun-solaris2.8/libstdc++-v3/include/sparc-sun-solaris2.8/bits/atomicity.h:122
122       *__mem += __val;
(gdb) where
#0  __atomic_add(int volatile*, int) (__mem=0x0, __val=1)
    at 
/opt/build/eric/gcc/sparc-sun-solaris2.8/libstdc++-v3/include/sparc-sun-solaris2.8/bits/atomicity.h:122
#1  0xff2e3678 in std::ios_base::Init::Init() (this=0xff35b95c)
    at 
/opt/build/eric/gcc/sparc-sun-solaris2.8/libstdc++-v3/include/streambuf:437
#2  0xff2e16a0 in __static_initialization_and_destruction_0(int, int) (
    __initialize_p=-13256356, __priority=0)
    at 
/opt/build/eric/gcc/sparc-sun-solaris2.8/libstdc++-v3/include/iostream:77
(gdb)

Everything is fine on Thu Oct  2 03:49:53 UTC 2003.

-- 
Eric Botcazou

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Bradley Lucier]

It seems that the g++ errors disappeared yesterday, but many of the 
libstdc++ ones remain:

http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00446.html

Brad

On Friday, October 3, 2003, at 09:22  AM, Bradley Lucier wrote:

> Just a note that there were several thousand g++ and libstdc++ test 
> failures on 3.4 mainline from yesterday afternoon on sparc and 
> sparc64; full test results are at
>
> http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00119.html
>
> Brad

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Eric Botcazou]

> It seems that the g++ errors disappeared yesterday, but many of the
> libstdc++ ones remain:
>
> http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00446.html

The C++ errors have not totally disappeared: on 10/02, I had 7 unexpected C++ 
failures while there are still 42 unexpected failures according to your 
report. I don't think we have recovered from the disastrous change yet.

-- 
Eric Botcazou

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Andreas Tobler]

Eric Botcazou wrote:
>>It seems that the g++ errors disappeared yesterday, but many of the
>>libstdc++ ones remain:
>>
>>http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00446.html
> 
> 
> The C++ errors have not totally disappeared: on 10/02, I had 7 unexpected C++ 
> failures while there are still 42 unexpected failures according to your 
> report. I don't think we have recovered from the disastrous change yet.
> 

With the patchlet we (mostly Pétur) found, the situation improves:
http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00450.html

See also the follow up thread from this one.
http://gcc.gnu.org/ml/libstdc++/2003-10/msg00170.html
and back.

Still awaiting comments.

Andreas


RCS file: /cvs/gcc/gcc/libstdc++-v3/src/locale.cc,v
retrieving revision 1.95
diff -u -r1.95 locale.cc
--- src/locale.cc       2 Oct 2003 23:06:13 -0000       1.95
+++ src/locale.cc       9 Oct 2003 18:37:34 -0000
@@ -387,7 +387,6 @@
  #ifdef __GTHREADS
      if (__gthread_active_p())
        __gthread_once(&_S_once, _S_initialize_once);
-    else
  #endif
        {
         if (!_S_classic)

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Eric Botcazou]

> With the patchlet we (mostly Pétur) found, the situation improves:
> http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00450.html

It's far better, indeed.

> See also the follow up thread from this one.
> http://gcc.gnu.org/ml/libstdc++/2003-10/msg00170.html
> and back.

Thanks for your effort in sorting this out!

-- 
Eric Botcazou

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Andreas Tobler]

Eric Botcazou wrote:

>>With the patchlet we (mostly Pétur) found, the situation improves:
>>http://gcc.gnu.org/ml/gcc-testresults/2003-10/msg00450.html
> 
> 
> It's far better, indeed.

I have a multilib test running which shows better. I post as it finished.

> 
> 
>>See also the follow up thread from this one.
>>http://gcc.gnu.org/ml/libstdc++/2003-10/msg00170.html
>>and back.
> 
> 
> Thanks for your effort in sorting this out!

It was mostly learning for me, but quite useful.

Andreas

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Benjamin Kosnik]

So, what is the status of this patch? If it's to be something like this,
can you do a ChangeLog entry?

thanks,
benjamin


Index: src/locale.cc
===================================================================
RCS file: /cvs/gcc/gcc/libstdc++-v3/src/locale.cc,v
retrieving revision 1.96
diff -c -p -r1.96 locale.cc
*** src/locale.cc       9 Oct 2003 21:17:11 -0000       1.96
--- src/locale.cc       13 Oct 2003 15:01:03 -0000
*************** namespace std
*** 395,406 ****
  #ifdef __GTHREADS
      if (__gthread_active_p())
        __gthread_once(&_S_once, _S_initialize_once);
-     else
  #endif
!       {
!       if (!_S_classic)
!         _S_initialize_once();
!       }
    }
   
    void
--- 395,403 ----
  #ifdef __GTHREADS
      if (__gthread_active_p())
        __gthread_once(&_S_once, _S_initialize_once);
  #endif
!     if (!_S_classic)
!       _S_initialize_once();
    }
   
    void

Re: Heads up: Several thousand g++ and libstdc++ test failures on sparc[64]

[Andreas Tobler]

Benjamin Kosnik wrote:

> 
> 
> So, what is the status of this patch? If it's to be something like this,
> can you do a ChangeLog entry?

Ah!?

Certainly can I write a ChangeLog entry, if this IS the solution.
But what do I write?

Workaround for a solaris bug?

Andreas