2 new regression on darwin6.1

2 new regression on darwin6.1

[Andreas Tobler]

Hi,
so far I couldn't find what is the reason for, but on darwin6.1 I get 2 
regressions by this evening.
IOW, the tests passed with last nights cvs on trunk. It must have 
happend during the last 8 hours.

Here the failing TC's:

FAIL: gcc.dg/cpp/assembler.S spurious warning (test for bogus messages, 
line 7)
FAIL: gcc.dg/cpp/assembler.S (test for excess errors)

Right now I run the g++ suite, so I can't interrupt.

Regards,
Andreas

Re: 2 new regression on darwin6.1

[Aldy Hernandez]

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

 > Hi,
 > so far I couldn't find what is the reason for, but on darwin6.1 I get
 > 2 regressions by this evening.
 > IOW, the tests passed with last nights cvs on trunk. It must have
 > happend during the last 8 hours.

 > Here the failing TC's:

 > FAIL: gcc.dg/cpp/assembler.S spurious warning (test for bogus
 > messages, line 7)
 > FAIL: gcc.dg/cpp/assembler.S (test for excess errors)

This might be a dejagnu bug.  It may also be some silly OS bug when
spawning lots of processes.  It gets confused, and you end up getting
bogus errors.  I don't know if this is fixed in Jaguar, but I know
that in previous versions you'd have to do "make check" a few times
and report the union of the runs :).

Perhaps Jim Ingham can elaborate on this.  He had explained the bug to
me in more eloquent words.

...and perhaps, it's an actual regression failure..

Aldy

Re: 2 new regression on darwin6.1

[Andreas Tobler]

Aldy Hernandez wrote:

> This might be a dejagnu bug.  It may also be some silly OS bug when
> spawning lots of processes.  It gets confused, and you end up getting
> bogus errors.  I don't know if this is fixed in Jaguar, but I know
> that in previous versions you'd have to do "make check" a few times
> and report the union of the runs :).
> 
> Perhaps Jim Ingham can elaborate on this.  He had explained the bug to
> me in more eloquent words.
> 
> ...and perhaps, it's an actual regression failure..


Hm, I guess the latter since I run a Shebs tweaked gcc which has the 
-fslow switch integrated. My results so far are like this:

                 === gcc Summary ===

# of expected passes            20186
# of unexpected failures        21
# of expected failures          62
# of unresolved testcases       3
# of unsupported tests          139

Andreas

Re: 2 new regression on darwin6.1

[Richard Earnshaw]

> >>>>> "Andreas" == Andreas Tobler <toa@pop.agri.ch> writes:
> 
>  > Hi,
>  > so far I couldn't find what is the reason for, but on darwin6.1 I get
>  > 2 regressions by this evening.
>  > IOW, the tests passed with last nights cvs on trunk. It must have
>  > happend during the last 8 hours.
> 
>  > Here the failing TC's:
> 
>  > FAIL: gcc.dg/cpp/assembler.S spurious warning (test for bogus
>  > messages, line 7)
>  > FAIL: gcc.dg/cpp/assembler.S (test for excess errors)
> 
> This might be a dejagnu bug.  It may also be some silly OS bug when
> spawning lots of processes.  It gets confused, and you end up getting
> bogus errors.  I don't know if this is fixed in Jaguar, but I know
> that in previous versions you'd have to do "make check" a few times
> and report the union of the runs :).

No, I think it's real.  It came up for me last night on an arm-elf run.  I 
don't recall seeing that one before.

R.

Re: 2 new regression on darwin6.1

[Neil Booth]

Richard Earnshaw wrote:-

> No, I think it's real.  It came up for me last night on an arm-elf run.  I 
> don't recall seeing that one before.

-ansi was being passed to the preprocessor when preprocessing assembler
(-ansi implies -std=c89).  I've fixed the testcase.

IMO the testsuite needs to default its flags based upon a file's
extension, rather than the current method which is hard to control.

Neil.

Re: 2 new regression on darwin6.1

[David Edelsohn]

	I see the regressions on AIX as well.

David

Re: 2 new regression on darwin6.1

[Andreas Tobler]

Neil Booth wrote:
> Richard Earnshaw wrote:-
> 
> 
>>No, I think it's real.  It came up for me last night on an arm-elf run.  I 
>>don't recall seeing that one before.
> 
> 
> -ansi was being passed to the preprocessor when preprocessing assembler
> (-ansi implies -std=c89).  I've fixed the testcase.

Verified, working now.

Running /Volumes/xufs/gccsrc/gcc/gcc/testsuite/gcc.dg/cpp/cpp.exp ...

                 === gcc Summary ===

# of expected passes            829
# of expected failures          4

Thanks,

Andreas

Re: 2 new regression on darwin6.1

[Jim Ingham]

Expect bugs don't seem to be the cause of this, which is great, but 
just for background...

There were some kernel bugs that were causing expect to hang waiting 
for a program that exited.  One bug caused wait to sometimes not return 
when the program died, this one was fixed in Jaguar.  Another caused 
loss of output from a program that had died (when the program was 
killed the output buffers were not being flushed or something).  This 
one can be worked around by judicious sleep's before exit (I think the 
folks here hacked around it this way in gcc - Stan or somebody would 
know more).  It is also fixed, but not in a released kernel.

With these two fixes, expect now seems to run both the gcc & gdb 
testsuites well on X.  Actually, gdb was only hit by the former bug, so 
its testsuite runs fine on straight Jaguar.

Jim

On Thursday, October 10, 2002, at 10:26  PM, Aldy Hernandez wrote:

>>>>>> "Andreas" == Andreas Tobler <toa@pop.agri.ch> writes:
>
>> Hi,
>> so far I couldn't find what is the reason for, but on darwin6.1 I get
>> 2 regressions by this evening.
>> IOW, the tests passed with last nights cvs on trunk. It must have
>> happend during the last 8 hours.
>
>> Here the failing TC's:
>
>> FAIL: gcc.dg/cpp/assembler.S spurious warning (test for bogus
>> messages, line 7)
>> FAIL: gcc.dg/cpp/assembler.S (test for excess errors)
>
> This might be a dejagnu bug.  It may also be some silly OS bug when
> spawning lots of processes.  It gets confused, and you end up getting
> bogus errors.  I don't know if this is fixed in Jaguar, but I know
> that in previous versions you'd have to do "make check" a few times
> and report the union of the runs :).
>
> Perhaps Jim Ingham can elaborate on this.  He had explained the bug to
> me in more eloquent words.
>
> ...and perhaps, it's an actual regression failure..
>
> Aldy
>
--
Jim Ingham                                   jingham@apple.com
Developer Tools
Apple Computer

Re: 2 new regression on darwin6.1

[Aldy Hernandez]

> With these two fixes, expect now seems to run both the gcc & gdb 
> testsuites well on X.  Actually, gdb was only hit by the former bug, so 
> its testsuite runs fine on straight Jaguar.

Hallelujah!  Now I can go back to running tests on Darwin.

Thanks for the info.

Aldy

Re: 2 new regression on darwin6.1

[Stan Shebs]

Jack Howarth wrote:

>    I wonder are you using RedHat's expect patches? On debian, we
>were getting spurious false failures earlier this year which have
>been all eliminated by that fix. It is described in the debian 
>bug report...
>
>http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=156854&archive=yes&repeatmerged=yes
>
>I did forward that patch upstream to the expect maintainers but I don't
>think its in a new release yet.
>                         Jack
>
>
The Darwin problem is a kernel bug manifesting as unflushed buffers
when a process exits.  We have a couple people trying out a fixed
kernel, although my guess is that the fix probably won't appear in
a Software Update, since only a small number of users are hardcore
enough to run dejagnu... :-)

Stan

Re: 2 new regression on darwin6.1

[Jack Howarth]

    I wonder are you using RedHat's expect patches? On debian, we
were getting spurious false failures earlier this year which have
been all eliminated by that fix. It is described in the debian 
bug report...

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=156854&archive=yes&repeatmerged=yes

I did forward that patch upstream to the expect maintainers but I don't
think its in a new release yet.
                         Jack