WG: FAIL SDL-2.529_1 v5.10.1 GNU/Linux

Hi Chad,

looks like these everytime different errors are due to thread unsafty.

I hope we find out what went wrong... I guess we don’t need more information
from you.

Thanks, Tobias

-----Ursprüngliche Nachricht-----
Von: Chad Davis [mailto:[email protected]] 
Gesendet: Donnerstag, 3. Februar 2011 15:11
An: Tobias Leich
Betreff: Re: FAIL SDL-2.529_1 v5.10.1 GNU/Linux

Hi Tobias,

Thank you for following up.

I've been trying to track this down, but am having a really difficult
time isolating this.

First, Alien::SDL is a dependency, and that was already causing some
difficulty. I was not installing from within the CPAN shell, but would
test successfully from the bash shell. After I undefined PERL_MB_OPT
(which had been defined in my environment because I use local::lib, I
was able to install Alien::SDL from within the CPAN shell. Can't
decide if this was problem with CPAN.pm or with Module::Build or with
Alien::SDL.

The SDL errors also were different between the CPAN and the bash
shell. And the errors were different on different runs.

One run
mixer_channels.t
Can't coerce UNKNOWN to string in gvsv at
/usr/local/perl/Ubuntu-10.10-x86_64/lib/perl5/Test/Builder.pm line
2287.


Another run
mixer_channels.t :
All 34 subtests passed
(Seems to die after the 'expire_channel' test or on 'delay definedness
madness test \#13')


Another run
t/mixer.t ....................... ok
panic: restartop
# Tests were run but no plan was declared and done_testing() was not seen.
Unbalanced scopes: 12 more ENTERs than LEAVEs
Unbalanced saves: 121 more saves than restores
Unbalanced tmps: 3 more allocs than frees
Unbalanced context: 9 more PUSHes than POPs
Can't undef active subroutine during global destruction.
t/mixer_channels.t ..............
All 31 subtests passed


Finally, I noticed that if the tests are interrupted, e.g. SIGINT,
then any subsequent runs of mixer_channels.t will generally also fail.
This error was repeatable (and even resulted in a 'double free or
corruption' in glibc). I was able to undo it finally by changing the
volume on my sound card, then the tests all passed again.


Most of these were on the 2.529_1 but the 2.530 seems more stable, but
I don't know what changed specifically. I can get the mixer_channel.t
in 2.530 to die after 'expire_channel' if I try to interrupt the tests
at just the right moment, but otherwise the 2.530 seems less likely to
fail, though I haven't tested it 100 times yet.


Are these errors useful for you? Can I give you some specific output?


Chad


On Wed, Feb 2, 2011 at 22:43, Tobias Leich <[email protected]> wrote:
> Hi, we got your test report for SDL:
>
>
http://www.cpantesters.org/cpan/report/94899a96-2e50-11e0-bb60-39b910278d4b
>
> Thanks so far!
>
>
>
> The problem is that there is no helpful content within the report itself.
> May I ask you to send me a mail, with verbose out?
>
>
>
> I’d like to have the output of:   ./Build test --test_files
> t/mixer_channels.t –verbose
>
>
>
> Thanks in advance!
>
>
>
> Tobias