Compiler timeouts in the basic_string tests

[Andreas Tobler]

Hi all,

Stuart Hastings wrote:

>>WARNING: program timed out.
>>compiler exited with status 1
>>FAIL: 21_strings/basic_string/inserters_extractors/wchar_t/1.cc (test  
>>for excess errors)
>>Excess errors:
>>exit status is 1
>>WARNING: 21_strings/basic_string/inserters_extractors/wchar_t/1.cc  
>>compilation failed to produce executable
>>
>>I don't think this can have anything to do with the things I was  
>>testing (library changes in clause 23 stuff); it looks like a compiler  
>>problem.  No guesses about what might have caused it, other than that  
>>it must have been recent.
>>
>>I'm rerunning my tests on a different machine, just to rule out the  
>>possibility of something random like a hardware problem.  It would be  
>>nice to know if anyone else has seen this, though.
> 
> 
> I've seen one or two DejaGnu failures due to timeouts; don't recall
> the failing test offhand (and I'm too lazy to track it down right
> now).  I don't think it was a C++ test. 

 From my observations it has nothing to do with the compiler nor with 
dejagnu. It's simply vm exhaustion.

I recently got this too. I did not have it because I moved my swap away 
from the root disk to a partition which has about 1Gig free and which 
was/is not used for anything else.

Then I did an update to 10.3.2 and somehow my swap got moved to the root 
disk again. The root disk itself has about 1.7Gig of free space.

After this update I got this timeouts too.

Now I moved my swap back to this dedicated partition and the timeouts 
are gone.

Every time I run the libstdc++ testsuite I have to take care that no 
other memory (vm) consuming app is running. E.g, if I forget mozilla 
which uses about 400Meg I can count on a system freeze.

Now quitting all apps and running the c++ suite, I get every time a 
window popping up telling me the disk (swap disk) is full. (with some 
hints on what to do)
But the tests pass. (This happens on the test cases which use the 
set_memory_limits() function.)

While having the swap on my root disk (where I have more space) it 
continues to swap longer and therefor it possibly times out.

What we lack, my opinion, is a proper memory, in this case vm, 
limitation. We should be able to tell the application (the testcases) 
how much memory they can consume. Unfortunately the mechanism in the c++ 
testsuite doesn't work for darwin. And I wasn't able to find a proper 
way to do a limitation. Maybe one of you has an idea.

Can you agree on my observations?

Andreas