This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: ridiculous memory consumption in libstdc++ test suite


On Fri, Apr 27, 2001 at 02:50:21AM -0700, Zack Weinberg wrote:
> My current libstdc++ test results (mainline):
> 
>                 === libstdc++ tests ===
> 
> FAIL: 21_strings/ctor_copy_dtor.cc (execution test), static  
> FAIL: 21_strings/ctor_copy_dtor.cc (execution test), shared  
> FAIL: 21_strings/insert.cc (execution test), static  
> FAIL: 21_strings/insert.cc (execution test), shared  
> FAIL: 22_locale/ctor_copy_dtor.cc (execution test), static  
> FAIL: 22_locale/ctor_copy_dtor.cc (execution test), shared  
> FAIL: 27_io/ios_members.cc (execution test), static  
> 
>                 === libstdc++ Summary ===
> 
> # of expected passes            403
> # of unexpected failures        7
> # of expected failures          16
> 
> All the 21_strings failures are because I had to shoot the test
> process after it chewed up all the RAM on the machine and well into
> swap.  SIZE column in ps was >512MB in all cases.  The machine has
> 256 megs of real RAM and a gigabyte of swap.
> 
> I have tried setting ulimits before running make check, but they are
> ineffective.  I may be setting the wrong ones, or something may be
> resetting them.
> 
> I shouldn't have to kill tests or set ulimits in order for the test
> suite to be a well-behaved background job and not interfere with my
> ability to r/e/a/d/ U/s/e/n/e/t/ get work done while it's running.
> 
REF:
http://gcc.gnu.org/ml/gcc/2001-04/msg00663.html
http://gcc.gnu.org/ml/gcc/2001-04/msg00706.html

I've had similar problems with gcc-3_0-branch.. I've gone so far as to
provide some, hopefully useful, backtrace info on the
libjava.lang/klass.java test. The 21_strings/substr.cc libstdc++ test
I've had 0 luck determining the problem, as noone else seems to have
either of these problems, I wonder why?

Originally I went so far as adding a 5-min interval cronjob to just
killall lt-klass 
killall substr.exe

My bootstraps occur overnight, while I'm not physically present. Now,
I've just modified both tests to exit rather than run their test code.
Horrible, but otherwise my bootstraps would just hang on those tests and
never complete...


Gordon Sadler



Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]