ridiculous memory consumption in libstdc++ test suite

[Nathan Sidwell]

Zack Weinberg wrote:

> 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

>
> 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 wonder, could this be related to bug 2517 which reports a
memory leak with string class? I have not verified.

nathan

--
Dr Nathan Sidwell :: http://www.codesourcery.com :: CodeSourcery LLC
'But that's a lie.' - 'Yes it is. What's your point?'
nat...@codesourcery.com : http://www.cs.bris.ac.uk/~nathan/ : nat...@acm.org

[Robert Lipe]

Zack Weinberg wrote:

> 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 have to set ulimits in my testsuite runs, too. Certain tests have
historically consumed all of VM otherwise and one of my OpenServer
handles this poorly. (UnixWare handles it very well but it has a much
smarter VM system.) I've found setting ulimit -v and ulimit -c in the
script that calls 'make check' to be an effective governor on the amount
of resources consumed.

Ulimit is sometimes a shell intrinsic and sometimes a binary. Be
sure the rlimits are being passed on through all the way down to the
children.

RJL

[Zack Weinberg]

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.

zw