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


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

FYI, recent commit


I don't know the exact time window but someone committed a change to
the library implementation which causes (at least):

	testsuite/thread/pthread2.cc & testsuite/thread/pthread3.cc

to gobble memory until they are killed by the VM system (on at least
FreeBSD/i386 4.9 with .5GB of VM; slightly larger 32-bit machines will
display ever growing allocations until >350MB is allocated for
pthread2.cc and the test then naturally terminates).

As an example, under g++-3.2.3 (only version quick at hand), the
pthread2.cc case runs in 17.5u (user-seconds) on the aforementioned
machine and never allocates more than ~2MB.  I suspect mainline has
been running at or near this level for some time.  I'd consider that a
reasonable QoI goal for that test (in terms of memory behavior).

With mainline today, after 17u is allocated to the process, it has
allocated >150MB...

Regards,
Loren

PS, hopefully this is obvious to someone that broke it; if not
ping me.


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