[ptr_container] memory usage causes problem with regression tests
Hello, for approximately a week now, I'm experiencing crashes when I run the regression tests. Victor told me he also noticed tests consuming unusual amounts of memory. I think I now found the culprit: the ptr_container test suite. This is the second time I notice the kernel run out of memory when once of the ptr_container tests is run. This is a problem for running the tests since the kernel then starts to kill processes. So this needs to get fixed ASAP. The ptr_set test output is: *** glibc detected *** free(): invalid pointer: 0x14347b4c *** Running 1 test case... unknown location(0): fatal error in "test_set": signal: SIGABRT (application abort requested) *** errors detected in test suite "Pointer Container Test Suite"; see standard output for details We need a way of portably specifying and enforcing time and memory limits for tests (it's not the first time I've been hit by tests running wild). We also need the authors of tests to make use of that feature, once we have it. I think displaying the time and memory used by a test in the regression results would be helpful (requires changes to Boost.Build, too, obviously). In this special case, it also would have been helpful if absolute times were printed in the output of Boost.Build, since all I have is the time-stamp from the error messages of the kernel but no way of correlating this time-stamp with the regression tests except for incidentally looking at the screen when the crashes happen. Regards, m
"Martin Wille" <mw8329@yahoo.com.au> wrote in message news:425911C8.5050207@yahoo.com.au... | Hello, | | for approximately a week now, I'm experiencing crashes when I run the | regression tests. Victor told me he also noticed tests consuming unusual | amounts of memory. I think I now found the culprit: the ptr_container | test suite. have we got any indication about which particular test that causes this behavior? -Thorsten
Thorsten Ottosen wrote:
"Martin Wille" <mw8329@yahoo.com.au> wrote in message news:425911C8.5050207@yahoo.com.au... | Hello, | | for approximately a week now, I'm experiencing crashes when I run the | regression tests. Victor told me he also noticed tests consuming unusual | amounts of memory. I think I now found the culprit: the ptr_container | test suite.
have we got any indication about which particular test that causes this behavior?
Other than the output for test_set I pasted? No. The tests still are running. Searching manually in the log files doesn't work too well. I suggest we wait until the results get published. (This will take another 12 hours, likely) Maybe, I'll find some time to run test_set manually, tomorrow. I hope gdb will print some more information. Regards, m
"Martin Wille" <mw8329@yahoo.com.au> wrote in message news:425920FA.1070707@yahoo.com.au... | Thorsten Ottosen wrote: | > "Martin Wille" <mw8329@yahoo.com.au> wrote in message | > news:425911C8.5050207@yahoo.com.au... | > | Hello, | > | | > | for approximately a week now, I'm experiencing crashes when I run the | > | regression tests. Victor told me he also noticed tests consuming unusual | > | amounts of memory. I think I now found the culprit: the ptr_container | > | test suite. | > | > have we got any indication about which particular test that causes this | > behavior? | | Other than the output for test_set I pasted? No. | | The tests still are running. Searching manually in the log files doesn't | work too well. I suggest we wait until the results get published. (This | will take another 12 hours, likely) I think I found the error in ptr_set.cpp. It was caused by insertion into a multiset from itself. I've committe the new test file which works locally. -Thorsten
Thorsten Ottosen wrote:
I think I found the error in ptr_set.cpp.
It was caused by insertion into a multiset from itself.
I've committe the new test file which works locally.
Thanks for fixing this so quickly! I'll update that file while the current tests are running. (gcc 3.4.3 and gcc 4 test runs should already take advantage of the fix) BTW, you should comment your commits to the CVS. I updated and learned only a .hpp file had changed. The CVS log didn't tell me anything other than "*** empty log message ***". I had to diff against the previous version of that file in order to find out that this likely is the fix you talked about. Thanks, m
"Martin Wille" <mw8329@yahoo.com.au> wrote in message news:425947D2.9070109@yahoo.com.au... | Thorsten Ottosen wrote: | | > I think I found the error in ptr_set.cpp. | > | > It was caused by insertion into a multiset from itself. | > | > I've committe the new test file which works locally. | | Thanks for fixing this so quickly! | | I'll update that file while the current tests are running. | (gcc 3.4.3 and gcc 4 test runs should already take advantage of the fix) | | | BTW, you should comment your commits to the CVS. I updated and learned I should do that....I'll try to improve myself -Thorsten
On 4/15/04, Thorsten Ottosen <nesotto@cs.auc.dk> wrote:
I think I found the error in ptr_set.cpp. It was caused by insertion into a multiset from itself. I've committe the new test file which works locally.
This test (ptr_set) is currently failing on a few different platforms with memory-access violation errors: Results Page: http://www.meta-comm.com/engineering/boost-regression/cvs-head/developer/ptr... Some failures (SunOS, OSF1, there is also Darwin and Intel Win32): http://tinyurl.com/ba8cv http://tinyurl.com/83fvf The test succeeds on my Linux machine when run standalone, but with valgrind --tool=memcheck there are errors detected. Please see the attached logfile. -- Caleb Epstein caleb dot epstein at gmail dot com
I'm resending this message since it may have been effectively lost as a response to an older thread. This test (ptr_set) is currently failing on a few different platforms with memory-access violation errors: Results Page: http://www.meta-comm.com/engineering/boost-regression/cvs-head/developer/ptr... Some failures (SunOS, OSF1, there is also Darwin and Intel Win32): http://tinyurl.com/ba8cv http://tinyurl.com/83fvf The test succeeds on my Linux machine when run standalone, but with valgrind --tool=memcheck there are errors detected. Please see the attached valgrind logfile. -- Caleb Epstein caleb dot epstein at gmail dot com
participants (3)
-
Caleb Epstein -
Martin Wille -
Thorsten Ottosen