Ticket #8357 (closed: fixed)
Clear memory leaks in the APITest package
Reported by: | Russell Taylor | Owned by: | Russell Taylor |
---|---|---|---|
Priority: | major | Milestone: | Release 3.2 |
Component: | Framework | Keywords: | Maintenance |
Cc: | Blocked By: | ||
Blocking: | #8361 | Tester: | Martyn Gigg |
Description (last modified by Russell Taylor) (diff)
The SNS Jenkins has a valgrind job which uses the unit tests as a hook into exercising the Mantid code itself. This of course means that it picks up any leaks in the test code as well.
Using this job and/or valgrind runs of your own, eliminate all the leaks/errors that come out of the test code itself. I've created this as an isolated ticket because I'd like to get a clear view of the errors we have in the actual Mantid code. All changes to this ticket should only be to code in the Framework/API/test/ directory!
When working on this, don't just add delete calls - consider whether the leaking call really needs to be a heap allocation; create the object on the stack if not.
Change History
comment:3 Changed 7 years ago by Russell Taylor
- Owner set to Russell Taylor
- Milestone changed from Backlog to Release 3.2
comment:4 Changed 7 years ago by Nick Draper
- Status changed from new to assigned
Bulk move of tickets out of triage (new) to assigned at the introduction of the triage state
comment:5 Changed 7 years ago by Russell Taylor
- Status changed from assigned to inprogress
Re #8357. Eliminate valgrind invalid write warning.
By making sure to unregister the observer at the end of the test.
Changeset: 845efec319cf1cd719cee923ad07370e65bf1885
comment:6 Changed 7 years ago by Russell Taylor
Re #8357. Fix leak in test by breaking cyclic reference.
The BoxController & BoxControllerDummyIO objects were each holding a shared_ptr back to the other.
Changeset: 72fed1a43f3622555347e63395766e3204c62082
comment:7 Changed 7 years ago by Russell Taylor
Re #8357. Fix memory leaks in tests.
By creating object on the stack instead of unnecessarily on the heap.
Changeset: c02e1bdf04ba6eacc4b14bdfbca7da0d932b9570
comment:8 Changed 7 years ago by Russell Taylor
Re #8357. Remove leak in test by adding components to instrument.
The fact that you have to call two methods in succession to get this right is probably not great design. Nor is the fact that you can 'mark' something that isn't in the instrument without any error or warning.
Changeset: 4b45a3284795ff43623134ad4fcfedfe1fb9da0e
comment:9 Changed 7 years ago by Russell Taylor
Re #8357. Eliminate memory leaks in tests.
By calling delete where appropriate.
Changeset: a295915fdba69374fbc2a04490c41b80896b1e46
comment:10 Changed 7 years ago by Russell Taylor
Re #8357. Eliminate memory leak in test.
By removing group from within group to eliminate cyclic shared_ptr reference. Possibly there's a real potential issue here, but would anyone really ever put a group within a group?
Changeset: 9533f817885addf622af52987b94f1908f386d1c
comment:11 Changed 7 years ago by Russell Taylor
Re #8357. Eliminate valgrind invalid write warning.
By making sure to unregister the observer at the end of the test.
Changeset: 80f3695a539a3a67ace1becf8091636ec88ce73d
comment:12 Changed 7 years ago by Russell Taylor
Re #8357. Fix leak in test by breaking cyclic reference.
The BoxController & BoxControllerDummyIO objects were each holding a shared_ptr back to the other.
Changeset: 7053b8961e800d64e54c51bc15c89341d0845e94
comment:13 Changed 7 years ago by Russell Taylor
Re #8357. Fix memory leaks in tests.
By creating object on the stack instead of unnecessarily on the heap.
Changeset: d97c3c8c89110d496960c9af431d5f64847693d1
comment:14 Changed 7 years ago by Russell Taylor
Re #8357. Remove leak in test by adding components to instrument.
The fact that you have to call two methods in succession to get this right is probably not great design. Nor is the fact that you can 'mark' something that isn't in the instrument without any error or warning.
Changeset: da65363bf7f7ab8d0b8789648b5b0ed5b8faa66d
comment:15 Changed 7 years ago by Russell Taylor
Re #8357. Eliminate memory leaks in tests.
By calling delete where appropriate.
Changeset: 067e4473293d2462a4caa61273494c83ffbd8005
comment:16 Changed 7 years ago by Russell Taylor
Re #8357. Eliminate memory leak in test.
By removing group from within group to eliminate cyclic shared_ptr reference. Possibly there's a real potential issue here, but would anyone really ever put a group within a group?
Changeset: ee89f1bb5030406f4c1859dddbe9b2b05391e74f
comment:17 Changed 7 years ago by Russell Taylor
Re #8357. Update to account for change in TestHelpers class.
The BoxControllerDummyIO constructor was changed to accept (and hold internally) a bare pointer to avoid a cyclic reference.
Changeset: 981868bdda2c8ae0fdb3e9778f1ce9b0476bb297
comment:18 Changed 7 years ago by Russell Taylor
Re #8357. Add a suppressions file for the API tests.
Valgrind is complaining about an OpenMP call in AlgorithmManagerTest. I don't know whether this is a real leak, but I suppose that there's not much we can do about it.
Changeset: 3058ccfae1cb69a4f53664df2cb91053504be900
comment:19 Changed 7 years ago by Russell Taylor
Re #8357. Fix memory leaks in test.
Ensure Property objects returned by loadProperty are deleted after use.
Changeset: 0de7d7e83ee1e436cfffc5089a50829e345d00bf
comment:20 Changed 7 years ago by Russell Taylor
- Status changed from inprogress to verify
- Resolution set to fixed
To test: This mainly needs to be tested by inspection. The way I can tell that all leaks in API/test itself are gone is because I run valgrind in Eclipse and observe the absence of a little red cross next to that directory.
comment:21 Changed 7 years ago by Martyn Gigg
- Status changed from verify to verifying
- Tester set to Martyn Gigg
comment:22 Changed 7 years ago by Martyn Gigg
- Status changed from verifying to reopened
- Resolution fixed deleted
I ran the APITest suite through valgrind in eclipse and I'm showing up an error in MemoryManagerTest.h:37, which uses an openmp loop. The stacktrace is:
==19901== 2,352 bytes in 7 blocks are possibly lost in loss record 36 of 92 ==19901== at 0x4C29DB4: calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==19901== by 0x4012074: _dl_allocate_tls (dl-tls.c:297) ==19901== by 0x7B6BABC: pthread_create@@GLIBC_2.2.5 (allocatestack.c:571) ==19901== by 0x774731B: ??? (in /usr/lib/x86_64-linux-gnu/libgomp.so.1.0.0) ==19901== by 0x86299F: _ZN17MemoryManagerTest13test_parallelEv (MemoryManagerTest.h:37) ==19901== by 0x862CD1: _ZN53TestDescription_suite_MemoryManagerTest_test_parallel7runTestEv (MemoryManagerTest.cpp:38) ==19901== by 0x9B6AB4: _ZN7CxxTest19RealTestDescription3runEv (RealDescriptions.cpp:96) ==19901== by 0x9B9FDE: _ZN7CxxTest10TestRunner7runTestERNS_15TestDescriptionE (TestRunner.h:94) ==19901== by 0x9B9ED0: _ZN7CxxTest10TestRunner8runSuiteERNS_16SuiteDescriptionE (TestRunner.h:79) ==19901== by 0x9B9D6B: _ZN7CxxTest10TestRunner8runWorldEv (TestRunner.h:63) ==19901== by 0x9B9B2E: _ZN7CxxTest10TestRunner11runAllTestsERNS_12TestListenerE (TestRunner.h:24) ==19901== by 0x9BFB6D: _ZN7CxxTest12XUnitPrinter3runEv (XUnitPrinter.h:30)
It looks like something that just needs a similar suppression to the one already present for AlgorithmManager?
comment:23 Changed 7 years ago by Russell Taylor
- Status changed from reopened to inprogress
Re #8357. Add suppression for another use of OpenMP in a test.
Changeset: 111f66ab17766b1392a4ac2f6739de9fd18fa325
comment:24 Changed 7 years ago by Russell Taylor
- Status changed from inprogress to verify
- Resolution set to fixed
That last one didn't show up for me (or in the Jenkins job) when running the entire API suite, though it did when I ran the MemoryManagerTest on its own. Anyhow, it's suppressed now.
comment:26 Changed 7 years ago by Martyn Gigg
The API test folder is now clear for me too.
comment:27 Changed 7 years ago by Martyn Gigg
- Status changed from verifying to closed
Merge remote-tracking branch 'origin/bugfix/8357_api_test_memory_leaks'
Full changeset: 4e89f08a33013f4b0628c312f176d500225ea2a4
comment:28 Changed 5 years ago by Stuart Campbell
This ticket has been transferred to github issue 9202