NotLinkingJiraYet fail test if subprocess aborts or segfaults #1112
Conversation
|
just a note to myself, two segfaults reported by ctest without any details or backtraces https://travis-ci.com/github/apache/qpid-dispatch/jobs/497981423#L4673 |
|
The problem with this is that the router not always segfaults or aborts, sometimes it just exits. Some tests do expect a !=0 ecode exit. The code has to look for expected ecode in the event handler, compare those two, and somehow print output (so it does not get lost) and either kill the test, or continue from the next testcase with restored routers. |
jiridanek
force-pushed
the
jd_2021_04_08_detect_dead_child
branch
2 times, most recently
from
9ca5b63 to
d96009d
Compare
TODO: * does this actually do the right thing? * is it necessary to instrument `unordered_move_stack` function? When there is a read or write to the poisoned area, the error message looks like ``` ==15792==ERROR: AddressSanitizer: use-after-poison on address 0x611000034dd8 at pc 0x7fdaa75fc713 bp 0x7fff2d0c8d80 sp 0x7fff2d0c8d78 14: WRITE of size 8 at 0x611000034dd8 thread T0 14: #0 0x7fdaa75fc712 in qd_hash_internal_remove_item /home/jdanek/repos/qpid/qpid-dispatch/cmake-build-debug-asan/../src/hash.c:131:30 14: #1 0x7fdaa75fb51d in qd_hash_free /home/jdanek/repos/qpid/qpid-dispatch/cmake-build-debug-asan/../src/hash.c:146:13 [...] ```
jiridanek
force-pushed
the
jd_2021_04_08_detect_dead_child
branch
from
d96009d to
0bb3c7a
Compare
|
I was thinking about few more things. First, reporting/dealing with leaked threads, https://issues.apache.org/jira/browse/DISPATCH-2323?focusedCommentId=17484381&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-17484381 and with adding rr --chaos, https://groups.google.com/g/mozilla.dev.platform/c/wJJUkf0GESE, which requires dealing with rr's inability (or unwillingness) to forward signals it receives to the inferior process under debugging. Solving this does not sound hard and should be interesting. (Btw, I first saw https://issues.apache.org/jira/browse/DISPATCH-2318 under rr chaos) |
No description provided.