Merge lp://qastaging/~mir-team/mir/only-one-way-to-run-the-tests into lp://qastaging/mir
Status: | Work in progress |
---|---|
Proposed branch: | lp://qastaging/~mir-team/mir/only-one-way-to-run-the-tests |
Merge into: | lp://qastaging/mir |
Diff against target: |
617 lines (+43/-530) 3 files modified
cmake/MirCommon.cmake (+43/-115) cmake/src/mir/CMakeLists.txt (+0/-10) cmake/src/mir/mir_discover_gtest_tests.cpp (+0/-405) |
To merge this branch: | bzr merge lp://qastaging/~mir-team/mir/only-one-way-to-run-the-tests |
Related bugs: |
Reviewer | Review Type | Date Requested | Status |
---|---|---|---|
Alan Griffiths | Needs Resubmitting | ||
PS Jenkins bot (community) | continuous-integration | Approve | |
Alberto Aguirre (community) | Needs Information | ||
Alexandros Frantzis (community) | Needs Information | ||
Review via email:
|
Commit message
Drop DISABLE_
This was used to not run the test executables to discover their tests when
cross-building. But since we're cross-building we're not going to run the tests
*anyway*, so just don't add the test targets instead.
Description of the change
Drop DISABLE_
This was used to not run the test executables to discover their tests when
cross-building. But since we're cross-building we're not going to run the tests
*anyway*, so just don't add the test targets instead.
Whenever we need to modify the environment the tests run in this gets in our way. I don't know any reason for itother than “you can't run test discovery while cross-compiling”, and that can be better served by just not running test discovery while cross-compiling :).
Unmerged revisions
- 2465. By Chris Halse Rogers
-
Remove some unnecessary Valgrind fiddling
- 2464. By Chris Halse Rogers
-
Re-add the mistakenly-dropped TSan handling
- 2463. By Chris Halse Rogers
-
Drop GTest test discovery.
There's no particularly good reason to have two different ways of running the testsuite.
Drop the one that isn't 3x slower on armhf CI due to valgrind process startup inefficiencies.
We are currently using this option in CI runs to reduce the test runtime [1]. This has a significant effect, since it avoids the startup cost of new processes which is very high with valgrind (especially on armhf).
Do we actually need test discovery any more? IIRC, its goal was to get more information when tests failed in CI, but we can get enough information by using ctest in verbose mode instead of using test discovery, which is exactly what we have been doing for a long time in CI anyway.
I propose we move in the opposite direction and remove the test discovery mode.
Needs discussion/ Disapprove
[1] https:/ /bazaar. launchpad. net/~private- ps-quality- team/pbuilderje nkins/trunk/ view/head: /hooks/ H15enable_ testing. in