Merge lp:~bregma/geis/fix-integration-tests into lp:geis
Status: | Merged |
---|---|
Merged at revision: | 242 |
Proposed branch: | lp:~bregma/geis/fix-integration-tests |
Merge into: | lp:geis |
Diff against target: |
99 lines (+16/-19) 4 files modified
configure.ac (+1/-1) libutouch-geis/geis.c (+15/-9) libutouch-geis/geis_private.h (+0/-8) libutouch-geis/geis_subscription.c (+0/-1) |
To merge this branch: | bzr merge lp:~bregma/geis/fix-integration-tests |
Related bugs: |
Reviewer | Review Type | Date Requested | Status |
---|---|---|---|
Chase Douglas (community) | Approve | ||
Review via email: mp+100266@code.launchpad.net |
Description of the change
These changes bring utouch-geis to the point where the entire test suite runs green on my local machine, at least under the following conditions.
(1) No other X server grabs the input devices. Running from a text console (VT) prevents this. Failures due to this are not failures in the uTouch stack.
(2) The captive X server will not shut down and start rapidly. All tests pass if run separately or when built and run in the same pass, but some tests will not start if run too rapidly together. This is effectively a failure of the test harness and not indicative of problems in the uTouch stack.
The changes include here include (1) setting the appropriate libraries so the testsuite will build, and (2) changing the order of teardown of some contained objects to avoid both a circular dependency in refcounted objects that sometimes caused a resource loss (no X connection teardown, memory leaks) and also a double-delete or reference of as deleted object.
It all looks reasonable to me.
I think we may need some work on xorg-gtest to make it a bit more robust. There are circumstances where the X server doesn't always shut down, and perhaps it should be starting server on a random port that is unused.