On Apr 12, 2007, at 10:25 AM, Doug Cutting wrote:
Nigel,
Something I committed yesterday seems to be causing a unit test in
trunk to fail (TestCheckpoint). I'm not sure how this happened. I
thought I made sure that each patch passed unit tests before I
committed it.
It's a non-deterministic failure.
Our Hudson patch build assumes that trunk is unbroken. When it's
broken, all newly submitted patches are incorrectly given -1.
But, you gotta admit that it gets peoples attention :-)
Is there any way to easily fix this? The dumb way would be to run
tests twice each time, once without the patch and once with it. A
better approach might be to trigger a build for each commit, and
explicitly keep track of whether trunk is broken, not running new
patch builds until trunk is fixed. But I don't know if that's
possible.
Or we just suspend the patch build when we discover trunk is broken.
Best would be if trunk was never broken...
+1. As Owen has advocated in the past, perhaps the patch process
should only +1 when 2 (or 3?) test runs ALL pass, instead of the
current state where just 1 of 2 attempts need to pass.
Nige
