homepage

It'd be nice if there was a facility to force a particular test to be 
re-run in verbose mode with the most recent build via the buildbot web 
interface. It would allow us to get a bit more information about 
what's going on when the buildbots report stuff like this:
test_property
test_cprofile
test_signal
command timed out: 1800 seconds without output, killing pid 214235
SIGKILL failed to kill process
using fake rc=-1
program finished with exit code -1
Neal or Martin, if neither of you have the desire/bandwidth to look 
into this, I'm more than happy to take it on. Suspect it'd involve a 
buildbot change on dinsdale right?

I think this will be fairly difficult to set up. If the clean buildstep
had been executed, you would have to rerun configure and compile before
you can run any tests.
Also, how would you communicate what specific test you want to run?
A "builder" (a lane in the waterfall) has a fixed sequence of build
steps, where one is "test", and a step can be described with a shell
command. The test step simply invokes "make buildbottest". I don't know
whether it's possible to pass per-build parameters to each step (except
perhaps to the svn step); you could ask for that on the buildbot mailing
list.
The simplest solution to achieve the desired effect is to create a
branch in subversion where the buildbottest target does something
different, and then ask the slave in question to build that branch.
In any case, no, I won't be able to look into this further anytime soon.

> I think this will be fairly difficult to set up. If the clean buildstep
> had been executed, you would have to rerun configure and compile before
> you can run any tests.
We could re-order and do clean first. That would leave all the build
artifacts in tact after a build which would be nice for some
debugging.
> Also, how would you communicate what specific test you want to run?
I agree here. My guess is it would be pretty hard to modify the
buildbot to support this. I don't have bandwidth to help. It would
be nice to have, but probably not a high priority.

> We could re-order and do clean first. That would leave all the build
> artifacts in tact after a build which would be nice for some
> debugging.
With the current setup, that wouldn't quite work. We can't run it before
configure, because we might have no Makefile to invoke the clean target,
and we can't run it after configure, as we run "make distclean", which
deletes the makefile.

I'm closing this as "won't fix". If there is a need to run specific tests in specific ways, it is best to check in code that does it as desired. If it is then unacceptable to run that change on all slaves, a branch can be created, and building the branch can be triggered in the UI.