417511 - Sisyphus/JavaScript Tests should clobber on existence of clobber file

Status: RESOLVED WONTFIX

(Testing Graveyard :: Sisyphus, defect)

Description

[Bob Clary [:bc] (inactive)]

It is safest for the browser tests, if they were run using clobber builds. While this would add additional time to the build, it is not that much of an impact considering these tests are not run as often. The certainty that there are no depends build issues in the test results outweighs any time consideration imo. The shell builds are actually clobbered each time by default anyway.

The change might be as simple as changing the instances of -B "checkout build" to -B "clean checkout build" in the runtests.sh steps however for a tree that has not yet been checked out, this will fail since it currently does a make -f client.mk clean in an existing tree. I could change the clean step to just rm -fR the firefox-(opt|debug) obj dirs which would not fail even in the event that the tree wasn't checked out. robcee, is an rm of the objdirs sufficient? It's certainly faster.

Comment 1

[:Gavin Sharp [email: [e-mail]]]

Do dependency problems often affect the JS tests in-browser? It doesn't seem to be a problem for the existing unit test machines.

Comment 2

[Bob Clary [:bc] (inactive)]

Its not a known issue, but considering I am only getting one build/test with the browsers per day I don't want to waste it on potential dependency problems.

Comment 3

[Bob Clary [:bc] (inactive)]

fyi, firefox win32 1.8.1 debug build failed this morning linking jsj3250.dll. I manually deleted the object directories and started a new build.

Comment 4

[Rob Campbell [:rc] (:robcee)]

this came up in triage today. do we need to do anything here, Bob?

Comment 5

[Bob Clary [:bc] (inactive)]

I'm not sure. I've been doing ok without clobber builds for the most part, so my inclination is to leave this alone. The actual fix would be trivial however and could be done with a quick edit to change the occurrence of "checkout ..." to "clean checkout ...". It might be better if this was controllable ala the way you guys do in the build environment. In that case, I think a file is checked in that will force a clobber when it is present. I could just as easily add something to sisyphus that would do a clobber in the presence of a file which could even be manually pushed to the machines where a clobber is desired. 

How about we morph this bug to allow the forcing of a clobber by pushing a file out to the buildbot slave. You can give the bug to me then. ok?

Comment 6

[Rob Campbell [:rc] (:robcee)]

yeah, that works for me and will probably be better for everybody. The bug is yours!

Comment 7

[(not currently active) Ted Mielczarek]

Mass move of Sisyphus bugs to Testing : Sisyphus. Filter on SisyphusMassMove to ignore.