388169 - Changing nsIPrintSettings.idl uuid causes printing reftest failures on win32 firefox unit test tinderboxes

Status: NEW

(Testing :: Reftest, defect, major)

Description

[Kenneth Herron]

Changing the UUID in widget/public/nsIPrintSettings.idl causes the printing reftests to fail on the two win32 firefox unit test tinderboxen (qm-winxp01 and qm-win2k3-01). The test failures are persistant as long as the change is left in. Backing out the change causes the tests to succeed again. The specific test failures are:

REFTEST UNEXPECTED FAIL (LOADING): file:///C:/slave/trunk_2k3/mozilla/layout/reftests/printing/blank.html
REFTEST UNEXPECTED FAIL (LOADING): file:///C:/slave/trunk_2k3/mozilla/layout/reftests/printing/272830-1.html

I first experienced this trying to check in a patch for bug 381631, and again trying to check in a patch for bug 315687. Eli Friedman was able to reproduce the effect checking in a uuid change by itself (see the sample url).

I've examined some of the relevant build logs, but didn't see an obvious explanation for the test failures. I don't have access to a win32 build environment, so I can't reproduce the problem.

CVS log for nsIPrintSettings.idl (to cross-reference checkin times):
http://bonsai.mozilla.org/cvslog.cgi?file=/mozilla/widget/public/nsIPrintSettings.idl

My first attempt to patch bug 381631:
http://tinderbox.mozilla.org/showbuilds.cgi?tree=Firefox&maxdate=1182710000

My second attempt to patch bug 381631 and Eli's test:
http://tinderbox.mozilla.org/showbuilds.cgi?tree=Firefox&maxdate=1183242000

My first attempt to patch bug 315687:
http://tinderbox.mozilla.org/showbuilds.cgi?tree=Firefox&maxdate=1184443447

Comment 1

[Kenneth Herron]

For a new example, see <http://tinderbox.mozilla.org/showbuilds.cgi?tree=Firefox&maxdate=1184522370>. 

This morning, I tried to patch bug 315687 again. For some reason, this triggered printing reftest failures only on qm-win2k3-01; qm-winxp01 stayed green. When I backed out the patch--reverting the uuid--qm-win2k3-01 went green, but the printing tests started failing on qm-winxp01.

I tried clobbering in conjunction with my checkin, but it doesn't seem to be working on the unit test boxes. See bug 388223.

Comment 2

[Eli Friedman]

kherron: would you like me to check in sometime when people are around in #build?

Comment 3

[Kenneth Herron]

Eli, if you want to do it, that would be fine by me. I probably couldn't try again until this weekend. Note Gavin indicated that only certain people could clobber the unit test systems.

Comment 4

[Reed Loden [:reed] (use needinfo?)]

This bug bit us again when I landed bug 193001. :(

Comment 5

[(not currently active) Ted Mielczarek]

I don't know if this is a bug in the reftest harness or a RelEng bug (does a clobber fix it) or what.