{ http://tinderbox.mozilla.org/showlog.cgi?log=SeaMonkey-Ports/1241902785.1241910813.3688.gz WINNT 5.2 comm-central unit test on 2009/05/09 13:59:45 REFTEST INFO | runreftest.py | Running tests: start. INFO | (automation.py) | Application pid: 4092 ### XPCOM_MEM_BLOAT_LOG defined -- logging bloat/leaks to c:\docume~1\seabld\locals~1\temp\1\tmpglqhsf\runreftest_leaks.log REFTEST TEST-UNEXPECTED-FAIL | | EXCEPTION: ReferenceError: gAutoconfVars is not defined REFTEST FINISHED: Slowest test took 0ms (undefined) } { [kairo@kairo.at - 2009/05/09 16:25:16] Not sure why and where it comes from, but this seems to be a common error on fresh build slaves of all platforms (others had it before being switched to SeaMonkey-Ports). Goes away on further cycles from the same machine. } Bug 420084 removed the involved code, but it's not going to land on 1.9.1 :-|

(In reply to comment #0) > build slaves of all platforms > (others had it before being > switched to SeaMonkey-Ports). Simply confirming this: { http://tinderbox.mozilla.org/showlog.cgi?log=SeaMonkey/1243629373.1243636856.15225.gz Linux comm-central unit test on 2009/05/29 13:36:13 }

http://tinderbox.mozilla.org/showlog.cgi?log=SeaMonkey2.0/1247222100.1247232546.19688.gz WINNT 5.2 comm-1.9.1 unit test on 2009/07/10 03:35:00

I had never had this error on my local Windows 2000, afair. Today, I tried twice to (clobber) build (optimized) Firefox 3.5 [which I seldom do], and I'm getting the error. This is (now) major because it prevents me from verifying that reftests/crashtests work/pass :-/ (which is precisely what I wanted to do for bug 483062.) I'd love to figure out what is going on! From what I checked, the autoconf.js file is in reftest.jar and contains the gAutoconfVars var definition as expected. I have no clue about what would be causing this error. David, any idea? NB: Next time I see it on a tinderbox, I'll try to download the build and compare it with a working one... (It happens regularly on SeaMonkey 2.0 boxes...) Until then, I'll save my local build so I can further test with them if need be.

(In reply to comment #3) > From what I checked, the autoconf.js file is in reftest.jar and contains the > gAutoconfVars var definition as expected. That was not entirely exact :-< I don't know why it happens, but I found what is wrong with autoconf.js: My 2 local Windows Firefox 3.5 builds: *objdir/layout/tools/reftest: 12 KB, content is right :-) *objdir/_tests/reftest/reftest/chrome/reftest.jar/content: 22 B, "var gAutoconfVars = {\n" only :-( > Next time I see it on a tinderbox, I'll try to download the build and compare > it with a working one... (It happens regularly on SeaMonkey 2.0 boxes...) It just happened :-> http://tinderbox.mozilla.org/showlog.cgi?tree=SeaMonkey2.0&errorparser=unittest&logfile=1251413374.1251421824.19742.gz&buildtime=1251413374&buildname=WINNT%205.2%20comm-1.9.1%20unit%20test&fulltext=1 WINNT 5.2 comm-1.9.1 unit test on 2009/08/27 15:49:34 *log: seems normal (autoconf.js: all lines and no error). *reftest.jar: 12 KB, but misses ending "};\n" line :-( http://tinderbox.mozilla.org/showlog.cgi?tree=SeaMonkey2.0&errorparser=unittest&logfile=1251416767.1251423861.10512.gz&buildtime=1251416767&buildname=WINNT%205.2%20comm-1.9.1%20unit%20test&fulltext=1 WINNT 5.2 comm-1.9.1 unit test on 2009/08/27 16:46:07 *log: seems normal (autoconf.js: all lines and no error). *reftest.jar: 22 B, "var gAutoconfVars = {\n" only :-( http://tinderbox.mozilla.org/showlog.cgi?tree=SeaMonkey2.0&errorparser=unittest&logfile=1251420427.1251426530.9754.gz&buildtime=1251420427&buildname=Linux%20comm-1.9.1%20unit%20test&fulltext=1 Linux comm-1.9.1 unit test on 2009/08/27 17:47:07 *log: seems normal (autoconf.js: all lines and no error). *reftest.jar: 15 KB, but misses ending "};\n" line :-( There is obviously something going wrong when adding autoconf.js to reftest.jar :-(

http://tinderbox.mozilla.org/showlog.cgi?log=SeaMonkey2.0/1252022081.1252031669.17345.gz WINNT 5.2 comm-1.9.1 unit test on 2009/09/03 16:54:41

http://tinderbox.mozilla.org/showlog.cgi?log=SeaMonkey2.0/1260156570.1260167528.31236.gz WINNT 5.2 comm-1.9.1 unit test on 2009/12/06 19:29:30

Does this happen only with make -j or -jN ? If so, maybe the build step that puts the file in the jar is running in parallel with the generation of the file?

(In reply to comment #7) That may be the cause: I just noticed that, for 'unittest' builds, SM 2.0 uses |mk_add_options MOZ_MAKE_FLAGS="-j3"|, whereas FF 3.5 doesn't use this option. Would it be possible to make this building task "not parallel"?

(In reply to comment #8) > (In reply to comment #7) > > That may be the cause: > I just noticed that, for 'unittest' builds, > SM 2.0 uses |mk_add_options MOZ_MAKE_FLAGS="-j3"|, > whereas FF 3.5 doesn't use this option. > > Would it be possible to make this building task "not parallel"? That might help debug the problem, but if it's the cause, wouldn't help in getting a real issue fixed (esp. as more and more developers ware compiling with parallel building options. Also, it would probably slow things down, which is not really helpful in our resource-strained situation.

(In reply to comment #9) > > Would it be possible to make this building task "not parallel"? I wasn't clear enough: I meant to "work around" '-j' for the file generation task... (Not to drop '-j3' from SeaMonkey build. Of course, we agree what that would do.)

This was happening "rarely" on our SM 2.0 'unit test' build, recently we switched to packaged 'crashtest' and 'reftest' tests runs, and it is now happening quite often :-/ I don't know what (change) caused this increase exactly, but it's even more visible as it happens on both tree columns at once. So a solution/workaround is more needed than ever...

Should this do it?

Comment on attachment 429400 [details] [diff] [review] (Av1) Add .NOTPARALLEL to ensure autoconf.js is fully generated before proceeding This is a sledgehammer approach. Just make the "make-xpi" target depend on autoconf.js, like: make-xpi: autoconf.js

Per your comment 13 ;-)

Comment on attachment 430649 [details] [diff] [review] (Av2) Make "make-xpi" target depend on autoconf.js [Checkin: Comment 15] http://hg.mozilla.org/releases/mozilla-1.9.1/rev/3151f58fac89

So many oranges for such a simple fix :-/