now that bug 523402 has landed, the next step is to generate the mobile specific test package. To do this follow these steps: 1) do normal build such as make -f client.mk build 2) cd to $(objdir)/mobile <- not the xulrunner objdir 3) cd mobile (i.e. objdir/mobile/mobile/) 4) 'make package-mobile-tests' 5) cd $(objdir)/mobile/dist 6) see fennec.*tests.tar.bz2 file if you could upload this to the location where the xulrunner tests.tar.bz2 file is located at that would help. In addition once this is done, we will want to run this on mobile tinderbox.

Mass move of bugs from Release Engineering:Future -> Release Engineering. See http://coop.deadsquid.com/2010/02/kiss-the-future-goodbye/ for more details.

Noting here, for the record: We'd love to see the test packages consolidated. I'm assuming this will be a "make package-tests" in objdir/mobile/... that grabs the xulrunner tests + fennec tests and creates a single fennec.*tests.tar.bz2 file. This feels like the right fix.

this is sort of a hack, but a solution that builds a single test package. To make it, do this: * cd objdir/mobile/mobile * make package-mobile-tests * cd objdir/mobile/dist * have a party with fennec-1.1a2pre.en-US.linux-i686.tests.tar.bz2 My concerns with this approach are: * if we change our directory structure to a single objdir, then this fails * if the original package-tests change much this could fail * there is no a11y and it fails with my chrome .jar packaging goop (another patch I am working on) Ideally I would like to see this bug morph into where we could do a: 'package-mobile-tests: make-stage-dir stage-mochitest stage-reftest stage-xpcshell stage-jstests' This would require some magic where I have those targets know they need to go to $(DEPTH)/../xulrunner/dist before running. The good thing about this patch is it only touches the mobile Makefile and no other Makefiles in the system.

I think I'd rather fix the mobile build so it's not doing a two-pass build with xulrunner+mobile, and instead just a single "mobile" build. Then this bug should just magically resolve itself. blassey patched this so it mostly works, it probably just needs some tweaking.

my understanding is we want both the single objdir and the dual objdir scenarios. I agree that having just the single objdir would make life easier!

Yes, please!

I'm just saying that I don't think we should spend a lot of time trying to make things work in the dual-objdir scenario. Everything is going to be a special-case hack there, and I don't think it's worth the time to fix each thing specially.

mfinkle or blassey, can you guys comment on the direction of a single objdir?

My patch in bug 557027 makes the single objdir solution work everywhere. I filed bug 557201 on switching our tinderbox builds to use it. Once we do that this bug should be WFM.

will we ignore the firefox browser-chrome tests as well?

If they're all in browser/, we won't ever copy them, since the Fennec build doesn't go there.

adding the patch for a single objdir and adjusting my mozconfig, I generate a single objdir successfully, but I end up with a handful of browser_*.js tests in browser-chrome that shouldn't be there.

I guess we should figure out where they come from and fix them. If they really are testing Firefox-specific functionality, then we should make them Firefox-specific.

with the landing of the single objdir, now we have a pile of browser-chrome tests from firefox that end up in the mochitest/browser directory. This patch fixes the makefiles to not include the browser chrome tests and files unless the app is 'browser'

update for bitrot.

Comment on attachment 443917 [details] [diff] [review] mobile-browser patch (2.1) This sucks, but short of "move all the tests to browser/" I don't have a better idea. It'd be interesting to see how many of these tests really do need to be browser-chrome tests, or if some of them could function as chrome mochitests.

updated patch for bitrot, verified clean pass on try server

checked in as d9dc6ee9190c

Fixed pymake bustage from this patch in http://hg.mozilla.org/mozilla-central/rev/8743dc102574

Comment on attachment 446197 [details] [diff] [review] mobile-browser patch (2.2) This has caused some tests to fail when run on non-Firefox browsers. >diff --git a/toolkit/content/tests/Makefile.in b/toolkit/content/tests/Makefile.in >--- a/toolkit/content/tests/Makefile.in >+++ b/toolkit/content/tests/Makefile.in >@@ -42,11 +42,15 @@ srcdir = @srcdir@ > VPATH = @srcdir@ > > include $(DEPTH)/config/autoconf.mk > > MODULE = test_toolkit_general > > XPCSHELL_TESTS = unit > >-DIRS = browser chrome widgets >+DIRS = chrome widgets >+ >+ifeq (browser,$(MOZ_BUILD_APP)) >+DIRS += browser >+endif For instance, the tests for bug 536567 and bug 500885 depend on the file "mockObjects.js" which is installed by toolkit/content/tests/browser/common.

I now see that you already filed bug 568733 about that. Sorry for the spam.

OK, so this patch disabled tests in at least docshell and dom when it landed. Please look over it carefully to see whether it happened to disable any other tests...

(In reply to comment #22) > OK, so this patch disabled tests in at least docshell and dom when it landed. > Please look over it carefully to see whether it happened to disable any other > tests... It may have, but I can verify that it is not the case now, MXR tells me we only ifneq mobile,... rather than ifeq browser,...

...and we don't have the s/DIRS/dirs/ problem other than what the dom patch (http://hg.mozilla.org/mozilla-central/rev/a9a1bbd4176f) fixed.