885154 - run mozconfig comparison checks on android and thunderbird

Status: RESOLVED WONTFIX

(Release Engineering :: General, defect)

Description

[[e-mail] (:bhearsum)]

Noticed this while reviewing bug 763903. There's probably a reason we don't run it, but it's unfortunate because it means that we won't run mozconfig checks for android...

Comment 1

[Phil Ringnalda (:philor)]

Yeah, you'll want to do something else, maybe more like the way check-sync-dirs.py gets run (but not as many times as it does); I don't think there's one single bug for the hopeless impossibility of make check on Android (which would be "on Linux64" since we don't build on Android), just multiple bugs on things like "make it possible to run cppunittests packaged instead of from an objdir" and "run them packaged" and "make it possible to run jit-tests packaged".

Comment 2

[(not currently active) Ted Mielczarek]

Right, I suspect the only real reason here is "most of the stuff that runs in make check doesn't make sense to run on a cross-compile".

Comment 3

[[e-mail] (:bhearsum)]

Ah right, duh. Morphing this bug, then.

I'm not sure the check-sync-dirs model works for mozconfig checks though, because they've got application specific logic to figure out which mozconfigs to look at. Maybe there's some preflight target inside the Android makefiles that we can put these into.

Comment 4

[[e-mail] (:bhearsum)]

The only target I found that seems remotely reasonable is "tools" in mobile/android. Nothing runs as part of it right now. Ted, is that a reasonable place to call compare-mozconfigs.py?

Comment 5

[(not currently active) Ted Mielczarek]

I don't really have any good suggestions here, we don't have a place for this at the moment. You could explicitly add it as a dependency of "default" in the root Makefile, so that it'd get run at the top of the build.

Alternately, maybe we could just fix "make check" to work on cross-compiles? This would probably just involve making sure we don't try to run C++ unit tests or the JS tests.

Comment 6

[[e-mail] (:bhearsum)]

Additional question from IRC, so I don't forget:

[16:29:24] <bhearsum> ted: re: https://bugzilla.mozilla.org/show_bug.cgi?id=885154#c5 - do you mean making something like "compare-mozconfigs" a dependency of default in the root Makefile, and then implementing that target in browser/Makefile.in and mobile/android/Makefile.in?
[17:18:20] <ted> bhearsum|afk: that'd be fine yes

Comment 7

[[e-mail] (:bhearsum)]

I'm going to scope creep this a little bit, because it's important that Thunderbird runs them too. Thunderbird should be an easy port -- add the whitelist, and the target to run the checks.

Comment 8

[[e-mail] (:bhearsum)]

Gps, to avoid extra iteration (like in bug 763903), would you mind giving feedback on the approach talked about in comments #5 and 6? No rush.

Comment 9

[Gregory Szorc [:gps]]

I think injecting an extra rule into the default target in the root make file is acceptable.

Comment 10

[Jason Yeo[:jyeo]]

Attachment: compare_android_mozconfig.diff

I have a WIP patch. I'm not sure if I am doing the make file stuff right.

Also, I pushed to try and I got this: https://tbpl.mozilla.org/?tree=Try&rev=618ece10f608.

There were some errors when checking the mozconfigs. Is it because some entries are missing in the whitelist?

Comment 11

[[e-mail] (:bhearsum)]

Comment on attachment 787698 [details] [diff] [review]
compare_android_mozconfig.diff

Review of attachment 787698 [details] [diff] [review]:
-----------------------------------------------------------------

This patch looks like it's on the right track, but it's probably out of date now (my fault) :(. Needs some build peer feedback/review after it's been retested, too.