I've just been investigating make check on Thunderbird builds and noticed that on Firefox trunk: - OS X 10.6.2 mozilla-central build -- Runs make -k check in objdir/i386 - OS X 10.5.2 mozilla-central leak test build -- Runs make -k check in objdir/i386 AFAICT in the 32/64 bit universal build cycle (on the 10.6.2 build), the make check executables are being built twice - once for 32 bit and once for 64 bit. Therefore, shouldn't the 10.6.2 builder be running make -k check against objdir/x86_64 so that we're testing the 64 bit version?

I just looked at the log for the latest "Firefox" Mac builds, and this is what I see: OS X 10.5.2 mozilla-central leak test build: make -k check in dir /builds/moz2_slave/mozilla-central-macosx-debug/build/obj-firefox OS X 10.6.2 mozilla-central build: make -k check in dir /builds/slave/mozilla-central-macosx64/build/obj-firefox/i386 (timeout 300 secs) OS X 10.6.2 mozilla-central leak test build make -k check in dir /builds/slave/mozilla-central-macosx64-debug/build/obj-firefox (timeout 300 secs) So it looks to me like the debug (leak test) builds are single-target, and run i386/x86_64 accordingly, while the universal opt build only runs make check on the i386 portion.

bhearsum helped me figure out that this is because platform_objdir is set here: http://hg.mozilla.org/build/buildbot-configs/file/513e4e263353/mozilla/config.py#l396 and then used here: http://hg.mozilla.org/build/buildbotcustom/annotate/ad10ae2a8e1a/process/factory.py#l975 Maybe we can just change that first line and then everything will work!?!?

(In reply to comment #2) > bhearsum helped me figure out that this is because platform_objdir is set here: > > http://hg.mozilla.org/build/buildbot-configs/file/513e4e263353/mozilla/config.py#l396 > and then used here: > http://hg.mozilla.org/build/buildbotcustom/annotate/ad10ae2a8e1a/process/factory.py#l975 > > Maybe we can just change that first line and then everything will work!?!? I suspect that won't work, because you'll then end up trying to the universal packaging in the x86_64 side of the build, when it expects to do it in the i386 side... I've not tested that assertion of course, but I'm pretty sure we had that issue when we started getting our initial 32/64 bit universals working on Thunderbird try server.

Just re-checked this, and it is an issue on the newer 10.7 builders.

Maybe we can say this is a feature, if we're testing 64bit on debug and 32bit on opt ?

I'd rather fix bug 992323 than spend any time on this. We'll eventually be able to stop doing universal mac builds once we unsupport enough OS X versions.

This is not worth fixing not that the 32-bit side is going away.