mozilla :: #build

15 Mar 2017
00:03rillianbrson's RFC suggests `overflow-checks = true` in the Cargo.toml's profile section.
00:16froydnjrillian: acrichto outlined a plan for adding support to cargo that I was going to try and tackle this week
00:16rillianfroydnj: is that a general-flags option?
00:17froydnjrillian: no, an `overflow-checks = true`; acrichto sounded like general flags was...difficult
00:20rillianmaybe general -C options, since they're supposed to be stable?
00:21rillianLooks like we can set RUSTFLAGS in the environment and cargo will propagate that. Not portable though.
00:21rillianI still can't figure out why profile.release.debug-assertions=true isn't working in my test crate
00:22froydnjis that test crate part of libxul? or something else?
00:23rillianI wrote a test thing to try and answer my question about Cargo profiles overriding settings in dependent crates.
00:26rillianfroydnj: compare 'cargo test' can 'cargo test --release' in https://github.com/rillian/cargo-override-test
00:27rillianthen compare 'cargo test --release' and 'RUSTFLAGS="-C overflow-checks=on" cargo test --release'
00:27rillian(with nightly rust)
00:35rillianI think cargo isn't passing -C debug-assertions
00:35tedfroydnj: https://bugzilla.mozilla.org/show_bug.cgi?id=1286294 FYI
00:35firebotBug 1286294 NEW, nobody@mozilla.org xpcshell invocation during startupcache precompile does not symbolicate crash stacks
00:38glandiumted: I'd rather kill the startupcache
00:44tedglandium: we were having this discussion just recently (in re: the cross-mac builds not having it)
00:44glandiumted: I know
00:47rillianted: I'm not sure what you mean about round-tripping your patches in bug 1346897?
00:47firebothttps://bugzil.la/1346897 NEW, giles@thaumas.net Apply 2-character indents to tooltool manifests
15:18makHi, today I keep hitting this build failure (no issues yesterday), a reboot doesn't help, removing AV doesn't help, reinstalling MB doesn't help. Any ideas?
15:19mak 9:38.45 0 [main] xargs 8784 fork_copy: linked dll data/bss pass 0 failed, 0xE3B000..0xF1886C, done 851968, windows pid 2052, Win32 error 299
15:19mak 9:38.45 xargs: cannot fork: Resource temporarily unavailable
15:19mak 9:38.46 NSModules are not ordered appropriately
15:23tedugh
15:40Callek_cloud9mshal: btw, your bug for repackage is https://bugzilla.mozilla.org/show_bug.cgi?id=1347576
15:40firebotBug 1347576 NEW, mshal@mozilla.com Create ./mach repackage that takes a tarball and returns a .dmg to support OSX signing
15:49mshalCallek_cloud9: cool, thanks!
19:01tedwe should upload config.status from every build so i can more easily figure out weird build configuration things
19:03mshalhah I was thinking the same thing :)
19:03mshalwhat did you need it for?
19:04tedi'm poking at this perf regression on the taskcluster mac builds again
19:04tedjust wanted to sanity check some assumptions
19:04tedi'm not entirely sure that config.status would have all the answers, though
19:05tedthe other thing i wanted to sanity check was "are both builds building the same set of source files" but that's really hard to tell with unified sources
19:07mshalis a try push sufficient? I wonder if you could just change the package to be a tarball of the objdir, and then grab that to poke around
19:08tedyeah, should be
19:08tedthat's a pretty good idea :)
19:09tedaltho it's gonna be huge if i tar up the whole objdir
19:09tedi guess that's someone else's problem
19:09mshalhaha
19:09ted*i've* got enough bandwidth and local disk
19:10froydnjuploading config.status and/or config.log would be helpful
19:10tedfroydnj: maybe you have thoughts here
19:11tedon my specific problem
19:11tedfroydnj: so comparing talos numbers between the buildbot and taskcluster opt mac builds
19:12tedthere's a regression in the tc builds, and it looks like the regression is primarily in some devtools tests, and those tests do a bunch of IPC messaging
19:12tedso! my current best hypothesis is that something in the IPC code is not getting built right in the tc builds
19:12tedthe tc builds are cross-compiles from linux, as you likely know
19:13froydnjinteresting
19:13tedIIRC we do use mach messaging for IPC on OSX, so i'm wondering if we are failing some configure test or something odd like that
19:15froydnjI'm coming up blank on weird things in the IPC stack that might be affected by cross-compiling
19:16tedokay
19:17tedhttps://dxr.mozilla.org/mozilla-central/source/ipc/glue/moz.build#107
19:17tedlike, i'm pretty confident we don't have the wrong value for OS_ARCH
19:20froydnjcould confirm that file is getting compiled via logs, but yeah, if OS_ARCH et al were broken, seems like lots of things would break
19:21froydnjmaybe, just maybe things like this: http://dxr.mozilla.org/mozilla-central/source/ipc/glue/SharedMemoryBasic.h#12 are busted?
19:21froydnjgetting the wrong shared memory implementation would hurt, but I wouldn't expect it to hurt devtools, more like graphics stuff
19:22froydnjah, compiling the mach version, ok, that theory's out
19:26tedyeah, i mean, if XP_DARWIN is not set the build ought to be super broken
19:27tedlike, i don't think the build is *broken*
19:27tedwcosta got tests running on the debug builds and they were very green
19:27tedwe only ran into one issue we could not figure out, and we wallpapered over it by increasing a timeout
19:28tedthe gtests were timing out, specifically there were some deadlock detector death tests that all took like an extra 90 seconds each
19:32froydnjted: one possibility is that we're not compiling with the same libc++ on both versions, and that's influencing the numbers somehow, since various parts of the ipc like std:: stuff
19:32tedhm!
19:33tedthat is a plausible scenario
19:33tedwe're still using the 10.7 SDK (): https://dxr.mozilla.org/mozilla-central/source/browser/config/tooltool-manifests/macosx64/cross-releng.manifest#19
19:34tedbut i think that should be the same for both builds
19:34froydnjwell
19:35froydnjI think we compile libc++ when we compile the clang we use
19:35froydnjand clang might use that in preference to other things
19:35tedah
19:35froydnjthere was also some weirdness that the 10.7 SDK is *not* the same as what you get when you use xcode
19:35tedwell, at first pass it turned out we were using clang 3.9 for the tc builds and 3.8 for the buildbot builds
19:35tedbut wcosta did a try push with 3.8 for both and the regression was still there
19:36tedfroydnj: oh yeah?
19:37tedi might just break down and do a try push that packages the entire objdir like mshal suggested
19:37tedso i can look for differences in generated files etc
19:44rillianmshal: did you want to review the updated tooltool manifest indent patch in bug 1346897, or should I carry forward your r+?
19:44firebothttps://bugzil.la/1346897 NEW, giles@thaumas.net Apply 2-character indents to tooltool manifests
19:45mshalrillian: nah if it's just the new files, r+
19:45rillianmshal: ok, thanks!
19:46rillianI updated the manifests generated by taskcluster/docker/recipes/install-mercurial.sh too for completeness sake.
16 Mar 2017
No messages
   
Last message: 159 days and 13 hours ago