mozilla :: #build

21 Apr 2017
00:20tedgps: yeah, so, not sure that VS offline downloading thing wants to work in TC for some reason
00:21tedprobably would be easier if i could see the desktop
00:21tedi'll spin up an EC2 windows instance tomorrow and try it
00:27gpsted: worst case we seed the offline download file in tooltool and have a TC task to turn that into an artifact/archive
00:27gpsat least that way the conversion from offline installer to toolchain is deterministic
00:27tedyeah
00:27tedright
00:27tedwould be nice if they hadn't stopped shipping ISOs
00:28gpsit also limits our exposure to verification failures
00:28gpse.g. if we need to change the toolchain artifact, we can tweak the layout *without* having to download new content
00:29tedyeah
00:29tedhaving a cached set of installer files and a task to build the package from that is still way better than "build it by hand"
02:15Silne30Hmmm...Can't seem to get git cinnabar to work.
02:15Silne30Access denied for the public key.
15:19mkaplyWhat does "You do not have sufficient scopes" mean when trying to trigger builds on try?
15:19mkaplyhttps://pastebin.mozilla.org/9019605
15:33nalexandermkaply: it means you don't have permissions.
15:33nalexandermkaply: which might mean you don't have Level X access.
15:33mkaplynalexander: this same thing worked yesterday. Maybe it says I'm logged in but I'm not?
15:34mkaplyThat's a really bad error. Plus it pops up and immediately goes away. So unless you're staring at the screeen, you don't see it
15:35mkaplyNope, logging out and relogging in same error. This all worked yesterday. Weird.
15:36mkaply(I even tried the try request that worked yesterday. Fails today)
15:36nalexandermkaply: ask in #taskcluster. That looks like internal fialure.
15:36mkaplynalexander:tx
16:27froydnjso I have to fix cargo linker issues for a servo bug, which affects webrender and cross-compiling programs as well
16:28froydnjmy current idea is to define a wrapper script that invokes the actual linker and any necessary LDFLAGS
16:28froydnjthe name of the linker to invoke and the arguments thereunto would be passed in via environment variables
16:29froydnjI'm sure there are all sorts of corner cases with shell quoting and whatnot there
16:29froydnjthe question is...do we care?
16:30froydnjI'm not inclined to write out the wrapper script every configure, though I guess that's kind of similar to what we do for e.g. gold
16:32froydnjor would we really be better served by a linker-wrapper.in script that configure substituted appropriate values in?
16:32froydnjI think you could probably construct some of the same corner cases there, but maybe it's less likely to matter?
16:52tedfroydnj: seems like generating a wrapper script at configure time is totally reasonable
16:52tedi do wish cargo had better support for passing down "here's the linker to use and also here are the arguments it needs"
16:57rhelmerhm. so I have a patch that moves system add-ons into the omni.ja, it works on in an unpackaged build but the packaged build isn't putting the files I want into omni.ja
16:57* froydnj tries the envvar approach, but assumes try will show him nine reasons why that's not feasible, even if it works locally
16:58rhelmerit is just hanging around in browser/chrome/features - I am doing it the same way e.g. devtools and webide work, with a features.manifest and entry in package-manifest.in etc.
17:10froydnj...and all the windows builds are falling over. that's good
17:13froydnjhm, cargo, it's super-unhelpful to tell me that the build failed and provide no other explanation
17:15froydnjah, here, the script is not a valid win32 application
17:15froydnjwhy not? it's marked +x
17:16froydnjmaybe it doesn't like #!/usr/bin/env bash
17:51froydnjhm, still doesn&#39;t think that&#39;s a valid script >.<
18:44tedrillian: oh, there&#39;s actually a manual step to update cargo for rust builds? TIL
18:45rillianted: per above, cargo in nightly in what the rust repo&#39;s cargo submodule says it should be. I opened https://github.com/rust-lang/rust/pull/41450 to get your change distributed.
18:45tedfroydnj: because cargo is just calling CreateProcess with it
18:45rillianer, above in #rust-infra
18:45tedrillian: thanks!
18:45rillianted: I also have alt-cargo repacking working
18:45froydnjted: yes, of course, doh
18:45rillianI&#39;ll put a script up for your review in a bit, if that&#39;s cool?
18:46tedrillian: sure! i will be gone in ~1.5 hours tho FYI
18:46tedfroydnj: i honestly don&#39;t know what the right thing to do on windows would be
18:46tedfroydnj: i guess just punt since we&#39;re not cross-compiling anything there (except x64->x86)
18:47froydnjted: atm I&#39;m just disabling on windows because 1) we don&#39;t cross-compile there and 2) we don&#39;t need the linker fixups like we do on mac
18:47tedyep
18:47froydnjif we need something there, I figure we can cross that bridge when we come to it
18:47tedyeah
18:56froydnjargh, cargo is using the &quot;target&quot; variables for the building of host binaries--e.g. build scripts--and that&#39;s falling over for some reason on osx
18:56* froydnj does yet another try push to fix this
18:58froydnjI can see how that logic makes sense, but it seems like it would be better to separate host and target configuration :(
19:08rillianted: script is up in bug 1358590
19:08firebothttps://bugzil.la/1358590 NEW, giles@thaumas.net Add --cargo-channel switch to repack_rust.py
19:17tedfroydnj: there&#39;s a whole bug about that, did you miss that?
19:18tedwe were hitting it when people were building win32 builds in a mozillabuild shell with the env vars setup for 32-bit MSVC
19:19tedcargo would go to build build scripts as host binaries, but use the vars from the env, and accidentally generate 32-bit libs and then fail
19:21froydnjted: hah! I saw that bug, but I did not make the connection =/
19:29* froydnj considers setting up automation so that --verbose is always passed to Cargo
19:29froydnjI know people groused about it for local builds, but it is really handy
19:31tedfroydnj: yeah, totally do that for automation
19:32tedi pretty much never am sad when CI logs are overly-verbose
19:55froydnjhm, MOZ_CARGO_WRAP_LD has sccache in it
19:55froydnjI wonder if that&#39;s causing issues
19:57froydnjactually, better question, why is this thing compiling cssparser when stylo isn&#39;t enabled?
19:57froydnjis this cargo weirdness again?
19:59rhelmerglandium: hey - I am trying to move system add-ons into the omni.ja, and I have it working except for the fact that the presence of an install.rdf seems to cause a XPI to be created
19:59rhelmerglandium: is there some way to override that so I can have the install.rdf inside omni.ja?
20:23tedfroydnj: shouldn&#39;t, sccache will just punt on things that aren&#39;t actually compilation
20:55rillianfroydnj: can we use bindgen in default builds yet?
20:59froydnjrillian: no, but we can certainly talk about doing so; stylo will need it anyway
21:02rillianfroydnj: mmm. Alfredo wants to hook the rust mp4 parser up to MOZ_LOG, which means calling C++. We might be able hold off by doing in in C++ glue code instead of rust.
21:03rillianfroydnj: is requiring everyone to install llvm what&#39;s holding us back?
21:04froydnjrillian: essentially, yes, and maybe some configure bits to ensure that we always have an appropriate LLVM to use
21:04froydnjrillian: what, you would bindgen logging methods and whatnot?
21:05rillianthat&#39;s the suggestion
21:05rillianthe alternative is to pass strings in a callback.
21:05rillianbut that&#39;s not helping us convert the rest of the module to rust :)
21:17nalexandergps: sigh, sorry for the noise on Bug 1254355. I&#39;m paging this stuff back in and making errors that I should catch.
21:17firebothttps://bugzil.la/1254355 FIXED, nalexander@mozilla.com Move preprocessed targets out of mobile/android/base/Makefile.in
21:17nalexanderIf any build peer can steal the trivial review at https://reviewboard.mozilla.org/r/39323/ I would appreciate it.
21:20rilliannalexander: lgtm, but not my usual area.
21:20nalexanderrillian: ta! I&#39;ll get it green on try (damn l10n repacks!) and see if I can get a GENERATED_FILES peer (chmanchester? mshal?) to stamp it too.
21:22* mshal looks
21:29mshalnalexander: is reviewboard showing more than it is supposed to? I&#39;m not sure how this fits the definition of &quot;trivial&quot; :)
21:30rillianmshal: I had to click on the patch name to get a sensible view
21:31rillianthere were four changes after that
21:32mshaloh, just adding $(abspath) in a few places?
21:32rillianyes
21:33mshalahh I went to https://reviewboard.mozilla.org/r/39323/ and clicked &#39;Show changes&#39;, which went to https://reviewboard.mozilla.org/r/39323/diff/5-6/ - and that was a bit of a mess :)
21:35mshalnalexander: well r+ for adding abspath where necessary - what actually breaks without them?
22:24Silne30How do I add some tests to moz.build to be run in automation?
22 Apr 2017
No messages
   
Last message: 155 days and 19 hours ago