mozilla :: #build

8 Aug 2017
00:43mjfBack after vacation for a week and my Ubuntu builds are failing locally. I did system updates, ./mach bootstrap ; ./mach clobber ; hg purge --all to no avail.
00:43mjfCould not find LLVM/Clang installation for compiling stylo build-time
00:43mjfAny suggestions?
00:54mjfMaybe a better question is, should I expect ./mach boostrap to install llvm-config?
15:32* froydnj wishes that hg had the equivalent of the shell's `command` builtin
15:49mjffroydnj: is that comment related to the hg log -G question you asked me? :-( If so, Im sorry.
15:50froydnjmjf: did I ask on IRC some days ago? if so, probably :)
15:50froydnjmjf: it's totally OK! I dropped the ball on try to fix it, and now we have a place for it
15:51mjfNo, this is in bug 1388213
15:51firebot NEW, ./mach bootstrap doesn't install llvm-config on Ubuntu
15:53mjffroydnj: I figured maybe the last question you asked me in the bug, and then your wish here might have been tied together.
15:53froydnjmjf: ah, I understand the question now. yes, the wish was related to the last comment
15:54mjffroydnj: Would it help you if I re-ran mach bootstrap with that alias temporarily disabled to see if it fixes the problem?
15:54froydnjmjf: that would be helpful, yes!
15:55* froydnj sees that HGRCPATH exists
15:56mjffroydnj: it definitely changes things. Im seeing clang.tar.gz downloading now...
15:57froydnjmjf \o/
15:59mjffroydnj: Thank you for thinking of that. I guess at some point in the future it will be safe to leave that alias enabled when running bootstrap?
15:59froydnjmjf: bootstrap should handle this gracefully, yes. I should write a patch
16:02mjffroydnj: Im rebuilding now, and Im getting much further in the build so it appears that was the issue.
16:02froydnjmjf: excellent
16:02mjffroydnj: thank you again for the help!
17:07tedfroydnj: is that bindgen bug, just looks like we don't pass down various SYSTEM cflags that stylo's bindgen script winds up needing to include headers
17:07firebotBug 1341234 NEW, bindgen doesn't pass CFLAGS for optional system dependencies
17:15froydnjted: ah, thank you!
17:15froydnjI am not exciting about manually maintaining that list of flags to pass down
17:18froydnjmaybe there's some list in or something that has all the third-party library flags that we need
17:21tedyeah, i dunno :-/
17:21tedwish we could make the stylo code maintain this, like "i'm going to include these headers that might be from system libraries"
17:22tedi love having epiphanies:
17:22firebotBug 1331049 NEW, DeadlockDetector death test block for 90s each and cause gtest failure due to timeout on osx debug T
17:22tederahm: ^^
17:23erahmted: omg
17:23tederahm: right???
17:23* erahm does a happy dance
17:24tedin fairness, we spent a *long* time trying to figure that out to fix the talos regressions
17:24tedit's pretty non-obvious
17:28erahmted: wait I thought we fixed that bug though
17:29erahm(the x-comp one)
17:30tederahm: a different patch landed that made it not a problem for talos, because we disabled some sandboxing setting
17:30tedbut we haven't fixed the actual path issue with the cross-builds yet, it got hung up in a bunch of other issues
17:31tedi *think* wcosta is finally out of the woods there
17:31tedyou can tell it was a fun bug because our solution involves a custom patch to valgrind
17:32ted(although that's not as bad as it sounds since sewardj wrote the patch)
17:32wcostayeah, valgrind issue is fixed, I am currently rebasing path patch
17:32tedwcosta: hooray!
17:33tedwcosta: see the scrollback, i think it'll also fix that gtest timeout that we worked around
17:35* wcosta looks
17:36wcostaI feel like in a series season finale, when I old mysteries are brought up and explained
17:37wcosta* when all
18:10gpsahh cool
18:10chmanchesteroh, nice!
18:10chmanchesterI must have last read those a few weeks ago
18:10tedsomeone filed a bug, they were hitting an edge case there
18:10tedchmanchester: sometimes stuff changes out from under you so quickly
18:18tedit currently breaks on the showIncludes parsing
18:19ted...which is why i revisited that other patch
18:25chmanchesterapparently we discussed this about a year ago in bug 1297373
18:25firebot NEW, Don't regenerate build ID for every build on developer machines
18:25tedtechnically buildid.h is just in GENERATED_FILES:
18:25tedbut the has a FORCE rule
18:31gpsted: i think showIncludes is about how far i got as well
18:32gpsi remember hitting something with path normalization
18:32gpsand thinking "oh no, this is going to be a rabbit hole" and bailed
18:39tedyeah there are probably some things that are going to wind up being terrible
18:41tednot bad: [task 2017-08-08T18:05:39.712572Z] INFO - MainThread - Finished in 35.837s
18:41tedmy upload script uploaded 3k+ files to s3 in ~36s
18:42tedincluding the time to sha-512 hash and gzip each file
18:42tedand download the tarball containing all the files from the index in the first place and unpack that
18:43gpsyou can do a lot of things in taskcluster tasks pretty quickly if the startup overhead is low!
18:43tedoh well, the *total* task time was longer
18:43ted[taskcluster 2017-08-08 18:05:40.564Z] Successful task run with exit code: 0 completed in 282.055 seconds
18:44ted~30s to fetch and load the docker image
18:44ted76.9s to fetch the clonebundle
18:45gpsi've been having IRL conversations with TC people about more intelligent scheduling so we get higher cache hit rate
18:45ted~2min to fetch the changeset from my try push
18:46gpsmore like 7s
18:46gps[vcs 2017-08-08T18:03:19.452269Z] adding remote bookmark release
18:46tedmaybe only 1min, and the other minute is something else?
18:46gps[vcs 2017-08-08T18:03:53.922962Z] (pulling to obtain d34b0a4cb676f0ff0b752f0aeac510cca68b1ffc)
18:46gpsthere is some dark matter in there
18:47gpsi need to fix this upstream
18:47tedoh, sorry, it's pulling + updating
18:47gpssparse checkouts will help a bit
18:47gpsbut the work to getting mach and things working with sparse checkouts will be... fun
18:48gpsoh, if anyone wants to weigh in on bug 1388471, that would be helpful
18:48firebot NEW, Add a configuration option for "this is a local development build"
18:48gpsi can't recall what our current state of all these special #define is
18:49gandalfgps: do you know the ETA for mercurial 4.3 release?
18:50gpsAugust 10
18:55tedgps: heh, highlight.js doesn't do a very good job here:
18:56gpsthere was a thread on the clang ml a few days ago about a new c++ documentation tool
18:59gpsoh, my unstated point is that any tool that isn't using an actual tokenizer or ast is doomed to fail horribly
19:00tedpygments does better:
19:00tedyeah, absolutely
19:00tedi'm sure highlight.js is just using regexes
19:01tedtoo bad, it'd be nice to just do this client-side
19:01gpsthey pretty much all do, sadly
19:01gpsdon't look at the pygments source code
19:02gpsit's too bad that languages don't publish a parsing grammar in a well-defined format
19:02gpsand/or offer easy-to-use "language services" to tokenize or parse an input
19:03tedwell C++ is a goddamned nightmare to parse
19:03tedprismjs seems to do OK
19:03ted(I pasted it into their "try it out" web form)
19:03gpsand it changes too much between major releases and even has large variances between implementations
9 Aug 2017
No messages
Last message: 13 days and 18 hours ago