mozilla :: #build

19 Apr 2017
00:33glandiumgps: r?
00:35gpsglandium: uh, did you get that patch somewhere?
00:36glandiumgps: what do you mean?
00:36gpsb/c i shouldn't be reviewing C/C++ in the Firefox repo unless it is a rubber stamp
00:36glandiumgps: it's for the command line brotli tool, it's not going to be built in firefox ever
00:37gpsbut it is a non-trivial patch. i'd feel better if this weren't done as an IRC review
00:38glandium(even if we get the brotli encoder in firefox, it won't use that file)
00:38glandiumgps: it's in the bug too
00:38glandiumin the second patch
00:38glandiumnow that i reopened mozreview to be able to push
00:39glandiumdammit, why did it eat one commit?
00:39glandiumaaah because it's public
00:40glandiumok, now fixed
00:41glandiumgps: I don't know why you think it's non-trivial, though
00:43gpsglandium: i don't think it is simple enough to warrant an IRC review
00:43gpswait - did you link to the correct try revision?
00:44gpsi see one review for brotli in my queue and it is touching build system
00:44glandiumgps: I folded in patch #2 in the bug
00:44gpsi hate r+ carry forward
00:45gpsoh, i completely misread the patch on try. failed to see it was itself a patch
00:45gpsyeah, this probably was trivial enough. derp.
00:47glandiumand the third patch needs a r+ again because of the mixup which reset its r+
00:49gpsthat makefile patch doesn't get any less strange with age
00:49gpsi can't believe we had an if else with identical one-line branches
00:49glandiumgps: you make me want to look how we got there
00:50gpsi'm scared i'm responsible for it somehow
00:54glandiumthat's how we got there:
01:03tedshitty makefile conditionals, is how
01:15glandiumgps: if you want to steal 1356927 and 1356929 from ted, I won't mind :)
01:15glandiumI want to get as much landed before I go off for two weeks :)
01:15glandium+as possible
01:17gpsglandium: you just got 4 r+s
01:21glandiumchmanchester: gentle ping on 1356933 :)
01:23gps[task 2017-04-19T00:55:05.466841Z] Generating tasks for test xpcshell on platform linux64-jsdcov/opt
01:23gps[task 2017-04-19T00:55:23.765401Z] Generated 2574 tasks for kind test
01:23gpsi think i found some sub-optimal python
01:24gpsi'm gonna go with list membership in a loop
01:24glandiumgps: O(n*m)?
01:24gpsor O(n^2)
01:24gpsdepending on the loop
01:25gpsdon't have time to look at it now. but that's ~20s in every push
01:28gpsbug 1357608 filed
01:28firebot NEW, ~20s pause in task graph logs
01:31gpsupdate: it spends the CPU time validating schemas
01:32gps6s CPU time in copy.deepcopy(). that's nice and random.
01:33glandiumgps: fwiw, it doesn't seem particularly slower on tests, it just seems generating tasks is slow, and there are a lot of them for tests
01:34glandiumthere's a reason I didn't want to generate the whole task graph for mach artifact toolchain
01:36* gps heads home
01:44philorglandium: should I be suspicious of on your push?
01:45glandiumphilor: maybe or maybe not, let me double check
01:45philorworld: wtf is this "Error 254" we keep getting everywhere?
01:46glandiumphilor: ah, it's a CLOBBER thing
01:48glandiumted: how does sccache handle errors?
01:49philor"Mac builds in automation require a clobber for a change in which ranlib they use" a close enough summary?
01:49glandiumphilor: yes
09:43tedglandium: what kind of error?
16:31Callek_cloud9glandium: how did you identify tooltool stuff to use the mach command... ?
16:54gpsted: the sccache hit rate is surprisingly high,c7ec4b6c4b7f2e59facdf77b037fe6d2b7968531,1,2%5D&series=%5Bmozilla-inbound,6413ad537250391d73a7e2128366dc02f48ee8d6,1,2%5D&series=%5Bmozilla-inbound,047921bde63b76270add4c1ec503cd8eccea02e5,1,2%5D&series=%5Bmozilla-inbound,f7b10e9548437f201be9b557a288d5668fb6baf0,1,2%5D&series=%5Bmozilla-inbound,102548c53902600d728a62ef6ee43
16:54gpsaside from one push, hit rate >75%
16:54tedgps: yeah, i did not expect that
16:55gpslooks like that one low hit rate push touched a handful of dom and layout headers
16:56tedi was thinking it'd be interesting to do graphs like glandium did for his initial sccache work, % of changesets vs % hit rate
16:56gpsthe number of pushes with a 100% hit rate is also promising
16:57gps got 100%
16:57gpseven though it touched C++ (backout though). that's cool
16:58gpswe really need the builds to detect when we can substitute an artifact build
16:59tedi know we've talked about it, but ideally we'd get to a point where artifact builds are just the build system pulling everything from cache
17:01gpsideally the build backend just does that (i think that's what you said)
17:05nalexanderI'd be really happy if we could teach the build enough about our dependencies to get there, but I have a hard time believing we'll do that while we still use make.
17:06nalexanderSuch systems are inherently hash-based, not timestamp based.
17:13tednalexander: yeah, i can believe that
17:20tedalso, i noticed we're not using sccache on the buildbot osx builds
17:20tedand i'm pretty sure we used to
17:20tedso i'm not sure what broke there
18:06mshalmaybe we turned it off accidentally when universal builds went away?
20 Apr 2017
No messages
Last message: 5 days and 1 hour ago