mozilla :: #build

16 Mar 2017
01:20ewongmshal ping
02:38mshalewong: pong
02:46ewongmshal: got a moment?
02:46ewongI realize it's probably end of day for you..
02:47mshalewong: I can try :)
02:48ewongmshal wrt to bug 1345781.. might you have a bit of time to look at it?
02:48firebot NEW, bustage in wget-en-US due to PKG_PATH
02:48ewongthere's also that TB bug that I've lost track of
02:48mshalI haven't had time to look into your recent comment - is that the latest or did you end up trying other things?
02:49ewongoh wait.. getting confused :( I meant bug 1345422
02:49firebot NEW, Thunderbird Daily l10n repacks failing for Windows as of 2017-03-06
02:50ewongmshal bug 1345781 is a comment addition
02:50ewongjust for clarification
02:52ewongmshal: I do have a workaround for bug 1345781 which involves overriding the wget-en-US rule in and put it in suite/locales/
02:54mshalewong: does thunderbird need a different 7zSD.sfx input file? Or was the path resolution just ending up looking in the wrong plcae?
02:55mshalone of the things we were hoping to get with the files and the like is to represent building a Windows installer with the minimum amount of info (which I believe is just the directory to package up and the tag file)
02:56mshalI thought the 7zSD.sfx file was a "constant" input, but I could be mistaken
02:56mshal(it seemed to be constant for m-c, at least)
02:56ewongmshal the issue was with the path hardcoded with firefox in it.
02:58ewongmshal and it looks like TB's and Firefox's 7zSD.sfx is different
02:58mshalahh, ok
02:58ewongwhich is why I was hoping to replace that with some 'flexible' path.. however.. the problem with my patch is it assumes that '../../..' will always point to the 'right' place during building
02:59ewongsince I can't predict if it changes.. it could potentially break in the future
02:59ewongso I had debated whether to actually specify the path of the whole thing for MOZ_SFX_PACKAGE, but it makes zero sense as it would confuse anyone looking at it
03:00ewonglike "what is ../../../other-licenses/.." relative from?
03:00mshalyeah, seems simpler if we're already passing in a full tag path to also do a full 7zSD.sfx path
03:03mshalewong: r+ with some comments - sorry for the delay!
03:04ewongmshal: no worries. thanks for the reviews. and if you have time, can you advise me on bug 1345781 (wrt my comments)
03:04firebot NEW, bustage in wget-en-US due to PKG_PATH
03:04mshalewong: mind if I take a look at that tomorrow? I should have some time
03:04ewongmshal no.. I don't mind. any time is fine. thanks!! appreciate the help
03:05mshalnp - thanks for making this work!
03:09ewongthanks for taking the time to review the patches.
14:15froydnjwhat is the minimum glibc we support?
14:18froydnjah, probably new enough that I can use this without glibc version checking
14:48RyanVMfroydnj: 2.26 I think
14:48RyanVMsorry, 2.22
14:52larsbergted: froydnj: Just wanted to mention!topic/, in case MinGW Servo is something we need to support for you folks! There's some desire to deprecate it, but we can certainly keep it around (or just make sure it keeps building) if that's required.
14:54froydnjlarsberg: how complex is the mingw support in servo?
14:55froydnjlarsberg: there was some talk of standing up a mingw firefox build so that we stop breaking it so often
14:55froydnjlarsberg: I guess that would keep stylo building, which is probably what we care about atm
14:58larsbergfroydnj: It's not super complex, tbh. It's more that we really don't want anybody developing or debugging against it and especially not using it on Windows instead of MSVC. The biggest hole is that there's a lot of work to make the perf not totally suck (e.g., with font selection) so we try to steer people away from it pretty hard.
14:59larsbergI think keeping it building isn't too bad, modulo switching to cross-compiling using the msvc rust+cargo toolchain, as rust/cargo on mingw are in a kinda awkward state right now.
15:00* larsberg suddenly realizes Firefox will probably hit the cargo mingw issue, too
15:13tjrSo we are working on standing up the mingw firefox build
15:14tjrI'm not very familiar with Servo, but I had thought it wasn't in Firefox? Or are components of it there? Or maybe it might be some day?
15:14Ms2gerComponents are in Firefox by now
15:15Ms2ger"Stylo" is the project to integrate the Servo style system into Gecko
15:28RyanVMyeah, I thought Tor used mingw for their buidls
15:29* RyanVM wonders idly if we could get them onto clang-cl by the time we ship esr59 (assuming that'd be acceptable to them, too)
15:31tjrI am not so idlying wondering that! christoph is going to talk to them about it at the dev meeting next week.
15:32tjrI need to understand more about clang-cl though. I know it's meant to be a cl.exe replacement using the clang backend, but is it designed to run as a drop-in on windows only, or can it be easily run on linux also?
15:32RyanVMisn't the Linux version just clang?
15:33tjrWell, tor cross-compiles on linux for windows
15:33RyanVMI....have no clue
15:33tjrDoes it aim to have feature-parity with cl.exe (meaning it would implement CFG) or would it be more clang (having CFI)? And replace CFI with other things that may make a difference like SEH)
15:33tjrYea me either :-p
15:34RyanVMtjr: further confusing is that MSVC also has a clang backend that isn't clang-cl
15:35RyanVMtjr: I mean, we're working on running ASAN on Windows care of clang-cl, for whatever that tells you
15:54Ms2gerfroydnj, tjr: is there a bug for that mingw builder?
15:57froydnjtjr: it's clang features (so CFI, for instance) with a cl.exe-compatible command-line interface
15:58froydnjtjr: I'm 90% certain you could cross-compile for windows from linux with it
17:51GeKotjr: google has a bug for chrome that might shed some light on it:
17:53GeKo(i don't know how mature clang-cl is to support our cross-compiling from linux -> windows, though)
17:54tjr"For now, link.exe, mc.exe, and the idl compiler would still need to run under wine, but that seems to just work."
17:54tjrThat's a hard blocker
18:14GeKoyes. on the other hand that's from two years ago. there might have been some improvements meanwhile.
18:14GeKothanks for the mail
18:15tbsaundetjr: you could use the wine idl compiler same way mingw builds do, and I don't really see a reason you can use ld in place of link.exe for clang-cl builds
18:15tbsaundeerr, I mean I think using ld to link would be fine
18:15tbsaundenot sure what happens for mc now for mingw builds
18:16tbsaundeand of course you'd need to convince the build system to do the right thing for this configuration
20:49RyanVMso apparently we don't bundle the debug msvc runtimes with our CI builds
20:49* RyanVM wonders if that's on purpose or not
20:49RyanVMour *debug* CI builds, that is
20:50RyanVM(had to walk a Softvision tester through grabbing the redistributable so he could test some builds)
21:41dmajorRyanVM|biab: the debug CRT is not redistributable
21:43glandiumRyanVM|biab: we don't use the debug CRT
21:57dmajorI'm pretty sure we do
21:57dmajor(in debug builds)
21:58dmajorI've certainly had to copy it into VMs to test stuff
22:00glandiumdmajor: I'm pretty sure we don't
22:04RyanVMglandium: dunno, his builds were crashing on startup until he installed it
22:06glandiumRyanVM: that would a bug then, because it's certainly not intended
22:25* dmajor wonders if it kicks in by default #ifdef DEBUG or _DEBUG or something
22:30dmajorglandium: hmm, xul.dll from a recent debug build on inbound depends on the non-debug CRTs
22:30dmajorI wonder if some other file in the package accidentally uses debug CRT?
22:31dmajorI suppose I could test this easily
22:40dmajorhmm, still not seeing anything link to the debug CRT
22:40* dmajor wonders why he had to copy it into his VM then
17 Mar 2017
No messages
Last message: 12 days and 9 hours ago