mozilla :: #build

20 Apr 2017
08:10Mavericksmach run <test_file> launches a nightly browser and doesn&#39;t execute the test
08:10Mavericksit displays the entire content in the test file instead
08:10Maverickswhat could be the issue ?
08:22DexterMavericks, that&#39;s not the right channel
08:22DexterMavericks, #introduction was
08:22DexterMavericks, but you quite before I could answer you there :)
08:22DexterMavericks, the right way to test is to use ./mach test <test_file>
08:22Dexter./mach run just runs an instance of the built Firefox
11:31MavericksThanks Dexter
14:37tedi didn&#39;t think MSVC versioning could get any more confusing
14:37tedbut somehow they have managed it
14:38tedI installed VS2017, it has a C++ compiler that reports its version as 19.10.25019
14:38ted(that&#39;s the usual level of confusing)
14:39tedwell, not quite, i guess, in that internally VS2017&#39;s version number is 15.1
14:40tedso previously you had three version numbers to deal with: the 2017, the 15.1, and the 19.10.25019
14:40tedwith 2017 they have split all the things it can install up into components, and one of those components is the C++ toolchain
14:41tedand for some godforsaken reason, the C++ toolchain component in VS2017 has the version number 14.10.25017
14:42froydnjI would love to hear the explanation for msvc&#39;s version schemes
14:43tedi would not be shocked to find that it mirrors their org chart
14:45Aryxnew boss = new version number?
14:46tedwell, each chunk of the org chart makes up its own version number
14:47catleeand then you xor them together
14:47froydnjand they are consistent with previous versions
14:47froydnjbut not with other arms of the chart?
14:47froydnjcould definitely see that
14:50tedyeah
14:50tedeach team has a totally reasonable and consistent versioning scheme
14:50tedthey just don&#39;t make any sense when mashed up with the other bits that ship in the same product
14:51tedwe did used to have separate gecko and firefox version numbers
14:51tedbut we squashed that
15:26teddo we know if anyone is building successfully with VS2017?
15:26teddmajor: ?
15:26dmajorted: it built fine a few weeks ago
15:26tedokay
15:27tedi&#39;m hacking up configure to be able to detect it
15:27tedand my build failed
15:27tedprobably just something regressed
15:28* dmajor reads backscroll
15:29dmajorted: does _MSC_FULL_VER at least match one of those other versions?
15:29teddmajor: yeah, it matches the cl.exe version
15:30dmajorted: where do you see the toolchain component version?
15:31tedhttps://irccloud.mozilla.com/pastebin/qzCif2UF/
15:31teddmajor: C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Tools\MSVC\14.10.25017
15:31dmajorted: oh, in the path - that&#39;s awful!
15:32tedwell
15:32tedthey give you an easy way to find the default toolchain version: C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary\Build\Microsoft.VCToolsVersion.default.txt
15:32dmajoroh and putting the product edition in the path us yuck too
15:32tedi guess you can have multiple toolchains installed side-by-side, which is not terrible
15:32teddmajor: are you just joining us? :)
15:33tedthey did at least provide APIs (and a commandline tool that uses those APIs) to find all this stuff
15:33tedhttps://blogs.msdn.microsoft.com/vcblog/2017/03/06/finding-the-visual-c-compiler-tools-in-visual-studio-2017/
15:33tedanyhow, once i get this working i am going to poke at getting builds working in WSL
15:34tedOK
15:34tedhttps://irccloud.mozilla.com/pastebin/a7ZQ6PBb/
15:35tedis the compile error i&#39;m getting
15:35tedi will file it
15:35dmajorted: oh, yeah, Span is new, that&#39;ll explain it
15:36dmajormy money&#39;s on https://dxr.mozilla.org/mozilla-central/source/mfbt/Span.h#56
15:38dmajorwould have been nice to have https://bugzilla.mozilla.org/show_bug.cgi?id=1318193 :)
15:38firebotBug 1318193 NEW, nobody@mozilla.org Tier-2 Windows builds with VS2017
15:38tedyeah
15:38tedmaybe i will take a crack at that as well
15:39dmajorespecially now that 2017 is the default download
15:39tedhttps://bugzilla.mozilla.org/show_bug.cgi?id=1358169 if you want to actually *fix* that bug
15:39firebotBug 1358169 NEW, nobody@mozilla.org mfbt/Span.h compile error with MSVC2017
15:39tedi don&#39;t think i know enough C++ to do so
15:40dmajorthat file was quasi-imported from another codebase and I bet the _MSC_VER >= 1910 path wasn&#39;t tested when it hit m-c
15:41dmajorthe big hammer would be to turn off constexpr like we do for the older compilers, but there&#39;s probably a more correct fix
15:42* dmajor kicks off a vs2017 build over lunch
15:44tedthx
15:49larsbergted: oh, sorry, didn&#39;t realize you were discussing VS/VC/toolkit version numbers here and not just on twitter. Feel free to ask away, though most of the history is exactly what you would expect
15:49tedhah
15:49tedlarsberg: the actual historical stuff is interesting but we can discuss that over a beer in SF :)
15:49larsberge.g., products were separate or were one-off things for {Windows, Office} and consolidating version numbers would require stuff like landing changes to core build infra scripts in those projects
15:50tedyeah, totally understood
15:50larsbergBut yeah, probably better over a beer in SF
15:50tedit&#39;s just funny to see so many different and confusing version numbers wind up sorta prominently displayed
15:50larsbergMainly just to us (build system owners). Marketing kinda had an edict that the version numbers never be user-visible
15:50larsbergb/c otherwise they threatened to force us to change them every rebrand
15:51tedhah!
15:51larsberge.g., they tried to make us change the version number to &quot;2013&quot; after visual studio .net
15:51tednice
15:51larsbergand we had to explain the PE format and limitations of the major version number in the binary and that we&#39;d have to rev the OS loader to support...
15:51tedi do like that the cl.exe version number is basically consistent all the way back to the very first release
20:46rillianglandium: cinnabar as been very slow pulling from mozilla-unified on my mac lately. Behaves normally on linux.
20:46rillianam I using the wrong version again? :)
20:46tedgps: https://tools.taskcluster.net/task-inspector/#a8WCM3JoQs24TmHazB3rhg/
20:46teddunno if that will *work*
20:50* rillian rebuilds the helper from the release branch just in case
20:51tedrillian: oh, that reminds me, can you point me at what&#39;s involved in repacking our existing rust toolchains with a newer cargo?
20:53rillianted: so there&#39;s a script in taskcluster/docker/rust-build/repack_rust.py
20:54rillianit either needs modification to pull a cargo version from a different channel manifest that rustc
20:54rillianor you can use the &#39;package_cargo&#39; method to package cargo separately and override the one in the rustc tooltool archive
20:54rilliandid your patch land?
20:59tedit did, yeah
21:00tedalthough i don&#39;t know how rust nightly picks its cargo snapshots?
21:02tedhttps://irccloud.mozilla.com/pastebin/7t8aS9Km/
21:03rillianted: I don&#39;t either
21:04rillianted: we could also build our own cargo of course, but my script for that has bitrotted.
21:04tedokay
21:04tedi will find out where cargo nightlies come from and try the package_cargo thing
21:24Silne30Hello, all. I am trying to create a new mach command for a test harness that I have written to run some tests. It&#39;s similar to the firefox -ui tests.
21:29gpsted: that&#39;s it?!
21:35Silne30i am using firefox-ui-tests mach command as a template.
21:36gpsSilne30: the first question I have is whether you need to write a new harness. harnesses are difficult to maintain. does an existing harness not suit your needs?
21:36Silne30gps: I built a harness on top of the Firefox-UI-Harness.
21:37Silne30There were some specific things for telemetry that needed to be implemented.
21:37Silne30gps: https://hg.mozilla.org/mozilla-central/rev/ea85026945a7
21:38gpsSilne30: so this is a flavor of marionette tests. i don&#39;t think you need a new mach command.
21:39gpsnew task in automation to run just these tests, sure. but you should be able to reuse mach commands for running marionette tests
21:39Silne30gps: I had talked to Henrik about it and he mentioned that I would need one. What do you propose?
21:40gpshow do we run marionette tests with mach today?
21:41gpsmach marionette-test
21:41Silne30So I can run mach marionette-test and it will point to my harness?
21:42gpswell, if the test manifest is defined in moz.build, then `mach marionette-test <directory|manifest>` or `mach test <directory|manifest>` should just work
21:43gpsyour new flavor of tests inherit from a slightly different python test class
21:43gpsthat&#39;s done at a level below mach
21:43gpsto mach, they are all marionette tests
21:45Silne30gps: Can you help me with that new task in automation?
21:45Silne30What do you mean by that?
21:45Silne30Treeherder?
21:46gpstreeherder just displays what automation runs. read https://gecko.readthedocs.io/en/latest/taskcluster/taskcluster/how-tos.html
21:47Silne30I see.
21:47Silne30So you don&#39;t think a new task is necessary.
21:47Silne30New command, rather.
21:47Silne30gps: ^
21:48gpscommand, no. task, maybe
21:48gpsif you define the test manifests as regular marionette manifests in moz.build, it should all &quot;just work&quot;
21:48Silne30gps: Can you comment on my bug then?
21:48gpsand which bug is that?
21:49Silne30https://bugzilla.mozilla.org/show_bug.cgi?id=1349597
21:49Silne30So I don&#39;t look like I am trivially closing issues.
21:49firebotBug 1349597 NEW, jdorlus@mozilla.com Create Mach Command for Telemetry Tests
21:52gpsdone
21:53Silne30gps: Really appreciate that.
21 Apr 2017
No messages
   
Last message: 155 days and 17 hours ago