mozilla :: #ateam

5 Jan 2017
01:42agashlinDoes anyone have advice for getting an official release branded Try build?
01:42agashlinI'm trying to understand how the configuration is different from what I'm seeing locally, where I can do MOZ_OFFICIAL_BRANDING=1 in browser/confvars.sh
01:52agashlinSpecifically I'm trying to get a build of the stub installer, and setting MOZ_STUB_INSTALLER=1 in confvars.sh at least worked for that. But I suspect somewhere there's a mozconfig coming into play that overrides that with a --with-branding
01:54agashlinprobably better to ask over in #releng, and during the day...
08:30whimbooTomcat|sheriffduty: hi
08:30whimboore https://bugzilla.mozilla.org/show_bug.cgi?id=1326174#c18
08:30bugbotBug 1326174: Marionette, normal, hskupin, RESOLVED FIXED, For unsupported commands in chrome context throw UnsupportedOperationError
08:30whimbooformerly it was great to see which files has merge conflicts
08:30whimbooyou missed to add it this time
08:34Tomcat|sheriffdutyyeah i think its bettter to make more verbose comments instead that just needs rebasing one
08:36whimbooTomcat|sheriffduty: so for this i cannot reproduce the conflict
08:36whimboomaybe a depending commit landed with another merge after you tried that one
08:37whimboomaybe it was the function formatting one for driver.js
08:38Tomcat|sheriffdutyyeah i think it was driver.js
08:39whimboo++
08:40whimboogood that we uplifted this formatting patch
08:40whimboootherwise lots of corrections would be necessary the next weeks
08:40whimbook, stepping out for a while. bbl
09:42Tomcat|sheriffdutywhimboo: uplift worked
11:35AutomatedTestermy wife is amazing... she was out and brought me back a flat white
11:35jmaher|afkwhat is a flat white?
11:35jmaher|afkis that like a white house?
11:36jmaher|afkahh, wikipedia https://en.wikipedia.org/wiki/Flat_white
11:36jmaher|afknice AutomatedTester !
11:37AutomatedTesterjmaher|afk: its the best way to have a coffee with milk
11:38jgrahamI thought it was a caucasian in a traffic accident.
11:38jmaher|afkusing 'flat' confused me as I switched to Queen English whilst parsing your message and it had confusion
11:38jmaher|afkjgraham: heh, so many interpretations
11:41tedhuh, we run the mochitest server processes anew for every dir in run-by-dir?
11:41tedi guess that makes sense to minimize potential server state leakage between sets of tests
11:41jmaher|afkted: yes
11:42jmaher|afkted: there is leakage- although we could clean that up
11:43tedOK
11:44tedmmm
11:44* ted is poking at logs from https://bugzilla.mozilla.org/show_bug.cgi?id=1325683
11:44jmaher|afkis there a problem with doing that? or just not optimal
11:44bugbotBug 1325683: General, normal, nobody, NEW , Many treeherder logs are missing crash stacks
11:44tedi think the problem here is that these assertions in his test logs are assertions in the xpcshell process running httpd.js
11:44ted(just one more reason not to use xpcshell for our web server in mochitests...)
11:45jmaher|afkah, interesting and a good reason to explain what is happening
11:45tedjmaher|afk: just not optimal, but it makes sense
11:46tedi really wish we had found the motivation to port everything to use a python httpd for mochitest at some point over the years
11:46tedoh well
11:46jmaher|afkted: it probably isn't too late
11:49tedOf course :)
12:09whimboojmaher|afk: quick fyi... with my upcoming patch the Marionette tests on Android will soon be green... https://treeherder.mozilla.org/#/jobs?repo=try&revision=ea9dc34d7f57c974dc3b498e936ff35b83d08d72&filter-tier=1&filter-tier=2&filter-tier=3&filter-searchStr=android
12:50Ms2gerted, we've got wptserve now ;)
13:44tedMs2ger: i would be totally open to using wptserve in mochitest
13:44tedthe sticking point last time we discussed it was all the .sjs in the tree
13:44* Ms2ger would prefer getting rid of mochitest altogether :)
13:45jgrahamWell wouldn't we all
13:45jgrahamBut achievable goals and all
13:46tedright
13:46tedunless you've also got someone handy to rewrite tens of thousands of tests
13:46tedhave you got that?
13:46Ms2gerYou?
13:48jgrahamI dont think that you will get sign off on ted rewriting all the tests
14:01atojgraham: Todays gold from geckodrivers issue tracker: If you some of you have time to spend in it it is very easy to reproduce. All you need is to have a build set up with Jenkins or TfsBuildAgent to run on a VM (WinServer2012).
14:02AutomatedTesterwat
14:04atoYeah, Ive got plenty of time to spend.
14:04atoJust not doing that.
14:16Ms2ger<ato> Yeah, Ive got plenty of time to spend.
14:16jgrahamAh, the kind of easy where you spend a week setting up the environment.
14:17Ms2gerWanna rewrite all the mochitests?
14:17* jgraham wonders if someone set this as a goal for Ms2ger and he&#39;s trying to subcontract
14:17atoMs2ger: There are some more meaningful things I want to get around to first. (-;
14:18atoWell I read a story that some programmer for a large American corporation actually did this.
14:18atoHe outsourced his job to a team of programmers in India.
14:18atoAnd presented the work as his own.
14:20Ms2gerAnd got fired in the process
14:21atoAnd brought us this wonderful story!
14:21jgrahamOr promoted to management, it&#39;s hard to tell
14:22atoAutomatedTester: r? https://reviewboard.mozilla.org/r/101636/diff/9/#index_header
14:27AutomatedTesterato: the Mn tests don&#39;t look too happy for that run btw
14:28atoAutomatedTester: Theres one failing test that I broke in a fixup somewhere.
14:28atoAutomatedTester: Looking at that now.
14:28AutomatedTesteralso, there is a new logviewer in treeherder that doesnt show the log
14:28AutomatedTesteremorley: ^
14:29emorleyplease could you file bugs if there are issues :-)
14:29AutomatedTesterFailed to execute &#39;postMessage&#39; on &#39;DOMWindow&#39;: The target origin provided (&#39;https://taskcluster.github.io&#39;) does not match the recipient window&#39;s origin (&#39;https://treeherder.mozilla.org&#39;).
14:29emorleyand mark blocking bug 1315808
14:29bugbotBug https://bugzilla.mozilla.org/show_bug.cgi?id=1315808 Treeherder: Log Viewer, normal, helfi92, RESOLVED FIXED, New Log Viewer integration
14:29AutomatedTesterthat seems suboptimal
14:30AutomatedTesterle sigh
14:30emorleyI agree, (bug 1315808 comment 13), albeit that error message appears in working cases too
14:32whimbooato: i assume the top level context for content is the tab&#39;s window, and for chrome it is the chromewindow?
14:32emorleythe log loads for me fwiw
14:32AutomatedTesterI have blank in that area
14:33ted&quot;post_upload.py is running the go code at github.com/mozilla-services/product-delivery-tools/tree/master/post_upload.
14:33tedIt has the .py extension, so releng didn&#39;t have to modify their calling scripts as part of the cutover.&quot;
14:33AutomatedTesteremorley: looks like HTTPS everywhere was blocking it
14:33atowhimboo: I dont really know what you mean by that.
14:34jgrahamted: Wasn&#39;t there some similar story about support.google.com or so having .py urls for a long time after it stopped using python?
14:34whimbooato: https://w3c.github.io/webdriver/webdriver-spec.html#dfn-current-top-level-browsing-context
14:34tedjgraham: i dunno, but i can believe that :)
14:34atowhimboo: So a top-level browsing context isnt equivalent to anything in XUL land.
14:35whimbooato: so its the domwindow of the tab
14:35atowhimboo: Not quite.
14:35emorleyAutomatedTester: oh strange - the iframe target (the new unified log viewer) is https, so not sure how it is affected?
14:35atoWell whatwg.org is down.
14:36AutomatedTesteremorley: now enabling it and reloading its working
14:36AutomatedTestermaybe it was just a hickup
14:36AutomatedTesterbloody computers
14:37atowhimboo: A top-level browsing context is a browsing context which has no ancestral browsing context. Its effectively what is inside a tab in Firefox, but not quite at the superlevel that is <xul:browser>.
14:37whimbooato: https://www.w3.org/TR/html5/browsers.html#browsing-context ?
14:37atoEw, w3.org.
14:38AutomatedTesterato: what about 5.1
14:38AutomatedTesterthat&#39;s where we are these days
14:39atowhimboo: <xul:browser> is a frame which holds a web view in Gecko, so its roughly equivalent to that.
14:39RyanVMemorley: are you aware of any log viewer bugs already opened for issues like https://treeherder.mozilla.org/logviewer.html#?job_id=66423492&repo=try&lineNumber=10049 ?
14:39RyanVMi.e. going to some random lines in the log rather than the actual error lines shown above?
14:40whimbooato: sure. but when we work in chrome scope, the browsing context is tied to the chrome window
14:41atowhimboo: You could say that, but just be aware that the comparison isnt 100%.
14:41emorleyRyanVM: I don&#39;t think there&#39;s already a bug for that - could you file one?
14:41RyanVMwill do
14:42whimbooato: so lets say the dom window... window.document.defaultView
14:43ato<xul:browser> has a property that is a WindowProxy element, but also a lot of other crap that isnt part of the platform.
14:43atos/element/reference/
14:43Ms2gerwhimboo, TR/ is for trash
14:43whimboooh
14:43RyanVMemorley: bug 1328880, thanks
14:43bugbotBug https://bugzilla.mozilla.org/show_bug.cgi?id=1328880 Treeherder: Log Viewer, critical, nobody, NEW , Disagreement between failure summary and actual log line numbers on Windows TC builds
14:44atoMs2ger: Wed use whatwg.org if it wasnt broken.
14:44emorleyRyanVM: many thanks :-)
14:45whimbooato: so it looks like that for the bug I&#39;m working on I will have to add checks to driver.js and listener.js. Or check that both contexts are valid
14:46atowhimboo: OK, so in the context (hah!) of that bug we just need to check if the users _selected window handle_ is still open.
14:47atowhimboo: WebDriver doesnt operate on the selected browser/window/tab in the UI, but the one selected through the Switch To Window command.
14:48atowhimboo: This is referred to as the current top-level browsing context in WebDriver and represented as GeckoDriver#getCurrentWindow() in Marionette.
14:48whimbooato: getCurrentwindow returns chrome windows
14:49jmaher|afkwhimboo: exciting that we might have green Mn on Android
14:49whimbooso eg. open a tab, switch to it, and close it, will not throw NoSuchWindowError
14:49whimboojmaher: yeah. about 10 unit test files remaining to fix
14:50atowhimboo: It sometimes returns <xul:browser>s when the curFrame or or curBrowser is set.
14:50atowhimboo: Which I agree is a chrome thing.
14:50whimbooato: at least i cannot get the above testcase to throw the error
14:50whimbooi will investigate more
14:51atowhimboo: Basically we want a check that returns an error when a Firefox tab the user is sending commands to disappears.
14:51whimbooato: not only a tab. also a chrome window
14:52whimbooso i have to handle two hierarchies
14:52atowhimboo: Right, and getCurrentWindow() returns either a ChromeWindow or a <xul:browser> depending.
14:53atowhimboo: Which by the way is problematic, because theyre different prototypes.
14:53whimbooato: but both have a document property
14:53whimbooand as such a defaultView
14:59atowhimboo: I think you want to look at curBrowser in this case.
15:07whimbooato: so curBrowser.window is the chrome window. do we have a reference to the xul:browser dom window?
15:07whimboootherwise we store the wrong value for curBrowser.window
15:08atocurBrowser.browserForTab returns the currently switched-to or currently selected <xul:browser>, but we dont have access to the DOM window from chrome.
15:10whimbooato: interestingly the debugger doesn&#39;t show me curBrowser.browserForTab
15:11atoIts a getter
15:11atoSo it might look like a property.
15:12whimbooah
15:12whimbooin __rpto__
15:20jmaherRyanVM: I see you merged to ash yesterday, linux64 opt has some reftest failures on buildbot, but not on taskcluster- those are the same failures we see in linux32 tc reftest which gbrown saw that we had reftest.list manifest conditions to expect failure
15:20RyanVMlmao
15:20RyanVMHTH did I not notice those were even running there still?
15:21RyanVMfile a &quot;linux buildbot reftests still running on Ash&quot; bug and assign to me please?
15:21jmaherRyanVM: the good news is we will have reftests running on linux32 in tc- it will go in with the patch to run tc linux32 tests
15:21RyanVMsweet
15:22jmaherRyanVM: should we disable linux64 reftests on ash?
15:22RyanVManyway, those shouldn&#39;t be running on Ash at this point, was an oversight
15:22RyanVMwe shouldn&#39;t be running the buildbot reftests anywhere but esr45 at this point
15:22jmaherI was working on disabling devtools on linux32 debug in buildbot and saw exceptions for ash - which is why I looked
15:22RyanVMthe overall intent of Ash was to run all e10s tests on all platforms
15:22jmaherRyanVM: ok, let me just do it with my other patch
15:22RyanVMthat works, thanks
15:23jmaherRyanVM: wait, this is reftest-e10s in buildbot for linux 64 opt/pgo
15:23RyanVMsame story
15:23jmaherok, got it
15:24jmaherwhat about web-platform-tests in BB vs TC?
15:24RyanVMif they&#39;re running on TC, I don&#39;t care if we kill them on BB
15:24jmaheris Ash TC only where possible?
15:24jmaherok, great
15:24RyanVMif you&#39;re feeling so motivated, we also run a bunch of non-e10s win7 vm tests on Ash that don&#39;t need to be running there
15:24RyanVMi haven&#39;t been motivated enough to do anything about it since it&#39;s not really hurting anything
15:24RyanVMjust unnecessary
15:25RyanVMyou&#39;re reminding me that I need to get ASAN wpt running on Ash again, though (got lost in the BB->TC migration, but they were flaky anyway)
15:25RyanVManyway, not your problem :)
15:28jmaherRyanVM: do you know the repo/tool to list_builder_differences.sh ? I am on a new machine and don&#39;t have all my old repos sitting around
15:28jmahermaybe braindump
15:30RyanVMjmaher: https://wiki.mozilla.org/ReleaseEngineering:TestingTechniques
15:30RyanVM<3 my awesomebar for that one :P
15:32jmaher:)
15:35RyanVMman, I love seeing commits that are substantially more lines of added tests than actual code fix
15:35jgrahamDoes that opinion change if all those tests start failing intermittently?
15:35RyanVMnot when it&#39;s a different project&#39;s infra! :P
15:36RyanVM(sqlite in this case)
15:36jgrahamheh
15:37jmaher:)
15:38RyanVM(on a side note, the sqlite folks are awesome to work with)
15:39jmaherI assume their queries of you are lighter than most?
15:40RyanVM:)
15:41davehuntwhimboo: you&#39;re leaving on the Sunday?
15:52malayaleecoderjmaher: I am here, let&#39;s figure out how to get the &#39;when&#39; clause inside the test configurations :)
15:54jmahermalayaleecoder: sounds like fun!
15:55jmahermalayaleecoder: lets start with what changes you have made to get the &#39;when&#39; clause added to desktop-tests/tests.yml
15:58malayaleecoderjmaher: here is the diff for your reference, https://pastebin.mozilla.org/8958873
16:00jmahermalayaleecoder: I am not sure what the &#39;jobs-from:&#39; is used for
16:01malayaleecoderjmaher: from my observation, wherever you have the &#39;when&#39; clause, it is either from kind.yml or another *.yml file which as been declared under &#39;jobs-from&#39; in the kind.yml of that directory
16:05jmahermalayaleecoder: it is hard for me to tell, maybe we should go to #taskcluster and ask :dustin as he masterminded this
16:06malayaleecoderjmaher: sure, will do that
16:10tedso i fiddled with esprima a little this morning: https://pastebin.mozilla.org/8958876
16:10tedthat&#39;s some summary stats on &quot;stuff that&#39;s in our .sjs files&quot;
16:10Ms2gerDoes use strict even work with single quotes?
16:10* jmaher needs to learn what esprima is
16:11jmaherted: ah, so these 259 functions we would need to have a replacement for in python
16:12tedit&#39;s not 259 functions, that&#39;s 259 unique top-level statements
16:12jmaherand I expect things like CC, Ci, timer, BinaryInputStream, BinaryOutputStream, etc. are just using pythong specific functions
16:12tedwell, unique as in naming
16:13jmaherok
16:13jmaherI wonder what wpt uses
16:13tedso out of 351 sjs files in the tree, 196 of them just declare a handleRequest function, nothing else top-level
16:13jmaheroh, nice
16:13ted158 of them have other stuff, like variable declarations or Components.utils.import or whatever
16:14tedyou can see from the output there there are 31 uses of &quot;var BinaryInputStream = ...&quot;
16:14tedthose basically all look like
16:14tedconst BinaryInputStream = CC(&quot;@mozilla.org/binaryinputstream;1&quot;,
16:14ted &quot;nsIBinaryInputStream&quot;,
16:14ted &quot;setInputStream&quot;);
16:17jmaherted: isn&#39;t this just receiving data or reading a file? doesn&#39;t seem hard in python
16:17jmahergbrown: ++ on the linux32 tests !
16:17jmaherI just saw the pulsebot messages in #developers :)
16:17jmahergbrown: although one is r=jmaher and author is set to &quot;Joel Maher&quot;
16:18tedjmaher: nothing the sjs files do is hard in python
16:18gbrownjmaher: oops. I am a bad patch thief. :)
16:19tedin fact, i would bet that 90%+ of what they&#39;re doing would be way easier
16:19jmaherted: I thought there were some quirky things
16:19tedjmaher: well sure
16:19tedit&#39;s not *hard*, just quirky
16:19jmaheriirc chmanchester looked into this when he was a contributor and made great strides, but got stuck
16:19tedif we were to port them to wpt, i bet wpt has enough built-in stuff that half of them would be one liners
16:19jmaheroh, port the specific tests to wpt, that would be interesting
16:19jmaherand a migration path that would be more ideal
16:20jmaherhmm, gsoc project anyone?
16:20jgrahamI sort of imagine that ted means wptserve rather than actually wpt
16:20tedyes, that
16:20tedsorry
16:20jgrahamCertainly some tests should also be ported to wpt :)
16:21jmaherI ported one, then got in trouble by the lint police a few weeks later when it merged to another repo :)
16:21jmaherbtw, thanks jgraham for fixing those lint issues
16:26jgrahamjmaher: No problem. It is supposed to fail a lint on m-c before it ever lands, but idk if that ever get reenabled after it turned out that some lint jobs were running against the wrong tree, or similar.
16:27jgrahamahal|afk would know if he was around
16:35jmaheryeah, ahal is out sick today; either way we should figure out how to enable that again and move forward
16:36jmaherjgraham: btw, you had mentioned web-platform-tests have a test-lint (or some other name) for verifying new tests are stable- is that run via tc on pushes to autoland/etc. or is it via some other channel/project?
16:38jgrahamjmaher: Run via travis on github pushes
16:38Standard8jgraham: jmaher: it got fixed, as far as Im aware, at least central has lint tests all unhidden
16:38jgrahamOh, then I wonder why it didn&#39;t catch jmaher&#39;s push
16:39jmaherStandard8: I think there are a lot of types of lint tests, possibly we have it enabled for certain types of files or certain directories and not others
16:39jmaherStandard8: I know you have been working on eslint for frontend JS code :)
16:39jgrahamjmaher: The stability checker really wants to have access to your vc history to see which files were modified. Once we run wpt from source that will presumably be easy, at the moment it is not
16:40jgraham(one can imagine some trickery to write it out in the build job and pass it through the test zips, but honestly that seems like a lot of wasted effort)
16:43jmaherinteresting
17:04gbrownjmaher: I have a patch for tc linux32 devtools; do you want that in bug 1328915, or a separate bug?
17:04bugbotBug https://bugzilla.mozilla.org/show_bug.cgi?id=1328915 General Automation, normal, nobody, NEW , disable linux32-debug devtools tests on all branches (trunk, aurora, beta, release, etc.)
17:04jmahergbrown: oh, if you have a patch,please put it there
17:04gbrownok
17:04jmaherI have been struggling with list builder differences on win10
17:05jmaherI think I just got it working, but I can shortcut and look at your patch :)
17:05jmahergbrown: we should also disable via taskcluster :)
17:06gbrownmine is just tc
17:06jmaheroh
17:06jmaherI misread
17:06jmaherI will continue on the buildbot train
17:19jmaherRyanVM: as we now have linux32 testing in TC (as of 1 hour ago), do you want BB disabled for linux32 on Ash ?
17:20RyanVMkill it with fire
17:27jmaherato: thanks for your email reply to the RFC RFC
17:32atomcote: Apparently jgrahams email didnt make it to tools@ either. Could it be that posting to it from the newsgroup, which armenzg_lunch and wlach|afk seems to be doing, is the only way that works?
17:34atoCertainly I didnt receive it.
17:34jmaheryeah, I didn&#39;t receive it either
17:35jmahergbrown: odd, the failure on linux32 tc about flash-player, I had thought you had a patch to resolve that
17:35mcoteato: could be. maybe the list is busted
17:36gbrownjmaher: I did, but I didn&#39;t check that it worked on linux64
17:37jmaheroh
17:37jmaheralways something
17:38* jgraham also tries auto-tools
17:39jmahertoo bad we don&#39;t have a tools and productivity team to make our lives easier
17:41atoI think theyre all busy being productive.
17:43jmaherRyanVM: for ash, do these changes look good to you: https://pastebin.mozilla.org/8958902 ?
17:44RyanVMjmaher: lgtm, but I think there&#39;s more x64 jobs that could also go mirroring the 32bit changes you&#39;re making
17:45jmaherRyanVM: yeah, I am going under the premise that l32/l64 are going to be running 100% of tests on TC
17:45RyanVMyeah, feel free to wield a heavy axe towards Ash then
17:45RyanVMi haven&#39;t made any major efforts to clean that up as jobs/platforms have moved to TC
17:46RyanVM(and Ash predates a lot of that migration work)
17:46jmaherRyanVM: I believe linux64 debug is already reduced
17:48jmaheras in only tc jobs
17:48RyanVMexcellent
17:49RyanVMjmaher: TBH, I think we&#39;re getting close to being able to kill Ash (or at least the e10s team&#39;s need for it)
17:49RyanVMonce e10s-multi actually manages to land and stick on nightly
17:50jmahertrue
17:51jmaherRyanVM: with e10s-multi will we need additional test runs to test e10s vs e10s-multi ?
17:53wlach|afkato: jgraham: could you try subscribing to tools through https://lists.mozilla.org/listinfo/tools and trying to post again? I don&#39;t think it&#39;s a good idea to followup to auto-tools, which is a private list that only reaches a small subset of release & productivity
17:54RyanVMjmaher: that&#39;s somewhat debatable, the premise thus far has been that if tests are green with multiple content processes, they should be fine with a single one too
17:55RyanVMbut at least in theory, you&#39;re right that Ash could still serve a purpose as the opposite of whatever trunk is testing
17:55RyanVMso we don&#39;t have any merge day surprises
17:55RyanVM(though the simulation runs would hopefully catch those anyway)
17:56jmaherRyanVM: ok- I am just trying to think ahead here- I think your argument for just e10s is most likely the reality
17:56wlachato: jgraham: oh, I guess you already tried that. n/m
17:56* wlach throws up his hands in frustration
17:56jmaherRyanVM: I suspect we will turn off non-e10s tests in the near future, I understand the majority of stuff is finishing in the 53 branch
17:58armenzg_mtgjmaher: I will be there in a couple of mins; please start w/o me
17:58armenzg_mtgsorry!
17:58RyanVMjmaher: that&#39;ll be a fun discussion to have
17:58jgrahamwlach: Yes, I have posted to tools in the past
17:58atowlach, mcote: Interestingly, the From/Cc headers dont appear correct. to be sent to the mailing
17:58atowlach, mcote: List-ID: <dburns-directs.mozilla.com>
17:58RyanVMjmaher: i&#39;m going to suggest you (or whoever intends to propose it) come to the Friday e10s cross-functional meeting when that time comes
17:58wlachato: huh, dburns-directs?
17:58atoIt looks like the newsgroup to mailing list redirection is wrong.
17:59atowlach: Yeah, thats the header taken from the email I got from you.
17:59wlachato: oh, probably because I bcc&#39;ed to lmandel-all
18:00wlachbut I don&#39;t understand why that would affect sending to mozilla.tools
18:00atoOh right, I see that.
18:00atoWell mozilla.tools to tools@lists.mozilla.org works.
18:00mcoteif you haven&#39;t raised this as a service now/hub request, I would do that asap
18:00atoNot tools@lists.mozilla.org to mozilla.tools.
18:01atoI have hubbled it.
18:01mcotecool
18:01atoI will follow up with this new information.
18:01mcotegroovy
18:01jmaherrwood: hey, meeting?
18:03armenzg_mtgjgraham: ato mcote I&#39;m subscribed to the various mailing lists through here https://lists.mozilla.org/listinfo
18:04armenzg_mtgI either submit directly to the mailing list or via groups.google.com
18:04atoWhen you write to the list (not the newsgroup), what email do you use?
18:04rwoodjmaher: sorry, grabbing a room
18:05jmaherrwood: nice
18:06armenzg_mtgato: tools@lists.mozilla.org
18:09jgrahamI wonder if wlach cross-posting to so many places generated headers that confused the bridge
18:09wlachoh probably
18:09wlachI didn&#39;t realize that cross-posting would affect replies sent to only one place
18:10armenzg_mtgI&#39;ve had trouble with cross posting when posting to the firefox mailing lists
18:10armenzg_mtgI&#39;ve given up on using newsgroups
18:11atoIt might be crossposting, but Im still unable to send new topics to tools@lists.mozilla.org.
18:11jgrahamYeah, I just tried &quot;replying&quot; with a fresh message and it doesn&#39;t seem to have worked, so I guess that&#39;s not it
18:14atoI also tried mailing mozilla-tools@lists.mozilla.org, which seems to be where messages sent via the newsgroup end up. This address isnt published by Mailman at all.
18:14atoMaybe the solution is just use the Google interface.
18:15jgrahamWell so far today I have a broken wpt, a broken mailing list and a broken computer
18:15jgraham2017 was great, for a while
18:15wlachwell, at least one of them is (probably) not your fault
18:16atoFWIW I also tried leaving the mailing list and resubscribing.
18:25armenzg_mtgthe ******* team
18:25armenzg_mtgthe .* team
18:26wlachmailing list software just seems to get worse and worse every year
18:28armenzg_mtgchmanchester: I must say I think I had 2 quarters last year that I had a week or so
18:36jgrahamWell, on the upside my computer works again
18:36jgrahamOn the downside the fastest path to victory was reformating it and starting again
18:46maja_zfjgraham: any objections to making geckodriver accept strings like &quot;\uE050&quot; to represent code points?
18:47maja_zf(there&#39;s a whole class of key actions I can&#39;t test because it raises for anything that&#39;s not a 1-char string)
18:51jgrahammaja_zf: So that *is* a one char string https://is.gd/5Vz7p1
18:52jgrahamGeckodriver should accept > 1 char strings, but once you unescape that particular example you are supposed to end up with a single unicode scalar value (&#39;char&#39;)
18:54jgrahammaja_zf: Did that make sense?
18:55maja_zfjgraham: yes... but that &quot;conversion&quot; doesn&#39;t happen. Am I writing it wrong the message body then? something like {&quot;type&quot;:&quot;keyDown&quot;, &quot;value&quot;:&quot;\uE050&quot;} works fine with Marionette directly, but Marionette sees that as 5-6 chars and accepts it. Once it&#39;s passed through to event.js, it&#39;s interpreted as one unicode value
18:56jgrahammaja_zf: Unless I have wildly misunderstood how WebDriver is supposed to work, you are supposed to send a single character over the wire
18:58jgraham(afaict the entire insanity of the design is predicated on the use of single char storage in Java or something)
18:59maja_zfjgraham: then I don&#39;t see how a Python client would send over the &quot;Control&quot; key as a single character, which means that I&#39;m maybe misunderstanding unicode representation in Python
19:00jgrahamHmm, so not only are one-click loaners for wpt not implemented; I also have &quot;insufficient scopes&quot; to create one
19:01jgrahammaja_zf: In [2]: print u&quot;\uE051&quot;
19:01jgraham
19:05maja_zfyep, that&#39;s what i&#39;ve been doing.
19:05atomaja_zf: The control key is represented by a special Unicode character.
19:06atowlach: I think its maybe that other software keeps getting better, but mailing list software is stuck in the 1990s.
19:07wlachato: I feel like @ Mozilla we started with something simple that worked well (newsgroups over nntp), then kept on layering crap on top which made the whole contraption more and more brittle
19:07atomaja_zf: Also, in Python 2.7 strings arent Unicode by default.
19:08atomaja_zf: So &quot;\uE051&quot; is not the same as u&quot;\uE051&quot;.
19:08jgrahammaja_zf: Then I don&#39;t understand why you are getting more than one character over the wire
19:08jgrahammaja_zf: Got any logs or anything to share?
19:08ato>>> len(&quot;\uE051&quot;)
19:08ato6
19:08ato>>> len(u&quot;\uE051&quot;)
19:08ato1
19:09jgrahamwlach: Newsgroups are unfortunately pretty poorly supported these days
19:09wlachjgraham: true, that&#39;s a bit of a seperate issue :)
19:09atoI was going to say Usenet support limits you to very few clients.
19:10atoBut layering crap seems like a common enough phenomena (-:
19:10maja_zfato: sorry, that was just a typo in IRC. Thanks, I know about u&quot;\uE051&quot; vs &quot;\u...&quot; :)
19:10wlachI understand why we did layer all the things on top (mailing lists, then google groups, etc) but I feel like we are overall due for a bit of a rethink
19:11atomaja_zf: FWIW I wont get around to your review today.
19:11atowlach: The mailing list to newsgroup mirroring seems mature for retirement in my opinion.
19:12jgrahammaja_zf: Hmm, so the only thing I can think is weirdness around json
19:13maja_zfjgraham: ato: to clarify the problem, when I send |&quot;value&quot;: u&quot;\ue051&quot;|, webdriver.transport.HTTPWireProtocol complains about |&quot;value&quot;: &quot;\\ue051&quot;|. When I first saw this, I assumed it was an issue on geckodriver&#39;s end
19:14atoThis is odd:
19:14maja_zfi&#39;ll dig around in wdclient to see if it&#39;s doing anything weird to the string
19:14ato>>> json.dumps(u&quot;\uE051&quot;)
19:14ato&#39;&quot;\\ue051&quot;&#39;
19:14wlachato: I would be happy with plain old mailing lists that worked, yeah
19:14wlachato: I would also take out the google groups integration, mailman&#39;s barebones archive is fine imo
19:15maja_zfato: jgraham: note that this is not a problem when using marionette directly
19:15atowlach++
19:15maja_zfato: no hurry with review, i have comments to james to chew on
19:16maja_zfnom nom nom
19:16atomaja_zf: Are we only serialising dicts to JSON? https://github.com/w3c/wdclient/blob/master/webdriver/transport.py
19:17jgrahamato: That&#39;s the correct json representation
19:18atojgraham: &quot;\ue051&quot; is?
19:18jgrahamYeah
19:18ato&quot;\ue051&quot;.length == 1, but paste above has two \\.
19:20jgrahamIn the JSON text it may be represeted as \\ue051 i.e. with a single literal \
19:20jgrahamThe js side should decode that to U+E051 i.e. a single codepoint
19:21atoBut certainly https://github.com/w3c/wdclient/blob/master/webdriver/transport.py doesnt json.dumps.
19:21atoIndeed.
19:21atoAnyway, have to go (-:
19:22maja_zf>>> json.dumps(u&quot;\uE051&quot;, ensure_ascii=False)
19:22maja_zfu&#39;&quot;\ue051&quot;&#39;
19:22jgrahammaja_zf: Yeah, either representation is OK afaict
19:38jmaherarmenzg++
19:39armenzgjmaher: now time to use it in production!
19:39jmaher:)
20:01wlacharmenzg: big congrats on getting seta into treeherder btw
20:02wlachthis should hopefully make future modification and deployment much easier
20:20armenzgwlach: yes, working and reviewing 1800+ lines can get tiring
20:20armenzgthanks
20:20wlacharmenzg: at least future work will be more incremental, and easier to both review and revise
20:21armenzgyes :D
20:38maja_zfjgraham: i removed the &quot;single-char&quot; exception from webdriver-rust and unicode code points work now. whatever is considered a single unicode value in json (\\ue051) is a multi-char string in rust, so we need to let it through
21:55AutomatedTesternight all
21:58jgrahammaja_zf: Something still seems strange, I will have another look in the morning
21:59maja_zfthanks
22:01maja_zfbbl
6 Jan 2017
No messages
   
Last message: 172 days and 14 hours ago