mozilla :: #ateam

11 Jan 2017
11:44whimbooTomcat|sheriffduty: no tree closure for android build failures?
11:45Tomcat|sheriffdutywhimboo: autoland ? will back this out
11:45whimbooTomcat|sheriffduty: the backout you landed still shows the bustage
11:46Tomcat|sheriffdutywas a backout fix
11:46Tomcat|sheriffdutybustage bix
11:46whimbooso it also didnt fix it then
11:47Tomcat|sheriffdutyand so complete backout now
11:47Ms2gerjgraham, I think I figured out bug 1302621, will reply there after lunch
11:47bugbotBug Document Navigation, normal, nobody, NEW , Intermittent /html/browsers/browsing-the-web/history-traversal/browsing_context_name.html | Retaining on history traversal - assert_equals: Initial load expected "test" but got ""
11:50jgrahamMs2ger: Great!
12:05jmaher|afkTomcat|sheriffduty: I was wondering if you had any thoughts on linux32-debug in Taskcluster as tier-1 vs tier-2. I had sent an email two days ago and haven't heard from the sheriffs
12:10Tomcat|sheriffdutyyeah will do this today
13:10AutomatedTesterjust heading home. brb
13:50atowhimboo: Im a bit unsure what to do about which has gotten backed out twice.
13:50bugbotBug 1322862: Marionette, normal, ato, ASSIGNED , Callbacks and event listeners attached to content sandbox cause permission denied errors when inspected
13:51atowhimboo: For example I find it hard to believe that a change to execute script sandboxes affects bfcache navigation.
13:54whimbooato: i feel its a race
13:54whimbooin listener.js we only do "curContainer.frame.history.back();"
13:54atoYeah, the test failure in itself is certainly a race.
13:54whimboowe dont wait
13:54atoI can file a bug on that.
13:54whimboobut i understand your question
13:54atoBut I dont think its something new introduced in the patch.
13:55atoIm tempted to just try for autoland again.
13:56whimbooato: i dont actually understand why it has backed out
13:56whimboothe n jobs have been re-run dozen of times
13:56whimbooonly one was failing
13:56atoYeah, and debug builds are not the most stable.
13:56whimbooit shouldnt have been backed out afaics
13:57atoOK, thanks for looking. Ill try again and see if it sticks.
13:58whimbooato: i will put ni? on philor
13:58atowhimboo: Might be easier just to re-try.
13:59whimbooato: well, i would like to see more details
13:59whimbooato: oh, when you run try again, please check the fennec results. i think your test chagnes cause a lot of failures
14:01whimbooato: chunk 4 and 6
14:03whimbooato: bug 1223277 is btw a tricky one to fix
14:03bugbotBug Marionette, normal, hskupin, ASSIGNED , Clicking an element that closes the current window causes the session to hang
14:03whimbooato: imo we should find and fix those hang bugs first
14:03whimbooato: i mean before i work on other webdriver compat bugs
14:06atoThat is quite a serious bug, indeed.
14:19whimbooato: i have a similar other wonky situation with my changes for close window and failnig tests in execute_script
14:19whimbooi have no idea what's that but I improve this test module now
14:20whimboohave seen a couple of problems in there
14:21whimbooato: i assume that chunking is somewhat not working correctly
14:21atowhimboo: Hm OK.
14:21whimboothe chrome tests in there should never have run
14:21whimboobut they are not skipped
14:21whimboonot in the test nor in the ini file
14:21whimboothey start fialing now with addtional tests added
14:21whimbooalso the test_window_close_content test I cannot see executed
15:00whimbooato: the seen fennec failures for your patch might be caused by the same problem
15:03whimbooato: hm, I wonder why that following line doesnt raise when running in chrome with marionette
15:03whimboo self.marionette.execute_script, "return foo")
15:03whimboodoes marionette set a foo property?
15:04whimbooits for fennec only
15:07whimboointeresting. looks like its a left-over from a different test
15:10maja_zfafter much frowning and squinting, I think I've made performActions blocking
15:10maja_zf^ ato
15:11whimboomaja_zf: hurray
15:12maja_zfturns out we needed to use yield after all
15:12whimbooi assume your code in listener is a generator
15:12maja_zfnow it is :)
15:13whimboonot sure if the proxy can handle non-generator methods
15:13whimboomaja_zf: with the other r? we will have a 5th green chunk for fennec
15:14whimboonot sure how long
15:14whimboore the chunking issue
15:15maja_zfwhimboo: seriously, just skip everything else and bump up to tier 2
15:16whimboomaja_zf: not as long as the chunking issue hasn't been solved
15:17maja_zfwhimboo: why? if you skip the left-over tests, it doesn't matter how they are chunked.
15:17whimboomaja_zf: we have some tests which don't seem to be executed at all
15:18whimbooif len(tests) changes they will
15:18maja_zfso skip those too
15:18whimboothere is something horrible broken
15:18maja_zfwhimboo: is there a bug you can point me to re chunking?
15:19whimboomaja_zf: filed a couple of minutes ago
15:19whimbooyou are cc'ed
15:19whimboobug 1330321
15:19bugbotBug Marionette, normal, nobody, NEW , Chunking of tests seem to be broken (some tests are not executed)
15:20whimbooI will land my close window patch now. lets see how Mn1 behaves
15:21whimbooato: ^^
15:26whimbooato: question. if a test sets " = 3", should this still be set in another test?
15:26whimbooi think Marionette cannot clean it up itself, right?
15:29maja_zfwhimboo: this might answer your question
15:29whimboomaja_zf: sorry, I should have been clearer... its in chrome scope
15:30whimbooand the above attaches it directly to the window
15:32whimboowith the system sandbox it works fine
16:05atowhimboo, maja_zf: execute_script(" = 3") is roughly equivalent to execute_script(" = 3", sandbox=None).
16:05atowhimboo: Oh, chrome scope.
16:06atowhimboo: For chrome scope I think it will be preserved until the window object gets closed. I dont know if `window` in a modal dialogue is different than when youre on the main ChromeWindow window?
16:06atomaja_zf: Also, hurrah on perform actions!
16:06atoSorry, catching up with backlog because of 1:1.
16:09atowhimboo: Great work on shipping
16:09bugbotBug 1311350: Marionette, normal, hskupin, ASSIGNED , close() and closeChromeWindow() do not wait until the underlying window has disappeared
16:09atowhimboo: Itll help me fix the geckodriver.
16:10tcampbellis it possible to direct mozregression to start a specific nightly without bisecting?
16:11tcampbellalso I get 404s once the bisect goes from nightly to taskcluster
16:11whimbooato: ++ lets get geckodriver fixed
16:13whimbooato: k, so the test will reset .foo then
16:13whimbooits already up as r? :)
16:14atowhimboo: Yes, r+ on
16:14atowhimboo: All good changes.
16:14whimboohurray... only waiting for teh ty build
16:14whimbooand then landing
16:14atowhimboo: Just as a follow-up on that: To not have state preserved in execute_script, you need to pass sandbox=None.
16:15atowhimboo: This calls the WebDriver-conforming execute script implementation.
16:15whimbooato: even for chrome scope?
16:15atowhimboo: The API is a bit wonky in the Marionette Python client because of backwards compatibility.
16:15atowhimboo: It would work for chrome scope too, I believe.
16:15atowhimboo: But you wouldnt have access to a few things.
16:16atowhimboo: Notably Components:
16:16whimbooato: so for this test we could use it
16:16atosandbox=None calls sandbox.createMutable.
16:17whimbooato: should I check my patch to make use of it?
16:17atowhimboo: Nah, its fine.
16:23whimbooato: and now enjoy your holidays!!
16:32whimbooato: i think you should not push the patch again
16:32whimboowe have a green run for Mn
16:32whimbooat least not without updating the tests
16:40atowhimboo: Its not clear to me that its this patch that causes the intermittents?
16:42whimbooato: when you check al the commits before this failure doesn't happen
16:42whimbooas what I have seen you changed the way how we evaluate scripts
16:42whimboothe location we retrieve here is via execute_script
16:42whimboomaybe there is a side-effect
16:44atoI see. It only happens on very slow builds (i.e. debug) which could indicate that its timing related. Something like a race condition.
16:45whimbooato: OF is still kinda low for Mn jobs
16:45whimbooso any new introduced failure is clearly visible
16:45whimbook, heading out for a moment to build a snowman :)
16:46atoBuilding snowmen > Hacking on Marionette
16:47atoIn fact: Many things > Hacking on Marionette
17:08whimbooato: well, if the quality of the material would be good enough :(
18:30aashayhi people, i am new to open source. i know javascript, c, c++ and python. please guide me where to start working. I have already started working on bug #1282970. I have doubts regarding this anyone to help?
18:30bugbotBug Marionette, normal, nobody, NEW , Implement command to get single cookie
18:33AutomatedTesteraashay: what are your doubts?
18:36AutomatedTesteraashay: I have added acomment to that bug
18:45aashayAutomatedTester : how do i run only the cookie test file? do i have to give it as arguments?
18:45maja_zfaashay: yes
18:46maja_zfaashay: see also ./mach marionette-test -h
18:47jmahergbrown: that is nice
18:49aashaymaja_zf : thanks. that resolved my issue
19:27Noveeni want to contribute to the bug 1276128
19:27bugbotBug Places, normal, nobody, NEW , Convert xpcshell-tests in toolkit/components/places/tests/bookmarks/test_393498.js to Bookmarks.jsm API
19:27Noveencan someone help me here?
19:29aashaymaja_zf : when i write testcases for my function this shows up: "AttributeError: 'Marionette' object has no attribute 'get_cookies_by_name'"
19:29aashaycan help?
19:30aashaythe line on which the error is : cookies = self.marionette.get_cookies_by_name()
19:36maja_zfaashay: you can use import pdb; pdb.set_trace() before that line to set a breakpoint, then look at self.marionette in the debugger to learn more about it
19:36maja_zfif you're not familiar with pdb, see
19:38maja_zfbased on that you'll see that self.marionette is defined by the Marionette class in Search for |'class Marionette' path:testing/marionette| on to look at the code
19:50AutomatedTesteraashay: you're supposed to be adding that command
19:53AutomatedTesteraashay: you will need to update
19:57aashaythe get_cookie function in already does the required thing. then what is to be done?
19:58aashayi suppose the bug is to add a function in the driver.js. Am i right?
19:58aashayAutomatedTester : ^help
20:03aashaymaja_zf : thanks really. u got me started :)
20:04maja_zfaashay: we're in a meeting now but we'll get back to you about other questions later. :)
20:16aashayokay no problem. waiting for your reply :)
20:21AutomatedTesteraashay: you need to update driver.js to return a single cookie
20:22AutomatedTesteraashay: however, the python code is expecting a list so that will need to be updated too
20:25whimbooeven 6 chunks passing!
20:25whimbooso +2 today
20:25whimbooits only because tests have been movd around through chunks
20:26whimboousing the % operator to set the chunk id is funky
20:26whimboonice round robin
20:26whimboobut its hard to find tests
20:26whimbooand once a new test gets added everything changes
20:26whimboojmaher: how is the chunking done in Mochitest
20:27whimboosame way?
20:27jmaherwhimboo: yes, that is the current way
20:27jmaherwhich is frustrating
20:27whimboojmaher: additionally to that we seem to have bug 1330321 for marionette
20:27bugbotBug Marionette, normal, nobody, NEW , Chunking of tests seem to be broken (some tests are not executed)
20:28jmaherwhimboo: one difference might be that we do --run-by-dir on desktop, but I don't think we do on android
20:28aashayAutomatedTester : okay, got it. i'll try and write that. Also, how is any bug assigned to someone?
20:28jmaherwhimboo: oh, that doesn't sound good, but not impossible to fix
20:28whimboojmaher: it works locally but is busted remote on teh test machines
20:28whimbooi hope to figure it out until tomorrow
20:28jmaherpython versions? different libraries?
20:29whimboono idea. maybe we miss to package some tests?
20:29jmaherI am sort of brainstorming out loud, not sure if it really makes sense
20:29whimbooeg.i cannot find a newly added test in any of the chunks
20:29whimbooanyway. out for today
20:29whimboohave a good remaining day!
20:29jmaherlater whimboo
20:31maja_zfbye! way to go with the Fennec tests!
20:35aashayAutomatedTester : but what is the problem with the current implementation? sorry if i am asking many questions.
20:53jmahergbrown: and taskcluster proves it's strength yet again :)
20:56jmaherrwood: just used the add related configs option in perfherder, big win!
21:00rwoodjmaher: nice! :)
12 Jan 2017
No messages
Last message: 5 days and 20 hours ago