mozilla :: #ateam

 
27 Mar 2017
08:13igoldantopic
09:50whimbooTomcat|sheriffduty: hello.
09:50whimboothere seem to be build failures for webplatformtests
09:53Tomcat|sheriffdutyyeah for your push
09:53Tomcat|sheriffdutywhimboo: either its a #taskcluster issue or your push it seems
09:54whimbooTomcat|sheriffduty: i didn't change anything in that area
09:55whimbooits in tooltool actuallz
09:55whimboooh i think networkign
09:55whimbooImpossible to add file geckodriver-v0.15.0-linux64.tar.gz to cache folder /builds/tooltool_cache
09:58jgrahamYEah idk why that happens, but ato plans to remove the tooltool dependency soon, so it should go away for that reason
09:58Tomcat|sheriffdutyjgraham: but seems its perma failing now
09:58Tomcat|sheriffdutyand its a tier-1 thing, so either we close the tree and investigate or we backout
09:58atotooltool issue?
09:58whimboowait
09:58whimboothe failure is due to bug 1309765, right?
09:58bugbotBug https://bugzilla.mozilla.org/show_bug.cgi?id=1309765 web-platform-tests, normal, nobody, NEW , Intermittent TEST-UNEXPECTED-TIMEOUT | /webdriver/navigation.py | expected OK
09:59whimbooit started to fail now again
09:59atoAttributeError: 'NoneType' object has no attribute 'size'?
09:59Tomcat|sheriffdutyyeah see https://hg.mozilla.org/integration/autoland/rev/5573ab27d696d236e48c7d781ac5b7998204d95c
10:00jgrahamTomcat|sheriffduty: Ah, well the tooltool thing shouldn't be related to the failure
10:00Tomcat|sheriffdutyjgraham: yeah i think its https://hg.mozilla.org/integration/autoland/rev/5573ab27d696d236e48c7d781ac5b7998204d95c
10:00Tomcat|sheriffdutyjgraham: whimboo will back this out to clear the test failure
10:00whimboook
10:00whimbooso somehing else changed for navigation.py since my last push
10:00atoFor a moment I thought our office was on fire.
10:00atotoo.
10:00atoBut it was just a fire alarm test.
10:01jgrahamTomcat|sheriffduty: That sounds suspiciously like the system working as intended ;)
10:01atowhimboo: Do you understand what is the problem?
10:01whimboojgraham: i wonder when I will be able to finally land this patch! :D
10:01whimbooato: not yet
10:02whimbooit was all fine the last time
10:02atowhimboo: Yeah, AttributeError seems odd.
10:02whimbooato: where do you see attribute error?
10:02atowhimboo: In the bug you linked to.
10:02whimbooi dont find the wpt logs that helpful
10:03whimboooh, thats old
10:03whimboohttps://treeherder.mozilla.org/logviewer.html#?job_id=86594540&repo=autoland&lineNumber=1509
10:03atoOh OK
10:04atoWARNING - Impossible to add file geckodriver-v0.15.0-linux64.tar.gz to cache folder /builds/tooltool_cache
10:04whimboolooks like i have to rebuilt locally
10:04atoOSError: [Errno 13] Permission denied: '/builds'
10:04jgrahamato: That warning isn't important
10:04jgraham[task 2017-03-27T08:47:59.669537Z] 08:47:59 INFO - TEST-UNEXPECTED-TIMEOUT | /webdriver/navigation.py | expected OK
10:04atoWhy isnt it?
10:05jgrahamato: BEcause it always happens
10:05jgrahamI think you just made the test slower
10:05* ato sighs
10:05jgrahamYOu might need to add
10:05jgraham# META: timeout=long
10:05atoThat seems very likely.
10:05jgrahamas the first line and regenerate the manifest
10:05atoThe default timeout is very short.
10:05ato(For WebDriver tests.)
10:06jgrahamYEah, 10s by default 60s for a lone timeout
10:06jgraham(it's a problem for the new session tests where many tests have to open the browser)
10:08whimboojgraham: might it be this http timoeut again?
10:08whimboodid you get the fix pushed already?
10:10whimboojgraham: i have seen that you patched it somewhere, but which bug was it?
10:12whimboojgraham: oh bug 1318724! Is there anything which blocks us from getting this pushed?
10:12bugbotBug https://bugzilla.mozilla.org/show_bug.cgi?id=1318724 web-platform-tests, normal, nobody, RESOLVED FIXED, Webdriver tests unstable on Linux64 debug
10:13whimbooerr, why this is marked as fixed?
10:13atowhimboo: Unrelated, UI tests dont use HTMLElement.send_keys in the client?
10:14jgrahamwhimboo: I think that the http timeout would look different
10:15jgrahamBut I can land that now
10:15whimbooato: they do but not for update tests
10:15atowhimboo: So youre OK with the changs in https://reviewboard.mozilla.org/r/122114/diff/3#index_header?
10:16atowhimboo: We could send {id: , value: , text: } if needs be.
10:16whimbooato: see https://bugzilla.mozilla.org/show_bug.cgi?id=1348782#c8
10:16bugbotBug 1348782: Marionette, normal, dburns, NEW , update sendkeys command parameters to match Webdriver spec
10:17whimboobut if the workaround is fine, we could do it
10:19atowhimboo: Yeah, just wanted to confirm.
10:20whimbooato: k
10:21whimbooato: to be fully sure an example update should be run
10:21atok
10:43whimboojgraham: did we enable Wd tests on other platforms than Linux x64 debug lately?
10:43jgrahamwhimboo: I hope not
10:44jgrahamWell linux64-opt should be enabled
10:44whimbooi mean opt
10:44jgrahamBut we don't have geckodriver binaries uploaded for other platforms
10:44whimbooall the time? so nothing changed lately?
10:44jgrahamThis can change when it's in-tree
10:44jgrahamwhimboo: I don't think anything changed. What's the problem?
10:44whimbooi ask because the last push i did by end lf last week only run on linux 64 debug
10:44whimboohttps://treeherder.mozilla.org/#/jobs?repo=autoland&revision=1c0d8a46504a7e3fa3ea5524eee46553554efe6f&filter-resultStatus=testfailed&filter-resultStatus=busted&filter-resultStatus=exception&filter-resultStatus=retry&filter-resultStatus=usercancel&filter-resultStatus=runnable&filter-resultStatus=success&selectedJob=86035192&filter-searchStr=wd
10:45whimbooand debug is fine with the follow-up patch
10:45whimboobut now the failure is for opt, and pgo
10:45whimboomaybe its SETA?
10:46jgrahamCould be seta
10:46whimboolooks like i just hit it :(
10:46whimboowould be great to run wd tests for all linux builds when changes are made to them
10:47whimbooi can see the problem locally now, but wasn't last time. so i still think something else has an influence here
10:47whimbooi will check after lunch
10:50atoHopefully https://bugzilla.mozilla.org/show_bug.cgi?id=1340637 should help us get there.
10:50bugbotBug 1340637: Marionette, normal, ato, ASSIGNED , Import and build geckodriver in central
10:50atoFixing up an unrelated patch right now, but will work on clearing out the geckodriver PR queue later today.
10:51atoThen we can rewrite the git history for export, get a rubberstamp from gps, and put it in m-c.
10:51atoThen I can push the changes in that that changeset.
10:52jgrahamI'm looking at the new sesison stuff now so hopefully we can land that first too
10:52atoExcellent.
12:14whimboo|lunchato, jgraham: hah. the webdriver navigate.py test is wrong. `foo` is not really an invalid url. it just gets forwarded to the selected search engine
12:15jgrahamwhimboo: Forwarding it to the search engine is a bug
12:15jgrahamWebDriver isn't supposed to do that
12:16armenzghello? is this working?
12:16whimbooarmenzg: howdy Armen is here!
12:17armenzgI think it is working!
12:17armenzghello whimboo
12:18whimboojgraham: then we would need a different abstraction for webdriver. for marionette it should work when it comes to fx ui tests
12:18whimbooarmenzg: how are you?
12:18atoWebDriver is meant to return an error when given an invalid URL.
12:18atoI thought Marionette did that.
12:19atoI vaguely remember writing a patch for it.
12:19armenzgI'm good. I feel as it is my first day of school
12:19whimbooato: `foo` is valid to be entered into the url bar
12:19armenzgwhimboo: it is an odd feeling to be away for so long!
12:19whimbooarmenzg: yeah! how long was it actually? 2 months?
12:20whimbooato: which could also be a keyword as set by the user
12:20armenzgwhimboo: 10 weeks so a bit over 2 months
12:20whimbooarmenzg: wow. then you will notice lots of changes!
12:22atowhimboo: This was a long discussion in the WebDriver WG, but the conclusion was that there is no way to make WebDriver interoperable by pretending we interact with the address bar.
12:23atowhimboo: We hook into the navigate algorithm in HTML, which comes after any UA specific resolution done by the address bar.
12:24whimbooato: hm, you mean with curContainer.frame.location
12:24whimboointeresting
12:24atohttp://searchfox.org/mozilla-central/source/testing/marionette/listener.js#1000 ensures step 5 in https://w3c.github.io/webdriver/webdriver-spec.html#dfn-go is achieved.
12:25atoThis is why we `new URL(userInputtedURL)` so it throws on being passed a malformed URL.
12:25whimbooato: so I may have removed too much from navigate.js? I will have to check
12:25whimboothanks for pointing that out
12:25whimbooato: it doesn't throw in this case
12:26whimboolet me debug it
12:26atoOK.
12:26atoThis is pretty subtle, but Im glad a test caught it (-:
12:26atoato: Thank you former self!
12:26armenzgwhimboo: yes; a lot
12:27whimbooato: k, it throws but the failure doens't correctly end-up in driver
12:27atoarmenzg: Yeah, A-team/EngProd is no more. Welcome back (-:
12:27jgraham new URL("foo")
12:27jgraham TypeError: foo is not a valid URL.[Learn More]
12:27atowhimboo: Ah.
12:27jgrahamShould throw it seems
12:27armenzgato: this is why I never pushed for a batch of t-shirts
12:27armenzglol
12:27armenzgthe moment we call THIS and THAT it called for a split
12:27armenzg*we called it
12:28atoarmenzg: Dont worry, it will change again in 15 days: http://arewereorganizedyet.com/
12:28ato(-:
12:28jgrahamWell I'm not sure that the new structure doesn't have just as manu unrelated things together
12:28jgraham*many
12:28jgrahamWe shall see
12:29armenzgato: haha that site is jokes!
12:29atoarmenzg: You say that, but its predication was only off by four days this time (-:
12:29armenzgIf we use Sugru it will keep it all together
12:30whimbooato: looks like its caused by the return right after sendError()? Would that automaticlaly send an ok in the dispatcher?
12:31atowhimboo: It shouldnt unless the command is using the new dispatch technique.
12:31atowhimboo: sendError really means send an error in this case.
12:32mikelingcalixte: ping
12:32whimbooato: k
12:34atoI may have broken trace logging in Marionette.
12:36atoYes I have.
12:36atoWill file a bug and fix it ASAP.
12:37atoToday is a day of for n in {1..10}; do git mozreview push && done
12:37whimbooato: k
12:38whimbooato: so our unit tests pass because `foo` is added as path to the localhost:port
12:38ato-_-
12:39whimbooits getting funny
12:40whimbooato: btw. tracing still works with latest m-c
12:40atoI just pulled and it broke.
12:40whimboois it something else you pushed to autoland?
12:40atom-c was just merged.
12:40whimbooah
12:41whimbooso i will refrain from pulling
12:41atoIts related to the fallback prefs in https://bugzilla.mozilla.org/show_bug.cgi?id=1344748.
12:41bugbotBug 1344748: Marionette, normal, ato, RESOLVED FIXED, Set automation preferences in Marionette server
12:41atoBut Ill fix it right away.
12:42atowhimboo: OK, so in unrelated news I have dropped the controversial code changes from https://bugzilla.mozilla.org/show_bug.cgi?id=1337743 and the associated issues.
12:42bugbotBug 1337743: Marionette, normal, ato, ASSIGNED , Stop appending eAttemptQuit in quitApplication
12:42whimbooi have already seen that
12:42atowhimboo: Added a comment with a link to a follow-up for moving over the cancel notification to get us closer to shipping.
12:43atowhimboo: I also addressed your other issue about returning the data from quit-application, which makes a lot of sense.
12:45atoWill fix the broken fallbacks now, then work on the PR queue of geckodriver.
 
27 Mar 2017
   
Last message: 11 minutes and 1 second ago