mozilla :: #e10s

20 Apr 2017
00:17mrbkapelan: We still need beta approval for the experiement patches.
00:29pauljtAlex_Gaynor: fwiw, this is exactly the sort of thing I was hoping to capture on but it needs more work
01:34elanmrbkap: Gerry will approve them
01:34elanhe knows about them and will approve them before GTB
01:34elanapproval doesn't block anything besides actually building Beta 1, correct?
01:35elanwait do you have to manually uplift the patches after they are approved or is that done by a sheriff?
01:36mrbkapelan: normally a sheriff does it but I could also land manually.
01:36elanok, I don't think we need to go out of process because Gerry knows
01:37elanthat we need them for GTB
01:37mrbkapelan: did you see my comments above?
01:37mrbkapAbout my testing
01:39* elan reads
01:40elan"an extension"
01:40elana specific extension?
01:40elanmrbkap: ^
01:41elanit looks like bootstrapped add-ons
01:42elanwhich has a patch
01:42elanand then it sounds like we need to communicate an individual workaround for people who have opted into e10s
01:42elanand want to test e10s-multi?
01:44elan(1) two questions. If they disable browser.tabs.remote.force-enable, will they then be back in the eligible population for multi?
01:45mrbkapelan: yes.
01:45elan(2) is there a way to force enable multi and decide how many processes they want to use, I'm thinking yes because of that other bug that cited 32 processes
01:47mrbkapelan: yes again. Setting Dom.ipc.processCount overrides almost everything.
01:47elanok, I need to figure out if we can add that to the beta release notes
01:47elanwhich I have completely lost track of
01:47mrbkapExcept ineligabolity for e10s
01:47mrbkapSorry I'm on my phone right now.
01:48elanok, that's ok
01:48elanI need to take a break, it's been a long day
01:48elanI'm sure it's been a long day for you
01:48elanI can see if I can figure out how to articulate this
01:49elanbasically, what I am hearing is that Dom.ipc.processCount is an override if you are otherwise eligible...
01:49elanthat's complicated but it's making sense
02:24mrbkapelan: ok, I'm back on a real computer.
02:24mrbkapelan: what do I need to do?
02:39elan hi mrbkap:
02:39elanI don't think you need to do anything except watch bugmail in case something unexpected happens with the uplift
02:40elanI think you and I are basically making the judgement call that these two "issues" don't block beta 1
02:40elanand that smoketests show we don't introduce new risk to 54
02:40elanso I think the only next step here is to *possibly* consider getting a workaround in for the Beta release notes
02:41elanBeta two ships tuesday though
02:41elanso it's kind of like, this is our dress rehearsal, if you can get fixes in by Friday, we'll be better
02:41elandoes this sound reasonable?
02:42elan(I'm not sure a workaround will yield any statistical difference in the data we get)
02:43elanand Gerry just approved :)
02:44mrbkapelan: yep :)
02:44elanso at this point, the best thing you and I can do is log off
02:44mrbkapelan: great, sounds good.
02:44elanor just watch to make sure nothing crazy happens with the uplift
02:44mrbkapYeah, I'll keep an eye on my email.
02:44elanand be refreshed for another day tomorrow
02:44* mrbkap will be more refreshed tomorrow.
02:44elan^5 thank you SO much
02:44* elan too
02:44mrbkap^5 thank *you*
02:44elanbye thanks again for everything!
14:38Alex_GaynorNot sure if this is the right place to ask, if not hopefully someone can redirect me. I've got a small snippet of code in a test (a mochitest) that uses some functional that we'd like to remove from the content sandbox; however it's not an API that the web should ever have access to, so I dont' want to just add a new IPC method. Is there a common pattern for
14:38Alex_Gaynorhandling cases like this? An example solution might be sending some JS over the chrome process to run, or even just invoking a method on a service in the chrome.
14:50jimmAlex_Gaynor: a browser chrome test maybe - the test that runs on the chrome side has chrome permissions, so you can throw some js over to the content process that triggers that call and communicate with it through the message manager
14:51jimmAlex_Gaynor: look up in browser/base/content/test for lots of examples
14:51Alex_GaynorWas about to ask if there was a good example, thanks!
14:53jimmbtw there's a ton of info on our test suites up on mdn
15:01elanchutten: probably not
15:01firebotBug 1349363 FIXED, [e10s-multi] Beta 54 experiment
15:01elanhow bad is that and what do we need to do?
15:02chuttenelan: Basically you just need a data peer to sign off on it
15:02chuttenThe information's in
15:49mrbkapah crap
15:49mrbkapchutten: sorry
15:50chuttenmrbkap: I've done it too, no worries.
16:59Alex_Gaynorjimm: Are there any particular tests in browser/base/content that are a useful reference for communication between content<->chrome; there&#39;s quite a lot in here :-)
17:22jimmAlex_Gaynor: good place to start, look for ContentTask usage -
17:22jimmthat little helper may be all you need
17:22Alex_Gaynorjimm: hmm, how is ContentTask different from loadChromeScript?
17:22Alex_Gaynorwell, they sound like opposites I suppose :-)
17:24jimmhmm, let me look
17:24jimmI think loadChromeScript might be overkill
17:25jimmIf all you neerd to do is execute a little bit of script over on content, especially from a test, a ContentTask is the simplest way to do it
17:25jimmdepends on what you want to do I guess
17:26Alex_GaynorErr, I&#39;m already running code in content, I want to run a little bit of code in chrome; don&#39;t I?
17:28jimmhmm, usually when you run tests, things are controlled by script on the chrome side which then injects code over into content. can you give me a more detailed example of what you want to test?
17:32jimmnow I&#39;m curious what we use loadChromeScript for -
17:34jimmmochitest chrome and plain
17:36jimmso that does what you&#39;re looking for, but I&#39;m curious if maybe you can accomplish this without it.
17:36billmaklotz: ping
17:37aklotzbillm: pong
17:38billmaklotz: hey, I was wondering how much our a11y code relies on us running a windows event loop in the child process
17:40aklotzbillm: It mostly does not. In fact I specifically wrote most of it to avoid using the event loop. That said, there could be some weird unforeseen situation where it is being used, but probably not difficult to handle
17:41billmaklotz: cool, thanks!
17:41aklotzNo problem
17:42Alex_Gaynorjimm: Once I get this working I&#39;ll solicit some feedback on if there&#39;s a better way (I think I have the JS working now, but it turns out my JS remoting issues were hiding an underlying issue!)
17:46jimmbillm: I bet you could disable that now. just have to make sure the content process doesn&#39;t create any native windows, or if it tries to, those calls fail.
17:47aklotzbillm, jimm: One caveat: even though we don&#39;t use the message pump, I have still augmented our code in that loop to do specific things to process a11y events, so I would need to add that stuff to whatever the replacement is
17:48billmok, that&#39;s fine
18:34mrbkapGrover-QA: ping?
18:34Grover-QAHey there
18:36mrbkapGrover-QA: Hi, I&#39;m reading your latest comment on bug 1352388 -- what steps do you get *-multiBucket on Windows and *-test on Mac?
18:36firebot NEW, [Request] Softvision Testing for the e10s-multi System Add-On For Firefox 54 Beta Experiments
18:40Grover-QA:mrbkap We are getting test on Mac when a WebExtension is installed. We are getting multibucket on Windows as default or when a valid WebExtension is installed. If we installed a bootstrapped add-on, we get different values, but you&#39;re aware of that already.
18:42mrbkapGrover-QA: Do you have the mac build handy? Can you tell me the values of extensions.e10sMultiBlockedByAddons in that profile?
18:43mrbkapGrover-QA: also, what&#39;s the value of &quot;dom.ipc.processCount&quot;?
18:43Grover-QAmrbkap: One sec, a colleague was testing on that. I&#39;m going to see his workstation
18:43mrbkapGrover-QA: ok
18:49mrbkapGrover-QA: I&#39;m going to be AFK for lunch.
18:49* mrbkap will be back.
18:49mrbkapGrover-QA: Please answer here or in the bug and I&#39;ll see it when I get bakc.
18:50Grover-QAmrbkap: I will have answer for you shortly.
18:50Grover-QAmrbkap: extensions.e10smulti.blockedbyaddon = true dom.ipc.processcount = 1
18:54mrbkapGrover-QA: what was the extension that was installed?
18:55Grover-QAmrbkap: Tab Submit Basic, Tab Auto Refresh, and Incognito This Tab
18:56mrbkapFelipe: ^---
18:57mrbkapGrover-QA: oh and dom.ipc.processCount.web?
18:57felipeare all of these extensions webextensions?
18:58Grover-QAThe three I mentioned are WebExtensions
18:58Grover-QAGonna get you the value of this one then I am going to be taking an hour lunch break
18:58mrbkapGrover-QA: ok.
18:59mrbkapGrover-QA: where in the world are you?
18:59Grover-QAOur QA team is in Las Vegas, Nevada (Pacific Time Zone) - It is 12PM now
19:00Grover-QAmrbkap: dom.ipc.processCount.webLargeAllocation = 10 - Is this the value you were looking for?
19:00mrbkapGrover-QA: oh, great.
19:00felipevery strange to get different results per platform as there&#39;s no platform-dependent code
19:00mrbkapGrover-QA: no, if there&#39;s no .web that&#39;s expected, somewhat
19:00felipeI wonder if any of these addons are messing with anything
19:01mrbkapThis might be another symptom of not updating the cohort properly.
19:01Grover-QAmrbkap: He says there is no pref with that at all on Mac
19:01Grover-QA:StefanG_QA can you elaborate? He&#39;s the colleague testing on Mac
19:02StefanG_QAHey If you need me to check anything on the mac just ping me
19:03mrbkapGrover-QA: ok. Let&#39;s follow up after lunch.
19:04mrbkapStefanG_QA: too
20:08Grover-QAmrbkap: Ok, we are back
21:36Grover-QAmrbkap: ping Any update on the build with the patch?
22:15Grover-QAfelipe: ping
22:28felipeGrover-QA: pong
22:28Grover-QAfelipe: Hey, any update on the build for e10s multi addon with the patch fix? Blake said he was going to try to get one up and running today.
22:32felipeGrover-QA: we&#39;re still talking about the proper fix.. I&#39;m not sure if builds will be ready by today
22:33Grover-QAfelipe: Ok, thanks for keeping me posted. Need Info me on the bug when it&#39;s ready to go and we&#39;ll hop back into testing it :)
22:33felipealright, will do
22:38felipeGrover-QA: when you said:
22:38felipe&quot;On Mac, it is set on multibucket before webextension is installed, it is set to -test after WebExtension is installed. &quot;
22:38felipewhich webextension were you referring to?
22:39felipe(do you have a testplan doc with the links so I can look into what extensions are you using in your testing)
22:44felipeGrover-QA: do you know offhand the answer to that? (which addon did you use in that Mac test)
22:44Grover-QAfelipe: Tab Submit Basic, Tab Auto Refresh, and Incognito This Tab. All three of them caused those to change
22:45felipeok, thanks! I&#39;ll try to reproduce this here and see what&#39;s going on
22:45felipeGrover-QA: that was with Beta build 1?
22:45Grover-QAYes, Beta build 1.
22:45felipeor a custom try build?
22:50GankroAlex_Gaynor: jimm: wrote up the problem and my proposed solution:
23:11jimmGankro: thanks. I&#39;m going to plop that link into an email and forward it around to a couple people. will cc you in.
23:12Gankrojimm: i just realized I didn&#39;t fully explain the overhead of bincode, it&#39;s non trivial to overcome but I&#39;m optimistic
23:12jimmno worries
21 Apr 2017
No messages
Last message: 3 days and 20 hours ago