mozilla :: #jsapi

11 Jul 2017
03:29ptomato-Mcould anyone here vouch for my contributor profile?
03:29ptomato-MI just got r+ on a patch and bugzilla sent me a mail suggesting I create one :-)
03:30araiwhich bug? asking to the reviewer would be the easiest way
10:49Tomcat|sheriffdutytandem ping
10:49Tomcat|sheriffdutyjandem ping
10:49Tomcat|sheriffdutyjandem, hazard failure
10:49jandemTomcat|sheriffduty: looking
10:58jandemTomcat|sheriffduty: can you backout? this is a false-positive but a bit annoying to fix
10:58Tomcat|sheriffdutyyeah will do
10:59Tomcat|sheriffdutyjandem: done
11:09jandemTomcat|sheriffduty: thanks
13:02nbpjonco: Search for GC_MMU_50
13:03nbpas well as GC_MAX_PAUSE
13:05jonconbp: running GC in browser time changed the shape a several of these curves
13:07joncobrowser idle time, should be
13:07jonconbp: i don't think these results necessarily indicate problems
13:07nbpok, so are they useful?
13:09jonconbp: the regression alerts? yes they have been very useful
13:09jonconbp: I get an email when one of these happens btw
13:09nbpok, so no need to spam you over ird.
14:12tillwhat do people think about adding is<T> to Value, where T: JSObject?
14:13tillthis&#39;d expand to `this.isObject() && this.toObject().is<T>`
14:15araialso to<T> ?
14:16jandemmaybe isObject<T>/toObject<T> then, but I&#39;m not sure if Value should know that much about JSObject*
14:18tillyou mean because of code dependencies?
14:18jandem(and JSObject::is/to are engine-internal while Value is public)
14:18tillhrm, yes
14:18tillv8 has this, and it&#39;s nice
14:19tillbut the internal/public thing is probably a deal breaker
14:19jandemmaybe we can make it work somehow, or we could add ValueIs<T> or something
14:23tillIn the streams code I introduced this
14:23tilland it&#39;s used 33 times
14:29jandemah I see
14:55Tomcat|sheriffdutyjandem: bustage
14:55Tomcat|sheriffdutyshall i backout ?
16:07sfinkcan PseudoStack::pop gc?
16:07mrgigglessfink: No, |void PseudoStack::pop()| cannot GC
16:19smaugnbp: that was a good one, &quot;privately&quot;
16:20smaugnbp: hey, another thing. Maybe I asked already. How would one try to use the parsed JS cache?
16:20* nbp hides off the spot lights
16:20smaugwith speedometer
16:20smaugnbp: I could modify source code to not use idle time etc
16:20smaugjust for some testing
16:21nbpsmaug: I tested with the current harness, and it did not give any improvements.
16:21smaugI don&#39;t expect score to improve
16:21smaugnbp: I&#39;d like to see if the time to run the test would improve
16:21nbpsmaug: If you want to modify the harness, then you should make it wait a few hundreds milliseconds after the idle callback.
16:22smaugwell I was thinking to modify source code to cache eagerly
16:22smaugjust for testing
16:22smaugto see how it affects, or whether it affects at all
16:23nbpsmaug: there is an eager mode, with the strategy config (about:config, search for script_loader)
16:23nbpsmaug: but this will still wait for the idle callback.
16:23smaugright. I would just drop the idle part
16:23smaugrunning speedometer doesn&#39;t have too much idle time
16:23smaugcertainly not enough idle time to run GC/CC
16:24nbpsmaug: to save eagerly, you would have to replace these lines, by the call to the function:
16:25nbpsmaug: which should cause everything to be saved at the end of the load-end event.
16:26smaugok, thanks
16:26smaugnbp: where do we check if we have bytecode cache already?
16:26smaugI mean when loading the same script again
16:29nbpsmaug: in the ScriptLoadHandler
16:29nbpsmaug: grep for mDataType.
16:31nbpsmaug: MOZ_LOG=ScriptLoader,5 should report all the inner details of the bytecode cache.
16:33nbpstandups: Currently working on getting chunk-compressed source into the JSBC saved content.
16:33standupsOk, submitted #48082 for
16:40jcranmernbp: oops ;-)
16:40jcranmernbp: fwiw, I&#39;ve been having lots of fun with x86 asm in Rust
17:18nbpsmaug: any success?
17:18smaugnbp: didn&#39;t try that yet. Ended up profiling :)
17:18nbpsmaug: ok ;)
17:19smaugjandem: curious, have you profiled Elm-TodoMVC ?
17:20jandemsmaug: yeah, should help a lot
17:20firebotBug 1375505 ASSIGNED, Improve for-in/GetIterator performance
17:21smaugwe seem to also minorGC quite a lot with that
17:45* nbp off
18:00evilpieptomato-M: which embedding do you work on?
18:07evilpiejandem: for GetIterator, I think using CacheIR per jsop_iter site might be a good idea?
18:12jandemevilpie: yeah the current one last-cached for everything is silly
18:12evilpiedefinitely, I think I measured this a wile ago on speedometer
18:13evilpiejust having one IC per site already helped
18:14evilpiejandem: for this patch
18:14evilpiewe are basically not well typed, because useRegister expects a TypedOperandId, and we don&#39;t know if we have a symbol or object
18:15evilpieshould we not care, or maybe just split this into two ops CompareObject/CompareSymbol ?
18:16jandemevilpie: yeah i think that would be safest because it&#39;s guaranteed to work
18:16jandemevilpie: or one op but with a JSValueType
18:16jandemthat you can then pass to TypedOperandId
18:17evilpieok, I think splitting it would be simpler
18:17evilpiewe can even share most of the code, just have to setup the registers first
18:18evilpiegood catch with the BaselineInspector, would have noticed this only after removing the code
18:51ptomato-Mevilpie: javascript bindings for GNOME desktop&#39;s developer platform
18:59evilpieptomato-M: great, thanks for reminding me
19:01mrgigglesthe mozilla-inbound tree is now closed (assertions on OS X starting with the push for bug 1374888)
19:01firebot NEW, Crash in SharedLibraryInfo::GetInfoForSelf
19:12mrgigglesthe mozilla-inbound tree is now open
21:19jimbmrgiggles: who can review netwerk/protocol/http/HttpChannelChild.cpp
21:20jimbmrgiggles: pun
21:20mrgigglesjimb: There was a sign on the lawn at a drug re-hab center that said &#39;Keep off the Grass&#39;.
21:20mrgigglesjimb: mayhemer x 49, jduell x 20, mcmanus x 18
12 Jul 2017
No messages
Last message: 15 days and 1 hour ago