These are chat archives for systemjs/systemjs

26th
May 2017
Nick George
@Izzmo
May 26 2017 00:05
@buddyp450 0.19.x is not painful, going from 0.19.x => 0.20.x is a breaking change and was super painful when I first tried it..
Yury Delendik
@yurydelendik
May 26 2017 17:00
so we have interesting thing: we have lots of file to load, and systemjs is using fetchFetch. On chrome very rare fetchFetch's promise is not resolved. does anybody have a similar problem?
Guy Bedford
@guybedford
May 26 2017 18:21
@yurydelendik any chance you have a test case, even if it only happens rarely?
Yury Delendik
@yurydelendik
May 26 2017 19:09
@guybedford we are getting it during testing of mozilla/pdf.js#8396 it's on linux + chrome
@guybedford or mozilla/pdf.js#8284 -- I need to try it on linux locally, but inserting lots of console.log's in the system.src.js gives something like https://pastebin.mozilla.org/9022749
more files we convert to es6 or more files we add increases chance of hang
Guy Bedford
@guybedford
May 26 2017 19:28
@yurydelendik and no errors are being thrown at all? I will definitely have a look.
Yury Delendik
@yurydelendik
May 26 2017 19:33
no errors
Guy Bedford
@guybedford
May 26 2017 19:38
@yurydelendik how would I run this test suite locally?
Yury Delendik
@yurydelendik
May 26 2017 19:41
gulp test (but edit test/resource/browser_manifest.json first)
Yury Delendik
@yurydelendik
May 26 2017 20:18
@guybedford maybe it's specific to certain version of chrome (it was Google Chrome 57.0.2987.21 beta)
Guy Bedford
@guybedford
May 26 2017 20:19
This stuff can hit browser bug territory sometimes, like the Edge function apply bug
Yury Delendik
@yurydelendik
May 26 2017 20:19
looks good locally on chromium 58
yurydelendik @yurydelendik updated test bot to run/check all problem PRs
Guy Bedford
@guybedford
May 26 2017 20:20
:+1:
yurydelendik @yurydelendik gives up running quality stuff on "beta" version of browsers :)
Yury Delendik
@yurydelendik
May 26 2017 20:27
ha, still fails on VM
Guy Bedford
@guybedford
May 26 2017 20:28
on 58?
Yury Delendik
@yurydelendik
May 26 2017 20:29
yeah (http://54.67.70.0:8877/7726f778ca04d00/output.txt has "test failed chrome has not responded in 240s")
Guy Bedford
@guybedford
May 26 2017 20:32
Doesn't this line mean that it was able to render - TEST-UNEXPECTED-FAIL | eq tracemonkey-eq | in chrome | rendering of page 2 != reference rendering?
Yury Delendik
@yurydelendik
May 26 2017 20:34
Yeah, but since I replaced the browser it might not render exactly the same pixels -- that's okay for this case
Guy Bedford
@guybedford
May 26 2017 20:38
but in terms of loading, it would load everything then run the tests
or does it do further dynamic loading on-demand when tests require different internal functionality?
Yury Delendik
@yurydelendik
May 26 2017 20:41
every test loads a web worker and pulls all (JS module) files
it's not different from running /web/viewer.html too -- no dynamic loading is happening on main or worker threads
two reasons I can think of: fetch() is not "following up" on its rep.text() promise, or this promise is garbage collected
Guy Bedford
@guybedford
May 26 2017 20:46
you could try replacing fetch with xhr
I think just adding self.fetch = undefined in the worker before systemjs should make it do that?
should at least hopefully separate the ability to say its a systemjs or browser bug
yurydelendik @yurydelendik thinks he did that -- same result, means something fishy with Promises
Guy Bedford
@guybedford
May 26 2017 20:55
@yurydelendik would it help to try disabling Ignition?
Yury Delendik
@yurydelendik
May 26 2017 20:57
what's Ignition? :)
Guy Bedford
@guybedford
May 26 2017 20:57
chrome://flags/#disable-v8-ignition-turbo
argh actually it's not enabled in Chrome 58
but it's the optimizing js compiler so shouldn't affect gc though
just grasping :P
maybe widen the browser tests on chrome, to see if it happens in Canary etc?
if it's an unstable part of the browser then should vary across browser releases
Yury Delendik
@yurydelendik
May 26 2017 21:00
heh, another thought, maybe xvfb corrupts memory or something
Guy Bedford
@guybedford
May 26 2017 21:00
are you 100% sure the fetch call is not returning at all?
it's an awful one... sorry I can't be more help
Yury Delendik
@yurydelendik
May 26 2017 21:02
it least my logs say that https://pastebin.mozilla.org/9022749 notice "fetch" lines for other than "colorspace" files
Guy Bedford
@guybedford
May 26 2017 21:02
and there's no sync error being thrown by fetch or anything odd like that?
Yury Delendik
@yurydelendik
May 26 2017 21:02
last dump is state of instantiateDeps()
this log is coming directly from devtools console, so I did not see any another errors/warns
Guy Bedford
@guybedford
May 26 2017 21:03
do you have a link to the logged loader?
Yury Delendik
@yurydelendik
May 26 2017 21:06
http://54.67.70.0:8877/system.src.js <-- work of two people both puzzled
Guy Bedford
@guybedford
May 26 2017 21:08
I wonder if it's just for fetch in this case particularly?
for xhr I wonder where it would have moved to
maybe do try canary
is there a promises polyfill in use?
Yury Delendik
@yurydelendik
May 26 2017 21:13
no polyfills in chrome case