811136 - l10n.js loadResource is causing multiple sync reflows

Status: RESOLVED WORKSFORME

(Firefox OS Graveyard :: Gaia::L10n, defect)

Description

[:Margaret Leibovic]

I noticed this in a profile of the settings app (this is a profile of master, not with any of my other patches applied). Hopefully we'll avoid going through this code path on app startup, but it would be good to investigate/improve it for the fallback case.

http://people.mozilla.com/~bgirard/cleopatra/#report=6f3e4afaa621941963a04001291bfa218a1e6702&filter=%255B%257B%2522type%2522%253A%2522RangeSampleFilter%2522%252C%2522start%2522%253A5046%252C%2522end%2522%253A5195%257D%252C%257B%2522type%2522%253A%2522RangeSampleFilter%2522%252C%2522start%2522%253A55%252C%2522end%2522%253A108%257D%255D&selection=%255B%2522Startup%253A%253AXRE_InitChildProcess%2522%252C%2522Input%253A%253AnsInputStreamPump%253A%253AOnInputStreamReady%2522%252C%2522Input%253A%253AnsInputStreamPump%253A%253AOnStateTransfer%2522%252C%2522content%253A%253AnsXMLHttpRequest%253A%253AOnStopRequest%2522%252C%2522nsEventDispatcher%253A%253ADispatch%2522%252C%2522js%253A%253ARunScript%2522%252C%2522parseResource%252FloadResource%252Fxhr.onreadystatechange%28%29%2520%2540%2520l10n.js%253A205%2522%252C%2522parseResource%252F%253C%28%29%2520%2540%2520l10n.js%253A222%2522%252C%2522parseProperties%28%29%2520%2540%2520l10n.js%253A136%2522%252C%2522parseRawLines%28%29%2520%2540%2520l10n.js%253A147%2522%252C%2522loadImport%28%29%2520%2540%2520l10n.js%253A187%2522%252C%2522loadResource%28%29%2520%2540%2520l10n.js%253A199%2522%252C%2522Input%253A%253AnsInputStreamPump%253A%253AOnInputStreamReady%2522%252C%2522Input%253A%253AnsInputStreamPump%253A%253AOnStateTransfer%2522%252C%2522content%253A%253AnsXMLHttpRequest%253A%253AOnStopRequest%2522%252C%2522nsEventDispatcher%253A%253ADispatch%2522%252C%2522js%253A%253ARunScript%2522%252C%2522parseResource%252FloadResource%252Fxhr.onreadystatechange%28%29%2520%2540%2520l10n.js%253A205%2522%252C%2522parseResource%252F%253C%28%29%2520%2540%2520l10n.js%253A222%2522%252C%2522parseProperties%28%29%2520%2540%2520l10n.js%253A136%2522%252C%2522parseRawLines%28%29%2520%2540%2520l10n.js%253A147%2522%252C%2522loadImport%28%29%2520%2540%2520l10n.js%253A187%2522%252C%2522loadResource%28%29%2520%2540%2520l10n.js%253A199%2522%252C%2522Timer%253A%253AFire%2522%252C%2522nsRefreshDriver%253A%253ANotify%2522%255D

Comment 1

[Axel Hecht [:Pike]]

maybe this gets cheaper if we remove the original strings from the html? Or we at least get to this earlier by not having to compute lengths etc for the first round of strings.

Comment 2

[:Margaret Leibovic]

The confusing thing is that it looks like these reflows are being triggered by different calls parseResource, and we should only be calling translateFragement once, after all of the different resources have been loaded. However, I'm by no means an expert at reading these profiles, so it could be interesting to try different changes and see what the profiler says. Like as a sanity check, we could get rid of the callback to translateFragment and see what happens.

Comment 3

[Zibi Braniecki [:zbraniecki][:gandalf]]

No, this is a single parseResource, it's just nested because first parseResource parses .ini file that triggers loading for .properties files and then parsing of it. (that's synchronous).

Comment 4

[Zibi Braniecki [:zbraniecki][:gandalf]]

We refactored the whole l10n.js file which makes all I/O async (bug 914414), and also moved to a single .json (instead of .ini -> .properties).

Can you reproduce this bug on master?

Comment 5

[:Margaret Leibovic]

(In reply to Zibi Braniecki [:gandalf] from comment #4)
> We refactored the whole l10n.js file which makes all I/O async (bug 914414),
> and also moved to a single .json (instead of .ini -> .properties).
> 
> Can you reproduce this bug on master?

I have not built or run gaia in a very long time. I'm willing to bet this bug isn't relevant anymore.