506431 - UTF8 to UTF16 conversion showing up in startup profiles

Status: RESOLVED INCOMPLETE

(Core :: General, defect)

Description

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

All of our parsers are UTF16, but all of our files are UTF8.  The conversion is showing up between 5% and 8% in a warm startup profile.

There are a couple of ways to fix this, and we should look at all of them.

- Speed up UTF8 to UTF16 conversion.  This is straight C code now, and could probably be made much faster -- bug 506430

- Fastload XBL -- bug 94199 (?)

- Fastload CSS -- bug 196843

Comment 1

[Benjamin Smedberg]

Is it worth trying to make our chrome UTF16? That's probably easy to do at build-time, although you'll pay an I/O price...

Comment 2

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Yeah, I thought about that... tried a quick hack using iconv, but that didn't work out too well.  It would probably help in the warm start case, but hurt in the IO-bound cold start case, but I'm just guessing there.

Comment 3

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Vlad, have you rerun these numbers since we fixed Bug 506430?

Comment 4

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Nope, haven't touched this in a while.  It's worth a look.

Comment 5

[Worcester12345]

(In reply to comment #4)
> Nope, haven't touched this in a while.  It's worth a look.

Did you ever take that look?

Comment 6

[Justin Lebar (not reading bugmail)]

> Did you ever take that look?

Sounds like he didn't, and he now doesn't work for Mozilla.

Glandium might be interested in this bug, however.

Comment 7

[Mike Hommey [:glandium]]

I don't remember having spotted so much time spent on such conversions during warm startup when i looked some time in Q3 or Q4 2010 ; but I can surely take a look.

Comment 8

[Marco Castelluccio [:marco]]

what about UTF8 parsers instead of UTF16?

Comment 9

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

I think this bug has outlived its usefulness.