1346974 - stylo: base URI handling for inline style sets is super-slow

Status: RESOLVED FIXED

(Core :: CSS Parsing and Computation, defect, P1)

Description

[Boris Zbarsky [:bzbarsky]]

This is not going to be a surprise, but....

I redid a profile for the testcase from bug 1344131, now that we fixed the issues the first profile found.  The new time is 2.5x slower than the old time.

Most obviously, this is because we started passing in a base URI for this stuff.  On the new (2.5x slower) baseline, we have:

10% RefPtr<nsMainThreadPtrHolder<nsIURI> release/dtor -- there's free() under there, so you lose.

6% GeckoParserExtraData constructor, mostly doing malloc.

5% RefPtr<nsMainThreadPtrHolder<nsIPrincipal> >::~RefPtr() -- again, free().

26% url::ParseOptions::parse.  I'm not even going to look under there.

Anyway, all told that's 47%.  So fixing that will at least get us down to about where we used to be and then we can re-profile.

Bug 1343964 tracks that to some extent.

Comment 1

[Bobby Holley (:bholley)]

(In reply to Boris Zbarsky [:bz] (still a bit busy) (if a patch has no decent message, automatic r-) from comment #0)
> This is not going to be a surprise, but....
> 
> I redid a profile for the testcase from bug 1344131, now that we fixed the
> issues the first profile found.  The new time is 2.5x slower than the old
> time.
> 
> Most obviously, this is because we started passing in a base URI for this
> stuff.  On the new (2.5x slower) baseline, we have:
> 
> 10% RefPtr<nsMainThreadPtrHolder<nsIURI> release/dtor -- there's free()
> under there, so you lose.
> 
> 6% GeckoParserExtraData constructor, mostly doing malloc.
> 
> 5% RefPtr<nsMainThreadPtrHolder<nsIPrincipal> >::~RefPtr() -- again, free().
> 
> 26% url::ParseOptions::parse.  I'm not even going to look under there. 

This last bit should be fixed by my patches in bug 1347435.

> Anyway, all told that's 47%.  So fixing that will at least get us down to
> about where we used to be and then we can re-profile.
> 
> Bug 1343964 tracks that to some extent.

Yeah, I agree.

Comment 2

[Boris Zbarsky [:bzbarsky]]

Fixed by bug 1343964.