[Cocoa] Optimize ResourceResponse::platformLazyInit()

• Component: WebKit2
• Status: RESOLVED FIXED

[Chris Dumez]

Optimize ResourceResponse::platformLazyInit(). Most of the CPU time currently goes into getting the HTTP headers from CFNetwork:
Sample Count, Samples %, CPU %, Symbol
46, 0.0%, 0.0%, WebCore::initializeHTTPHeaders(WebCore::OnlyCommonHeaders, NSHTTPURLResponse*, WebCore::HTTPHeaderMap&) (in WebCore)
34, 0.0%, 0.0%,     HTTPHeaderDict::copyAsOrdinaryDict(__CFAllocator const*) const (in CFNetwork)
11, 0.0%, 0.0%,     CFDictionaryApplyFunction (in CoreFoundation)

We currently have 2 levels of initialization: CommonFieldsOnly & AllFields. With WebKit2, most ResourceResponses get sent over IPC and thus end up getting initialized twice, once with CommonFieldsOnly and then with AllFields.
This would cause us to call the expensive HTTPHeaderDict::copyAsOrdinaryDict() twice instead of once, simply to initialize the common HTTP headers first and then the uncommon ones later.

My proposal is to initialize all HTTP headers once, as soon as CommonFieldsOnly initialized is requested, so that we no longer copy all HTTP headers twice.

[Chris Dumez]

Created attachment 361509 [details]
Patch

[Alex Christensen]

Comment on attachment 361509 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=361509&action=review

> Source/WebCore/ChangeLog:22
> +        This patch updates ResourceResponse::platformLazyInit() to initialize all HTTP headers at once, as soon as CommonFieldsOnly

If we're doing this, why not just remove CommonFieldsOnly?

[Chris Dumez]

(In reply to Alex Christensen from comment #2)
> Comment on attachment 361509 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=361509&action=review
> 
> > Source/WebCore/ChangeLog:22
> > +        This patch updates ResourceResponse::platformLazyInit() to initialize all HTTP headers at once, as soon as CommonFieldsOnly
> 
> If we're doing this, why not just remove CommonFieldsOnly?

I would need to gather more data before doing such change. Also, this is likely not true for WebKit1 since no IPC.

[Chris Dumez]

Comment on attachment 361509 [details]
Patch

In a follow-up, I will look into removing the double stage lazy init altogether.

[WebKit Commit Bot]

Comment on attachment 361509 [details]
Patch

Clearing flags on attachment: 361509

Committed r241220: <https://trac.webkit.org/changeset/241220>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/47934216>