Regression(r220817): We should only copy the original request headers for Ping loads

• Component: WebCore Misc.
• Status: RESOLVED FIXED

[Chris Dumez]

We should only copy the original request headers for Ping loads (Beacon at the moment), not for all CachedResourceRequests.

[Chris Dumez]

Created attachment 318398 [details]
Patch

[youenn fablet]

Comment on attachment 318398 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=318398&action=review

> Source/WebCore/loader/cache/CachedResourceLoader.cpp:836
> +        resource->setOriginalRequestHeaders(WTFMove(originalRequestHeaders));

We should probably also set the headers in the case of DeferOption::Defer.
If not, originalRequestHeaders could just be a parameter passed to CachedResource::load().

[Chris Dumez]

Created attachment 318406 [details]
Patch

[youenn fablet]

Comment on attachment 318406 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=318406&action=review

> Source/WebCore/loader/cache/CachedResource.cpp:276
> +            platformStrategies()->loaderStrategy()->createPingHandle(frame.loader().networkingContext(), request, HTTPHeaderMap(*m_originalRequestHeaders), *m_origin, contentSecurityPolicy, m_options);

IIRC, IPC is taking const& parameters.
WK2 createPingHandle could probably be implemented with just a const&.

[youenn fablet]

(In reply to youenn fablet from comment #4)
> Comment on attachment 318406 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=318406&action=review
> 
> > Source/WebCore/loader/cache/CachedResource.cpp:276
> > +            platformStrategies()->loaderStrategy()->createPingHandle(frame.loader().networkingContext(), request, HTTPHeaderMap(*m_originalRequestHeaders), *m_origin, contentSecurityPolicy, m_options);
> 
> IIRC, IPC is taking const& parameters.
> WK2 createPingHandle could probably be implemented with just a const&.

So that we remove this extra header copy.

[Chris Dumez]

(In reply to youenn fablet from comment #5)
> (In reply to youenn fablet from comment #4)
> > Comment on attachment 318406 [details]
> > Patch
> > 
> > View in context:
> > https://bugs.webkit.org/attachment.cgi?id=318406&action=review
> > 
> > > Source/WebCore/loader/cache/CachedResource.cpp:276
> > > +            platformStrategies()->loaderStrategy()->createPingHandle(frame.loader().networkingContext(), request, HTTPHeaderMap(*m_originalRequestHeaders), *m_origin, contentSecurityPolicy, m_options);
> > 
> > IIRC, IPC is taking const& parameters.
> > WK2 createPingHandle could probably be implemented with just a const&.
> 
> So that we remove this extra header copy.

WK2:
loadParameters.originalRequestHeaders = WTFMove(originalRequestHeaders);

[youenn fablet]

> WK2:
> loadParameters.originalRequestHeaders = WTFMove(originalRequestHeaders);

Would the following work?
- Pass originalRequestHeaders as a direct parameter of NetworkConnectionToWebProcess::LoadPing. It is not used for other loads.
- Create the message with something like Messages::NetworkConnectionToWebProcess::LoadPing(loadParameters, originalRequestHeaders)

IIRC, Messages structures only needs const&.

[Chris Dumez]

(In reply to youenn fablet from comment #7)
> > WK2:
> > loadParameters.originalRequestHeaders = WTFMove(originalRequestHeaders);
> 
> Would the following work?
> - Pass originalRequestHeaders as a direct parameter of
> NetworkConnectionToWebProcess::LoadPing. It is not used for other loads.
> - Create the message with something like
> Messages::NetworkConnectionToWebProcess::LoadPing(loadParameters,
> originalRequestHeaders)
> 
> IIRC, Messages structures only needs const&.

Yes, I have confirmed Messages structures only need const&. I can do what you suggest although I am unconvinced this is a worthwhile optimization.

[youenn fablet]

> Yes, I have confirmed Messages structures only need const&. I can do what
> you suggest although I am unconvinced this is a worthwhile optimization.

Agreed, the main point of this patch is to limit the header copy to beacon.

As a follow-up of that patch, I think it makes sense to split the original headers from NetworkLoaderParameters since they are related to ping loads only.
It should be easy to add it back when implementing keepalive for fetch API loads, if we find these headers to be needed.

[Chris Dumez]

(In reply to youenn fablet from comment #9)
> > Yes, I have confirmed Messages structures only need const&. I can do what
> > you suggest although I am unconvinced this is a worthwhile optimization.
> 
> Agreed, the main point of this patch is to limit the header copy to beacon.
> 
> As a follow-up of that patch, I think it makes sense to split the original
> headers from NetworkLoaderParameters since they are related to ping loads
> only.
> It should be easy to add it back when implementing keepalive for fetch API
> loads, if we find these headers to be needed.

It's trivial, I'll do another iteration.

[Chris Dumez]

Created attachment 318423 [details]
Patch

[Chris Dumez]

Comment on attachment 318423 [details]
Patch

Clearing flags on attachment: 318423

Committed r220888: <http://trac.webkit.org/changeset/220888>

[Chris Dumez]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/33952143>