2 Dec 2003 15:26
Re: PATCH: Better cross-domain cookie detection [BR 66090]
Dawit A. <adawit <at> kde.org>
2003-12-02 14:26:22 GMT
2003-12-02 14:26:22 GMT
On Monday 01 December 2003 07:53, Waldo Bastian wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > On Mon December 01 2003 13:39, Dawit A. wrote: > > However, before committing this I have to track down a bigger problem. > > For some reason none of the meta data set in KHTMLPart::openURL makes it > > down to the http io-slave. The weird thing is the ones set in the loader > > do get there everytime. Have no clue why and I am attempting to debug > > it... > > The request is usually actually started by doing: > emit d->m_extension->openURLRequest > > Then konqueror starts the job, looks at the mimetype that comes back and > decides to hand the URL over to the khtml_part//openURL which then starts a > new job which will get the original slave+response handed over by the > scheduler. Ahhh... Good thing you mentioned that. I found the problem. It is konqueror's attempt to detect the mimetype that was causing this issue. The "cross-domain" meta-data needs to be added to BrowserRun::scanFile and possibly to KRun::scanFile as well. That fixes my problem. However, after several hours of investigation, I am completely dumb founded about how retrieving a URL works when it is a result of the user typing into the location bar. Konquerror makes the original request. Determines the content-type using BrowserRun which puts the job on hold. Konqueror uses the mime-type information to invoke the appropriate part/view/app to handle the request, in our case KHTMLPart. KHTMLPart then invokes openURL which makes another KIO request I guess using the io-slave on hold. This is where it gets interesting. I see the meta-data set by KHTMLPart::openURL getting all the way down to SlaveBase, grand-parent of kio_http :) However, so far as I can tell no re-request is made by kio_http! Even if "Use Cache" is completely disabled! I presume this is by design to avoid duplicating each request, right ? This behavior also results in all the meta-data sent by KHTMLPart to be completely useless (ignored). For example, doing View->Document Information absolutely shows nothing here. Hmm... -- -- Regards, Dawit A. "Preach what you practice, practice what you preach"