[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-core-devel
Subject:    Re: Fixing kio_http
From:       Rik Hemsley <rik () kde ! org>
Date:       2001-06-29 0:55:28
[Download RAW message or body]

#if Dawit Alemayehu
> On Thursday 28 June 2001 18:39, Rik Hemsley wrote:
> > My quick hack to ensure that there is a maximum of 2 http slaves per
> > host:
> >
> > KIO::Scheduler::ProtocolInfo *
> > KIO::Scheduler::ProtocolInfoDict::get(const QString &protocol)
> > {
> >     ProtocolInfo *info = find(protocol);
> >     if (!info)
> >     {
> >         info = new ProtocolInfo;
> > +        if ("http" == protocol.lower())
> > +        {
> > +          info->maxSlaves = 1;
> > +        }
> >         info->protocol = protocol;
> >         insert(protocol, info);
> >     }
> >     return info;
> > }
> >
> > Yes, I know that should be 2, but there's an off-by-one error
> > somewhere... the maximum is normally set to 3, which in real life is
> > 4.
>
> Hi Rik,
>
> Does the scheduler properly cache the other requests ?  I mean if there
> are more than two request what happens ?

They're just queued. This patch just limits the number of concurrent 
connections made to one site.

> A typical scenario for this
> is khtml downloading image links within a page in the background ?   I
> am not sure how this is handled.  Also you might want to change that to
> protocol.lower().startsWith("http") so that it also applies to https
> connections :)

Well, I'm not sure what to do about https - it has to be handled 
specially in various ways.

> Anyways, this is indeed a quick hack as you said :) We definitely need
> to find a more generic solution any/all protocol can use and the user
> can adjust :)  Waldo said he was going to  implement something for
> this, but he's on vacation this week I guess.  Otherwise, another
> solution which Waldo did not like  is to use connection-oriented mode
> in khtml and have it manage its own io-slave connections which would be
> more work for khtml, but also more flexiable since it can do whatever
> it wants that way.

Yes, this should be more generic, but it would be nice to fix this now 
and make it generic later IMO.

> BTW, the two connection is a recommendation and not a requirement,
> right ?  Does the spec say MUST or SHOULD when it comes to this topic ?

I don't know, but it's a bit impolite to use > 2 connections anyway. 
There are enough stupid programs around that do things like open multiple 
connections to one server in order to theoretically download a file 
faster. Grr.

Rik

[prev in list] [next in list] [prev in thread] [next in thread]