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

List:       kde-core-devel
Subject:    Re: kio_http speedup patch again
From:       Rik Hemsley <rik () kde ! org>
Date:       2002-03-30 15:27:53
[Download RAW message or body]

#if Dirk Mueller
> On Sam, 30 Mär 2002, Rik Hemsley wrote:
> > Here's my patch again, slightly altered from the last version, because I
> > don't think we need a CC_VerifyNonLazy - the code in khtml
> > (misc/loader.cpp) seems to force a reload at the right times anyway.
>
> I still don't like it.
>
> a) why is there that magic hardcoded 10 * 60 instead of using one of the
> constants in slave_defaults.h ?

Because that's temporary to show the idea. I would of course make it a
constant somewhere.

> b) Why is there a time limit at _all_ ? IMHO the images should either be
> reloaded every time or never.But making this based on some weird timing
> stuff is nonsense. Everybody will consider this behaviour, when it makes
> problems, to be a bug.

I don't think 'weird timing stuff' is 'nonsense'. I think that 
'If-Modified-Since' is not supposed to be accurate - it's meant to be a 
guide, and when implementing a client, you're supposed to allow some slack, 
to account for clock changes (drift/updating/etc.)

This patch simply adds some slack in kio_http, which is where it belongs,
IMO.

BTW, you didn't mention where you think this will cause problems. I have
yet to think of anything, so I'd be interested to hear.

> c) The only reason why it looks slow is because we're waiting for the
> validation before we show the result. If that is the slow part, then this
> is what we should fix. I don't understand why the actual I/O is that slow.

If you don't understand why the I/O is that slow, consider that we're doing
an HTTP request every time you click a link that takes you to a new page. Now 
think of people on slow-ish connections. Even if the connection isn't clogged
with other stuff at the time, the latency is big. In the order of seconds.

If you want to show the page, then update it when and if updated data is 
returned, how would you approach that ? Make two requests to the slave, one
for cached data and then one with CC_Verify ?

Showing the page and then changing it a second or two later doesn't sound like 
a good idea to me, anyway. This would be confusing for the user.

Rik

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