Re: meeting notes!

From: Robert Collins <robert.collins@dont-contact.us>
Date: Tue, 6 Feb 2001 22:33:19 +1100

The fix should be for headers to be metadata and their parsed presence the test in clientCheckTransferDone rather than checking if
we know their size :]

Seriously, as I can't track the bytes written via vprintf in Packer.c... I'm calling httpReplyParse in httpReplySwapOut with the
first mem_node of the object. A _huge_ hack and one I'm not terribly happy about.

I had a look to see if there was better logic I could in in clientCheckTransferDone (and there is in two fashions...) but done that
can be done _today_.

The two better checkers are: header presence as above,
and recieving a EOF from the store which makes clientCheckTransferDone 100% irrelevant and dead code.

I'm still discussing with you changes to the store Append API, and metadata headers is a non-trivial topic. However authentication
with TE cannot work until I squash this bug ... so I chose the ugly duckling way for now

Thoughts?

Rob

----- Original Message -----
From: "Adrian Chadd" <adrian@creative.net.au>
To: "Robert Collins" <robert.collins@itdomain.com.au>
Cc: <squid-dev@squid-cache.org>
Sent: Tuesday, February 06, 2001 10:26 PM
Subject: Re: meeting notes!

> On Tue, Feb 06, 2001, Robert Collins wrote:
> >
> > does that impact error pages at all? authentication for one wioll be killed by that - making some unit tests difficult :-]
> >
>
> Erm, it shouldn't. error pages aren't served as storage objects, there
> is a chunk of code which reads them and appends them through the normal
> channels.
>
> > I ask because I just found a nasty bug that has been *bugging* me for weeks with te - on errors IE would hang: the error data's
> > *http_reply->hdr_sz was set 0 - httpReplyParse is never called for error objects, thus the checktransferdone function always
returns
> > not done, and te never writes the termination 0\r\n\r\n
> > - I'm fixing this now but should I bother backporting the fix to HEAD, or will is it another special case that won't be needed
as a
> > result of the work you are doing in modio?
> >
>
> It depends what thefix is. Are you going to call httpReplyParse
> because it shoudl be, or what ?
>
>
> Adrian
>
>
> --
> Adrian Chadd "Romance novel?"
> <adrian@creative.net.au> "Girl Porn."
> - http://www.sinfest.net/d/20010202.html
>
Received on Tue Feb 06 2001 - 04:32:15 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:13:28 MST