Re: [Evolution-hackers] [Fwd: [Fwd: A problem in libsoup]]

[Joe Shaw <joe ximian com>]

On Wed, 2003-04-09 at 14:35, Joe Shaw wrote:
> On Wed, 2003-04-09 at 12:49, Dan Winship wrote:
> > Ah. OK, so the problem is that we need to differentiate "not in use",
> > "sort of in use" (in_use TRUE but no message queued), and "definitely in
> > use".
> > 
> > Does this work for you? (And JP, does it work for you?)
> 
> It doesn't work for me, because conn->message is set by the time we hit
> connection_death().  Sorry, I haven't had any time to look into it
> beyond this yet.

Ok, this doesn't work because the connection is always set on the
message.

Digging a little further, it seems that for regular HTTP connections
through a proxy things work because control actually does make it to the
soup_transfer stuff because soup-queue.c:start_request() is called
directly from soup_queue_connect_cb().  For HTTPS connections, though,
control is given back to the main loop, so the HUP comes in before
start_request() and the subsequent
soup_transfer_write()/soup_transfer_read() callbacks are attached.

I'm not sure how to fix this yet.  What a mess.

Joe