Re: xassert in pump.c

Sure, request entities present special problems and unless dealt with in
a clean manner there will always be strange situations and very hard to
debug.

The change I proposed was to kill the pump module by having all logic
about request message delimiting in client_side.c, and then invent some
kind of "abstract" data flow for the request entities from there to the
protocol implementations.

What is bad with the pump module is that it short-circuits the client
and server sides, making the locic at both sides a bit messed up and
hard to follow, especially so in error situations where one of the sides
have failed/aborted. For simple non-persistent PUT/POST requests this
might be OK as there is not much logic to care for in the first place,
but for FTP or persistent connections (as seen for example in the NTLM
patches) the situation gets quite hairy.

/Henrik

Duane Wessels wrote:
>
> On Fri, 2 Jun 2000, Henrik Nordstrom wrote:
>
> > Argh! Not the pump module again!
> >
> > Duane, do you still think that this beast shouldn't be killed?
>
> seriously...
>
> sure, we (you) can kill it.
>
> My feeling is that request bodies present special problems and
> circumstances. Whether you call it "pump" or whatever, it still may be
> difficult to get right. Since their occur infrequently, they are hard
> to debug. Since they are sort of complicated, they are hard to
> simulate.
Received on Sat Jun 03 2000 - 04:21:31 MDT