Re: Squid-2.5 bugs to kill

On Tuesday 12 August 2003 10.30, Robert Collins wrote:

> I won't be backporting, nor trying to find the cause of, the
> proxy_auth issue with 2.5 delay pools.

Fine. Someone else have to do it then if this is to be fixed for
Squid-2.5. If not we have to live with the bugs. Not a big deal for
us, but maybe to some of our users..

The delay_access bug is mostly on the list due to it causing sporadic
assertion failures, not due to proxy_auth not working well in
delay_access.

> Re the transition period, 3.0 should be a whole heap better than
> 2.3 was. And the whole point of new versions is to fix things that
> would require destabilising changes in stable versions, as well as
> adding new features.

Indeed. However, from a quality point of view it is a sad truth that
it will most likely take considerably time before Squid-3.0 is up to
the level of the current Squid-2.5 tree. This is from experience with
earlier releases and math on the amount of changes..

What I am attempting with Squid-2.5 is just to get rid of the bugs
which have noticeable impact on the intended use of Squid-2.5 to have
a rock-solid Squid-2.5 release. Authentication not working properly
or crashing Squid is a quite noticeable thing. The purpose of having
a rock-solid Squid-2.5 is actually to allow us to focus on Squid-3.0,
not to take resources away from Squid-3.0.

The maintenance of Squid-2.5 has stabilized considerably lately, and
things are now looking very good I think. I am quite happy with the
current 2.5 source tree even with these known bugs, but these are the
bugs which I think users have trouble with, and I do not think they
should be very hard to fix for Squid-2.5 without having to rewrite
stuff.

There is mostly NTLM related issues now, and personally I do not
consider NTLM truly suitable for production use until we have fully
got rid of the synthetic challenges, have support for overlapping
requests to not run out of helpers and are using the Samba NTLM
implementation via the Samba provided helper. But this is me and I
might be a bit pedantic on protocol issues..

As I know there is lots of people using NTLM in it's current state I
an not too keen on making changes in the already fragile synthetic
challenge procedures during Squid-2.5, but I do acknowledge there is
issues with the current challenges used preventing many people from
using it today and may be convinced to make changes if properly
backed up with indications that they won't break current working
setups. Personally I would prefer it time was spent on getting the
NTLM procedure correctly done in squid-3.X rather than to try to
patch up the known to be broken approach of Squid-2.5 however.

Regards
Henrik
Received on Tue Aug 12 2003 - 17:30:02 MDT