What follows is a summary of the discussions held on IRC tonight.
As already announced Henrik will be at the Linux-Kongress in a few weeks. The question on if he is presenting something or just attending arised and he is attending the Linux-Kongress (or actually the Netfilter Workshop held in conjunction) and offers to hold a BOF session on Squid if there is interest.
Then is was realised several of the Squid developers quite often were nearby each other but still has not met in person, mainly due to not knowing the location of each other. It was decided to try to arrange a get-to-gether in probably in December (Henrik, Robert, Kinkie).
In addition there was some apparent confusion on where Henrik lives. For the record he (I) lives in Stockholm, Sweden. Not Russia.
Then the two pending NTLM patches (Bug #910, #994) and their proper testing (or rather lack thereof) was discussed extensively. It was concluded these fixes are becoming relatively important and it should be given yet another attempt to get them into the release before the 2.5.STABLE7 release which is supposed to be the final Squid-2.5 release.
Next topic was Squid-3.0. All agree the Squid-3.0 development need to get going again after a long period of slow (or none) progress. Things are looking more promising for this to happen now however.
In correlation to this there will be attempts to arrange code/bug-sprints where some of the developers gets together electronically and physically and selecting one problem which needs to be solved. As a good candidate Bug #7 was identified (update of header data on cache revalidation), and there was some short discussion on how this can be solved. It appears most of the infrastructure is in place in Squid-3.0, but some pieces are still missing. Until the missing pieces are there an viable option is to spool the data to a new object is one way out, and will also allow for the Squid-3.0 interfaces to stabilise a bit without having to rush something to fix Bug #7.
Then at the very end (technically after I left writing this summary) the issue about NTLM and too many queued requests error came up. Henrik reminded that the correct way out is to allow for a larger number of challenges in circulation without depending on very many helper instances. This requires the stateful helper protocol to be extended with support for overlapping lookups much in the same manner as is already present for the stateless helper interfaces (redirectors, external acl lookups, basic auth, etc). This has already been discussed numerous times already.
Regards Henrik
