Re: Squid-2.5 bugs to kill

On Wednesday 13 August 2003 03.48, Robert Collins wrote:

> Thats certainly possible. I'm hoping that we can start shorter
> release cycles once 3.0 is out - the 2 week heuristic seems a
> reasonable one for release quality - and we have a much better base
> to perform incremental improvements on ...

Some stats from Squid-2.5 based on the CVS changelog:

Most critical bug reports arrived 1-6 months after the release, but
there is still major bugs reported every now and then.

Week 1, 2; mainly obvious issues which we really should have detected
and fixed before going STABLE, mostly packaging related (missing
files, misleading notes etc). The only real bug fix is an attempt to
make proxy_auth work for delay_access.

Week 3: one important bug introduced (delay_pools memory allocation
got broken, fixed again a few days later), the rest was more
cosmetics and minor fixes such as some squid_ldap_auth options not
working. The other bug fix was closing of referer_log on shutdown
which is mainly a cosmetic bug. (complained about open file on
shutdown)

Week 4: there was only cosmetics apart from the delay_pools
anti-breakage mentioned above.

Week 5: a couple interesting bug fixes. external_acl issues with "odd"
data, AUFS corruption, WCCP state stability, and a lot of cosmetic
bug fixes.

Week 6: Almost nothing.

Week 7: Major bug. AUFS corruption. Some minor bug fixes.
"request_entities on/off" added. Security update to use POST in
cachemgr.cgi when submittin the admin password. Filter to block
proxying of NTLM. More AUFS bug fixes triggered by the corruption
bugfix. Support for spaces in included acl data.

Week 8: Minor aufs fix (continuation of the above). wb_group update.

Week 9: squid_ldap_group update. CONNECT issue bugfix.

Week 10: Nothing.

Week 11: More packaging bug fixes to increase portability.

Week 12: Nothing

Week 13: Nothing

Week 14: minor aufs speedup

Week 15: squid_ldap_group update. aufs improvements (pipe completetion
mechanism). More packaging bug fixes (incorrect paths in scripts
etc). Security fixes in squidclient.

Week 16: Instability if requests is received while rebuilding the
cache. offline_mode not working well. chroot_dir simplified to check
paths inside the chroot, not outside.

Week 17: squid_ldap_group updated. --disable-hostname-checks configure
option.

more of the history can be found in the CVS logs or the 2.5 patches
page..

> Oh - cool. Well, like I said, there is hope for ntlmv2 in 2.5, a
> patch is in the works.

Excellent.

> Squid never generates challenges, so there is nothing synthetic
> going on

Currently all challenges are synthetic as the helpers do not have
access to the clients NEGOTIATE packet..

> - as long as the reuse count is forced to zero when ntlmv2
> is in play, we should be fully compatible with whatever the samba
> guys come up with for future NTLM support.

Great. Looking forward to the results. It would indeed be nice if we
could have NTLM running correcly (but ineefficient). Would save a lot
of headache I think. However, this is primarily a Squid-3 candidate
in my eyes but I won't object to have this done in 2.5 if you feel it
is good.

Regards
Henrik
Received on Wed Aug 13 2003 - 01:52:02 MDT