[Dnsmasq-discuss] A test release?

Matthias Andree matthias.andree at gmx.de
Sat Mar 31 18:54:32 BST 2007


Simon Kelley <simon at thekelleys.org.uk> writes:

> In the past, I have not generally made "release candidate" releases of
> dnsmasq. In general this has worked OK, but it has lead to some
> occasions when "full" releases have been superceded quickly and
> withdrawn because of bugs and regressions. The last release took from
> 2.36 to 2.38 to really stabilise, for instance.
>
> Now that the list exists, and has a set of regulars, I think it might be
> time to try making a release candidate and announcing it to the list,
> with a test time of a couple of weeks or so, before making a final
> release and announcing it to the world.
>
> What do people think? Could you/would you test a RC binary?

Simon,

while I don't pass for a regular on your list, because I mostly use it
to harvest release announcements and catch up the rest, I think I can
contribute to your decisionmaking (I deliberately ignored the followups,
blame me for any duplication).

I've often found myself in the situation of considering -rc or not, and
my observations are:

- rc tarballs get tested by bug reporters if you ask them

- else, people wait for the real thing, because they usually pick
  packages from/for their distro

- thus, rc don't help much cutting down on regressions

Current precedent:

fetchmail 6.3.6 set out as fixing important security issues (protocol
implementation errors (fallback behavior) leading to password
leaks). The -rc releases leading to 6.3.6 were good to shake out
remaining bugs not addressed by the early fixes.

However, the parts of the code that I broke were only reported after the
6.3.6 release, making necessary a 6.3.7 release, and further 6.3.6
regressions were reported against 6.3.8, and I plan to put out 6.3.8-rc3
tonight so that my faithful translators can catch up.

As much as we loathe releasing banana software, at the end of the day we
end up doing it anyways, albeit unvoluntarily, because -rc tests will
happen scarcely.

I hope this hasn't discouraged you in any way, but I have rarely seen
bug/regression reports against -rc releases even in long -rc phases.

(Oh, and -rc wore out for Dovecot, too, since it's at rc29... I
understand most of how this came to pass, but people are irritated
nonetheless.)

HTH
Matthias


________
¹ the alternative would have been to do closed-source-style plugs that
  detect and ward off a particular attack, rather than fixing the root
  of the problem - I chose to go for the latter.

-- 
Matthias Andree



More information about the Dnsmasq-discuss mailing list