Re: [Libreoffice-qa] [ANN] LibreOffice 3.4.6 RC2 test builds available

Bjoern Michaelsen píše v St 14. 03. 2012 v 18:38 +0100:
> While being open to further concrete proposals for improvement, I dont think
> there is a fundamental flaw with the release concept itself. The quality of
> 3.5.0 and and 3.5.1 show that the number of regressions in 3.4 were a one-time
> outlier. We have to make sure it does not happen again, but I see no general
> flaw with our release model.
>
> We need to be sure to detect and pinpoint regressions earlier and more precise
> than in 3.4 -- and we have that with unittests, subsequenttests, onegit and
> bibisect -- all means that we were still missing in the 3.4 series.
>
> Those were the parts that needed changing and they are changed already. It
> doesnt end there though: With gerrit and the currently ongoing efforts in QA in
> addition to the tools named above, we are getting even better early warning and
> enable ourselves to quickly and confidently fix regressions.

I agreed with everything that was said by Bjoern. I just want to add
that we need to be realistic:

1. Every software contains bugs and regressions. See "Every non trivial
program has at least one bug" at
http://www.murphys-laws.com/murphy/murphy-computer.html

2. IMHO, nobody is able to fix all regressions because any fix
potentially creates another regression.

3. Some regressions are not easy to fix because the new code is
completely different. We always need to consider the seriousness of the
regressions and the cost of the fix.

4. IMHO, we could not maintain too old code by volunteers. It is
frustrating and very expensive. The code is evolving. People forget the
old structures and functions, so they need to investigate a lot of stuff
when they did into the history. Also the testing is expensive unless you
do trivial changes and you are sure that it does not break anything. In
the real word, it is done only by a company if a customers pays for it.

5. We need 3.4.6 because 3.5 is too young and we still might miss some
regressions and nasty bugs. We review the fixes when backporting, so
3.4.6 should not need extensive testing. Some basic tests should be
enough.

You said that 3.4.6 would have been more or less useless for many users
because of the regressions. I do not think so. LO-3.4 is being used by
many users, so it is usable. Some users are conservative and do not want
to risk 3.5 too early. 3.4.6 brings some very useful fixes for these
people.

We might change our mind and provide 3.4.7 if people report too many
serious bugs in 3.5 and it takes longer time to make it usable. On the
other hand, it seems that 3.5 is going to be better that 3.4. I do not
see any reason to spend more time on this historic 3.4 code line right
now.

As described by Bjoern, we do a lot to improve the code, tools, and
processes. LO-3.5 seems to be better than 3.4, so we seems to go the
right direction.


Best Regards,
Petr