Re: 3.3-branch QA assessment

[Jan Hubicka <jh at suse dot cz>]

> >>>>> "Joe" == Joe Buck <jbuck@synopsys.com> writes:
> 
> Joe> There are 118 high-priority PRs.  One, about Java race
> Joe> conditions, doesn't look like it is really a regression
> 
> Actually, it is.  The new direct-threaded interpreter has the races,
> the old, slower interpreter did not.  So in theory it is possible for
> code which worked on the old interpreter to fail with the new
> interpreter.  If need be, though, we can turn off direct threading on
> the release branch.  It all depends on whether I can find the time to
> fix this bug soon.
> 
> Joe> I'm afraid folks are going to hate me, but I'm seriously
> Joe> suggesting just dumping the 3.3 branch, unless someone can
> Joe> explain to me why I'm completely wrong.
> 
> Doesn't this imply that our development plan doesn't model reality
> very well?
> 
> My impression is that much more work is being done on the trunk than
> on the 3.3 branch (not to discount the work that has been done on
> 3.3).  Would you say that is accurate?  Suppose we made a 3.4 branch
> and tried to stabilize it.  Would the situation be any different?  I
> think people would most likely continue to focus on their new
> development and leave the new release branch to mostly languish.

Actually my experience with building aplications with 3.3 is that most
of bugs went away in recent weeks.  3.4 does not even build our SPEC
right now, so it looks funny to discuss that 3.4 is more stable than
3.3.  3.3 does not seem to be having more problems than 3.2 in practice
as far as I can tell.

My impression is that the difference in PR is probably due to huge
efforts put into stabilizing C++ parser at one side and testing 3.3 on
the other and I don't think it is good metric of the reliablity of
eighter branch.

Honza
> 
> Tom