Ports support for RELENG_4 (Was: Re: Question about ports builds)

[Doug Barton]

Mark Linimon wrote:

> The burden of trying to keep everything working on 4 i386 branches, 3
> amd64 branches, and 3 sparc64 branches is too high at this point, especially
> with the degree of drift in such things as header files and base compiler
> between -4 and -5.  Of course, most of these things can be fixed given
> sufficient maintainer and committer interest, but at some point you have
> to conclude that you're in the realm of diminishing returns.

I think you're right about that, and my preferred method of operation for
the ports that I maintain has been to try and test on RELENG_4 whenever
possible, but not let not testing stop me from updating a port that works on
7-current and 6-stable. What I've found is that in those rare cases where
there is an intersection between a port that is broken on RELENG_4 and users
that care about that, I'm notified fairly promptly. If not by users, I
usually get a krismail to the same effect. :) Thus, things in my little
world don't stay broken for too long.

All that said, I'd love to officially drop support for RELENG_4, but I think
that until we drop support for RELENG_4 in the base, dropping support for
the ports would break faith with our users. At the same time, I think that
some bit rot at the edges (of the seldom-used ports) is natural, and not to
be mourned.

> there are 206 legitimate build errors on i386-4 now; that doesn't include
> any port already marked as BROKEN.  That's quite high.

The way you break those numbers down is interesting. On i386 there are 206
errors on -4, 277 on -5, 119 on -6, and 151 on -7. I would be interested to
know what the percentage of overlap is ... in other words what percentage of
the 119 ports broken on -6 also comprise the broken ports on the other
releases. Based on a cursory examination of the errors on -4, I'd estimate
that percentage to be quite high. Assuming that it's 90%, that means that 99
ports are uniquely broken on -4, which is a very small percentage of the
packages that actually built (0.097%). IOW, there is actually less bit rot
(in terms of unintended brokenness) than I would have expected.

Doug

-- 

    This .signature sanitized for your protection