Ability for maintainers to update own ports

[Charles Swiger]

On Nov 10, 2003, at 9:19 PM, Kris Kennaway wrote:
> On Mon, Nov 10, 2003 at 11:16:51PM +0100, Oliver Eikemeier wrote:
>> The first can be satisfied with something like pkgsrc-wip, and I 
>> always
>> wondered why we don't have a ports-FRESH and ports-TESTED, like we 
>> have
>> -CURRENT and -STABLE.
>
> Because even with a single, unbranched ports collection, committers
> can't keep it in working order without significant ongoing effort.  On
> average, several ports become broken on one of the supported
> architectures and versions, *each day*.  [ ... ]

Thanks for your insight into the scope of the problem, Kris: as one of 
the <portmgr>'s, I suspect that you've dealt with more than a few 
variations on how the ports system can break.

More to the point, I regretfully agree that branching ports probably 
would hurt more than it would help.  There is the issue of "MFC'ing" a 
port from HEAD to the -STABLE branch (or -TESTED, or whatever name)-- 
dealing with CVS branches is hard (or at least more error-prone) for 
even for developers who are relatively experienced.  It would add a 
significant amount overhead to the CVS repo, and multiply the number of 
environments to be tested, the number of failures, and so forth, just 
as you've said.

> If you start adding more branches to the ports collection, you're
> going to multiply the possible failure modes, and the net result will
> be that the number of errors accumulating in the ports collection will
> more than multiply.

Branching the ports collection is not a helpful solution to the problem 
of speeding up the rate at which ports are committed.

Branching ports might help out people who try to update their ports 
while running older versions of FreeBSD (yes, I understand that 
supporting older versions is at a low priority, but this would address 
some user expectations/complaints), and it might smooth "cataclysmic 
events" involving changes to the ports infrastructure, changes to 
-CURRENT which break lots of ports, etc.

However, I'm not sure we need to branch the ports CVS repository to 
address these issues: what happens if we use CVS tags to indicate which 
OS versions and/or hardware architectures a port works on?  One could 
then have successful builds on bento perform the tagging in an 
automated fashion.  This might provide many of the benefits, without 
incurring a lot of additional work on the part of committers.

-- 
-Chuck