New rc.d code merge timing (Was: Re: Portupgrade confused about
linimon at lonesome.com
Sat Jan 7 16:16:44 PST 2006
On Sat, Jan 07, 2006 at 07:02:23PM -0500, Mike Jakubik wrote:
> Shame on me for running -STABLE in production then. I've always done
> this though, but fortunately its never caused any serious problems. It
> is called STABLE after all.
There was a lot more thought went into this than you are giving us credit
for. I was the person on portmgr who originally resisted this change. But
after talking through it thoroughly with Doug I finally agreed that there
was not going to be a better time to do this than now. We had a two-month
window before the 6.1 cycle and then we are going to have what may be a
prolonged release cycle because we are going to try to do 5.5 and 6.1
Doug did indeed do testing on some of the major ports on -CURRENT and sent
mail to, and worked with, the maintainers to get as much of this right as
could be possible without more widespread testing. And with that 2-month
window, that was going to be as much testing as we can get.
The idea is to make 6.1 rock-solid, and with these changes, so that things
will be simpler and more reliable in the future. If I didn't think we had
a very good chance to do it before 6.1, I would have attempted to stop the MFC.
You need to have the background context that Doug and I have butted heads
before on other issues, so I'm not automatically going to agree with him on
anything. But in this situation he presented his case well, he did work
ahead of time, and has continued to follow-up everything that has been
brought to his attention.
It's unfortunate that we had to take 1 step backwards to take 2 steps
forwards but that's life. The portmgr team spends a great deal of effort
in running regression tests of bsd.*.mk to avoid these kinds of regressions,
but even this kind of work cannot give you a 100% guarantee that nothing
will go wrong.
> One question though, were the porters notified of the upcoming changes?
> And were they provided with the appropriate information to write the RC
> scripts before the merge?
Yes. Please see the archives (although some of this went on, as does a lot
of work on FreeBSD, behind-the-scenes).
Honestly, I believe that this was the best we could do in the absence of
a branched ports tree (which IMHO we do not have the manpower to maintain
in a timely fashion), and even if we had had such a thing, it _still_ might
not have been sufficient to ensure widespread testing.
More information about the freebsd-ports