Re: Debian doesn't have to be slower than time.
On Sat, Feb 16, 2002 at 07:44:35PM -0700, Joel Baker wrote:
> Not to mention things like, oh... /usr/share/doc. How long has it been,
> since that cutover was started? Limited duration legacy support allows for
> a great deal of flexibility.
> Release foo: declare /usr/share/doc the preferred method, ensure that the
> old directory is supported sanely, and file wishlist bugs on all packages
> that use /usr/doc.
> Release foo+1: make the use of /usr/doc a lintian warning, and upgrade any
> outstanding bugs from release foo.
> Release foo+2: remove legacy support; any use of /usr/doc is now a serious
> enough bug to prevent package release (and should indicate a problem with
> the maintenence, most likely...)
> Break. The. Task. Up.
This is what we tried to do, although the way the /usr/doc ->
/usr/share/doc task was broken up was a little different. (We're
currently approaching the end of phase 2 of 3.) Believe me, it's still
hard. There is no way you can get around the fact that every package has
to be rebuilt in this situation. While this doesn't sound like a big
deal, I've spent something approaching a solid month of my time (imagine
I didn't sleep ...) on this single task over the last year, and it is
not trivial no matter how you split it up.
To take an example: /usr/doc became a lintian warning in July 1999. That
in itself should suggest that the distance between release foo+1 and
release foo+2 is rather greater than you thought.
The main problem is that a large proportion of maintainers are just
plain inactive, and, while packages can eventually be forcibly orphaned,
you have to go through the same political shit for every damn one. (And
you *need* the political shit to a large extent, otherwise there
wouldn't be much point in having package maintainers at all.)
> Claims that this is impossible don't hold much weight with me, when people
> all around us *are doing it*. It is clearly, and observably, possible. Say
> that it's more work than you think is worthwhile, if that's what you mean,
> but don't claim it's impossible.
There is a point where infeasible and impossible become very nearly the
same thing. I'm not sure anybody fully appreciated how much work the FHS
transition was going to be at the beginning, and just how many packages
would ignore the transition for extended periods of time.
I wouldn't have put what aj said in quite the same way, but it is true
that a lot of people commenting on this don't appreciate the sheer scale
of Debian because they haven't worked on one of these transitions. Those
who do generally dismiss the problem by saying "well, they should remove
some bloat then" - I invite those people to join the QA group in
actually locating bloat that's appropriate to remove, although often
that turns out to be not an awful lot less work (in the short term) than
fixing the packages themselves.
Colin Watson [firstname.lastname@example.org]