[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Axiom-developer] Re: usual development
From: |
Mark Murray |
Subject: |
[Axiom-developer] Re: usual development |
Date: |
Sat, 19 Mar 2005 20:01:13 +0000 |
root writes:
> in CVS you mention a flow of:
>
> local change
> commit to HEAD
> merge Good(tm) changes to STABLE branch
> release STABLE branch every 3 months or so
>
> in Arch the HEAD==branch. rather than everyone working on the
> same set of sources the branches allow you to work on BSD related
> changes where Bill can work on Windows related changes. These
> get sorted at the merge-into-main step when the branch is considered
> working. Thus rather than have 10 developers working on the HEAD
> there are 1 or 2 developers working on individual branches. This
> makes it less likely that they will step on each other.
Right - but at some stage the issue of "Do my BSD changes break
Windows?" crops up, and in the "all commits go to HEAD model", it means
that thinking about this forms part of my work, it is not something
that needs to be re-engineered later, particularly if the other
developers and I are approaching the problem in horribly incompatible
ways.
It allows us to co-operate earlier, in other words.
> It also allows us to explore structural changes such as your
> suggestion of skipping the lsp subdir build, and complete rebuilds
> such as the SBCL ansi common lisp branch. These are not possible
> under a CVS model except as a new project.
Not really. The only project that I've seen which really tested CVS'
limits that way was FreeBSD's 3-year mission to redesign the SMP
fundamentals, and that was an _ENORMOUS_ piece of work that tore the
kernel apart.
We've reorganised the tree, we've upgraded the C compiler, we've added
new CPU architectures, we've sliced of big chinks of old code, and
we've played very nasty games with cryptographic regulations of the
incomprehensible kind. CVS worked, not always very well, but the
development model was roundly affirmed.
> Arch has its problems though. I haven't figured out how to kill off
> a whole branch. And the --no-pristine option does not seem to work.
Yeah :-(.
Arch is not very feature-rich. It may get there, who knows? :-)
I guess the most important part of what I'm on about, though, is how
the Axiom project is going to scale once (when!) it gets popular. Once
the submissions start pouring in, how are you going to cope?
I understand, (and fervently hope!) that it will be an extraordinarily
comprehensive mathematics package, and I'm suggesting that the Linus
Torvalds' Linux kernel development model won't scale very well. I'd LOVE
to be proven wrong :-).
M
--
Mark Murray
iumop ap!sdn w,I idlaH