Re: [Gnu-arch-users] On configs and huge source trees

[Ollivier Robert]

According to Ludovic Courtès:
> As someone mentioned, GCC could certainly be split into a number of
> projects.  This could have a positive impact on code quality -- the APIs
> would have to be well-thought, things would have to be "librarified" as
> Tom would say, which means that various components could be untangled
> and made independent and reusable.  The same goes for Linux: why do all
> these drivers, filesystems, etc. _have to_ be within the kernel tree?

I have in mind a specific example: /usr/ports in FreeBSD.  It is a CVS
repository with more than 124000 files in it in more than 32000
directories.  We could split it several different ways: at the port level,
at the category level and so on but we'd lose the atomic commit feature and
changeset generation, that is unacceptable IMHO.

> Of course, one may argue that it's not the job of the revision control
> tool to advocate a way to organize projects.

Arch is doing it by forcing people to think in terms of c--b--v.  This is
one of its main selling point!

> Arch' configs help keep track of dependencies, more than working around
> a performance problem.

It is more showing a set of categories as a single tree (or set of trees)
and I think we lose too much in the process.  It is pretty close to CVS
modules (although c--b--v is a module in itself too).
-- 
Ollivier ROBERT -=- FreeBSD: The Power to Serve! -=- address@hidden
Darwin snuadh.freenix.org Kernel Version 7.9.0: Wed Mar 30 20:11:17 PST 2005