On 27 Dec 1999, Perry E. Metzger wrote:

> Bill Studenmund <wrstuden@nas.nasa.gov> writes:
> > Why?
> > 
> > What does making cross compilers part of the base system give us which
> > having them(*) in the package system does not?
> 
> The package system is
> 
> 1) Designed for things that are not internally maintained by the
>    project

While this is true, there's nothing stopping us from making all of these
bits be in the package system. We already have packages which we
completely support, so this isn't much of a stretch.

> 2) Designed for things that are not core to project management and
>    maintenance. 

How so? To be involved in the project, you have to use ssh to connetct to
the machines. To update source, you have to use cvs. Both of these actions
strike me as being much more of a "core" activity to managing and
maintaining NetBSD than cross compiling. :-)

> The cross compiler system *is* internally maintained and *is* about to
> become critical. (Without it, I don't expect we're going to be able to
> get out future releases in a prompt manner.)

But how does this stop us from using the package system as the vehicle of
distribution? There's NOTHING stoping us from making (other than
asthetics) packages which reach over into BSDSRCDIR and grab source,
compile it, and install it.

It would be a NetBSD-only package. :-)

> > That strikes me as a lot of bloat. I mean we're the OS which doesn't want
> > to add perl to the base distribution, and here we're talking about adding
> > 10 cross compilers???
> 
> You don't have to compile all of them, any more than you are required
> to compile games.

Yes, but the proposal wasn't at all flexible in how it gets distributed.
There was talk of ONE distribution file, xcomp (name would need changing
to not conflict w/ the X compiler stuff). A la base.tgz, comp.tgz,
misc.tgz. There's no granularity in that. If you get this distribution,
you would be getting all of the cross compilers.

I agree distribution flexibility is good (only getting the bits you need).
I'm just seriously questioning why we would want to not use the package
system to do it.

> > Oops. Since object format counts in the matrix above, we have i386-aout,
> > i386-elf, sparc-aout, and sparc-elf above.
> 
> Unnecessary. All that is really needed is the ability to cross-build
> current releases.

That was not part of the original proposal. In fact, from personal
communication, Frank's main motivation was so that a current release which
migrated to ELF would be able to make an a.out ld.so which did the right
magic. So as per the original proposal, they are necessary. :-)