Doing it better 1999-08-02 g
###

From Reuben.

> But let other people use different drivers if they want, because that is
> the whole point. Right?

What I was thinking is that our thin hardware abstraction layer can easily
be implemented on lots of hardware, but that's a distraction for us;
however, we should implement it on hardware that lots of people have.

> Another point: people will be interested in a cut down Linux distribution
> as a base for other OSes. Especially if we do the work necessary to
> provide the same abstraction layer for a Java style piggy-back OS on top
> of Windows, Mac OS and a full Linux. I think the ability to run our stuff
> stand-alone or piggy-back is really nice, especially for development, and
> we should publish that part as a separate product.

This sort of thing has already been done, e.g. OSKit runs Kaffe, there's a
cut-down Linux (~10Mb) which runs gforth and a few things to make
networking plausible; there're also lots of "Linux on a floppy" distribs
plus things like PicoBSD (the core of FreeBSD, which uses ficl in the
bootloader). Although I'm not sure it's been done well yet, it should have
been by the time we're anywhere near, so we should be taking advantage of
it rather than rolling our own.

One of the nice things about Tau is that to get anywhere we absolutely
have to keep it clean and simple. Complexity to gain performance simply
isn't an issue (with the exception of a few areas we're interested in,
like code generation).

--
http://sc3d.org/rrt/ | certain, a.  insufficiently analysed