On 6 Dec 1999, Chris G. Demetriou wrote:

> * only recently have people started making cross-compilation really
> work.  of course, to be done sanely, this requires that there be some
> cross-tools sources which compile in a sane way.  There aren't.

I've been thinking about this a little more since you mentioned it.  I
think it's a great direction and remains highly relevant, but I don't find
present cross- limitations too discouraging either.  A few observations:

 o We shouldn't be too hard on ourselves.  NetBSD can produce cross tools
   that work, and we probably will.  For now, we are self-hosting on the
   EB's, and the build system is clean enough that you can get work done
   even on a slow board.  Wind River doesn't, cannot, and never will have
   a self-hosted toolchain.

 o The point of pkgsrc is to save work.  If it doesn't save work, don't 
   use it.  Putting cross compilers in pkgsrc is a nice idea from a 
   user's standpoint, but we need to do things that are maintainable with 
   minimal effort or else they will spend most of their time not working.
   pkgsrc also isn't good at dealing by hand with the chicken-and-egg
   problems that tracking -current and toolchains in general entail.  it's
   not a happy place to work on anything--it's more of a presentation system.
   NetBSD is clean enough on a low level that Presentation Wrapping isn't 
   as important as with vendor stuff.  we can concentrate on simply 
   minimizing time, work, and rework.  ``make install'' saves time for 
   emacs.  for cross tools, it just leaves me feeling confused.  ``i 
   installed what, where, and built it how, to cross from what to what, 
   and how do i make world?''

 o considering the audience of cross tools, and the fact that the build 
   system will need to be aware of cross anyway, maybe it's better to do 
   something a little more complex outside pkgsrc and document it.
   chances are the persron who uses the tools will learn something
   generally useful from working through such documentation anyway.

 o building NetBSD on Solaris would be really professionalish, but I 
   think the wrs stuff would sell okay if you had to run their host IDE on 
   vxWorks/i386--as long as it worked just as well.  most places will 
   afford a dedicated box for the ``host,'' and NetBSD installs easily 
   and runs on practically anything.  so this limitation is not so bad
   either.  i think it actually makes a somewhat good impression.  it's 
   even useful in the sense that the host is source-compatible with the 
   target, which means you can buy one EB and have ten other people 
   developing and testing the app parts of your software on cheap PC junk.
   what's that nasty embedded OS that emulates Win32?  same idea, only
   without sucking.  it's sensical enough from both a maintenance and
   usefulness standpoint that you could put some marketing spin on it and
   call it a Position rather than a Limitation.

In any case I've been tossing my opinions about way too much for one day,
so i'm going to take a rest for a bit.  appologies if you've already been
over this territory a dozen times.

-- 
Miles Nordin / v:1-888-857-2723 fax:+1 530 579-8680
555 Bryant Street PMB 182 / Palo Alto, CA 94301-1700 / US