> >Except, that i don't think that the strategy of:
> >	"make (at least parts of) it build with the native compiler,
> >	then switch to gcc when we've shown that something useful can
> >	happen"
> >is at all uncommon when doing ports...
> 
> Yes, but how much would we lose by just starting with gcc?  On a "modern"
> machine it appears to take 6-8 hours, maximum, to build gcc, including
> c++, objective C, and Fortran.  Now, older machines may come up, but even
> then, it's possible to build.

To be quite frank, if you're porting to new hardware, you can lose
almost infinitely to gcc.

I had to fix or hack around at least 2 "serious" gcc bugs when i got
started making gcc compile NetBSD/Alpha.

If you're running new hardware, there's no guarantee that gcc will
work reasonably well at all, any time near when you're starting the
port.

There's also no guarantee that (as in my case, on the alpha) the
compiler is less than a little bit quirky.  Also, just because a
vendor ships a compiler, doesn't mean it's an ansi compiler...


cgd