On Tue, 30 Jun 1998, David Forbes wrote:

> >      Hmm... needing one box of each type would be a pain thats true... what 
> >      about cross-compiling with gcc?  never looked into it so I dont really 
> >      know squat about it...
> 
> 	That could be quite messy, esp. from the point of view of port
> specific kernel options (like CPU type) at the configuration side, plus
> disc space overheads of having several cross-compilers/linkers.

I don't think it'd be too bad. My m68k-netbsd setup (on an i386) takes 14
MB. My i386 kernel takes 5.6 MB. So the cross-compiler takes up about the
space of 3 kernels. If you really had the space to let folks make random
kernels on your computer, then I don't think the cross-compilers would
take up much space. :-)

And kernels are usually the easiest things to cross-compile. Most of our
cross-compilation problems have to do with shared libraries, which aren't
used in compiling kernels.

> >      maybe keep a cache of kernels on line so that if you specify options 
> >      that someone else has already asked for before, it will just dig it up 
> >      from the cache instead of re-compiling it from scratch.  this would be 
> >      faster and reduce the number of compiles executed.
> 
> 	I think it would be quite tricky to do a direct comparision of two
> compiles, and keeping loads on disk would also take a lot of space, but
> keeping some from say, the last seven days along with their configuration
> details would be handy.  Also, regular builds of extremely common
> ("standard") configurations from -current? 

If it were my computer, there'd be a limited default list of kernels. If
you had a custom one, you'd have to chat about getting an account. You'd
log into a menu program which'd let you edit a config, config a kernel,
and compile. :-)

Take care,

Bill