> So to answer your question (which is a good one), the reason tracking
> -current isn't a problem is that the corss-compiling will use make build,
> which will install new headers into the DESTDIR, and then proceed to use
> them exclusivly for cross-make programs.

Ok, this leverages the existing practice quite cleanly.

I still have a minor beef with the basic 'make includes' model though; it
overwrites headers before installing the binaries that those headers are
supposed to document. This means that if you have to abort a build of a
module halfway through, you are stuck using the new headers if you try to
build anything else that uses them.

I was thinking to fix two bugs with one swat, but it occurs to me now that
"xinclude" doesn't really fix this, and the only final solution would be to
include headers directly from their BSDSRCDIR master copies (as the kernel
does, I believe) -- however we still need the ability to build new source
using old headers, and supporting both models is messy.

Probably this is considered to be a long-standing "acceptable risk" of the
BSD build philosophy, so I'll stop muttering about it.

Todd Whitesel
toddpw @ best.com