Ports cross-compilation

[Mark Linimon]

On Mon, Dec 05, 2011 at 08:54:55PM -0700, Warner Losh wrote:
> Well, the ONLY_FOR_ARCH stuff might be a good first-order clue.

This is why I want to move away from NOT_FOR_ARCHS and move exclusively
to ONLY_FOR_ARCHS; for embedded, the former is exactly The Wrong Thing.
Unfortunately, NOT_FOR_ARCHS is all over the tree.

The other problem is that if the port merely breaks on compiling
(rather than there being missing pieces that prevent it from even
getting that far) the 'best practice' is to set BROKEN conditionally,
instead of *ARCH*.  But the tree is rife with examples where the one
approach is done rather than the other.

To see an example of the union of the above:

http://pointyhat.freebsd.org/errorlogs/sparc64-9-latest/duds.verbose

That's the union of all cases of "do not try to build me" for sparc64-9.
Obviously a lot of the things are common to all archs, and one metadata
entry can cause many lines (e.g. "no linux on sparc64"), but it gives
a rough idea of the scope of the task (1794 entries).

mcl