[ cc'd to gnats-bugs@netbsd.org so it'll go on the PR. ]

Lennart Augustsson <lennart@augustsson.net> writes:
> "Chris G. Demetriou" wrote:
> > The whole point of the 'make includes' changes to the kernel was so
> > that kernel-only headers wouldn't have to be exported to user-land.
> 
> Indeed, but it seems that plenty of kernel-only headers are exported anyway.

Yes, however, most of those were exported because they were exported
before the 'make includes' changes.

Previously, little thought was put into exporting things to user-land,
and therefore it now takes careful thought before they can be
un-exported.  When making the 'make includes' changes, I couldn't put
in that 'careful thought' for every port, and so I didn't attempt to
do it for any.  For the MD code, it's something that each individual
port maintainer can do on their own if they have the desire.  (There
are probably plenty of MI headers which can be un-exported, too...)

The point of my response to this PR is that going forward, we should
be putting careful thought into what's exported _before_ it's
exported, going forward, so we can improve the situation.  Exporting
loadfile_machdep.h is not the correct solution the problem presented,
and exporting it only worsens this other problem.


> Someone should take a look at this and trim it down.

Yes, but as noted, it has to be done carefully.



cgd
-- 
Chris Demetriou - cgd@netbsd.org - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.