Re: Module paths [Re: CVS commit: src]

[Andrew Doran <ad%netbsd.org@localhost>]

On Sun, May 04, 2008 at 01:39:36PM +0000, Christos Zoulas wrote:
> In article <20080504123433.GC27619%hairylemon.org@localhost>,
> Andrew Doran  <ad%netbsd.org@localhost> wrote:
> > 
> >> i think we also need some version number in here.  kernel number
> >> changes matches when the kernel ABI has changed so, it seems
> >> a valid choice to make.  it means having a lot more to clean
> >> up if you use modules but i don't know of any other way that
> >> is actually safe.
> >
> >/foo/modules/5.0/cd9660/cd9660.kmod
> >
> >I'm tempted to say that the minor version shouldn't matter for releases but
> >it might be hard to stick to that kind of policy.
> >
> >/foo/modules/4.99.62/cd9660/cd9660.kmod
> >
> >For -current the entire version would need to be significant. One problem I
> >see is getting those into the set lists, but that's a SMOP a think.
> >
> >Sound reasonable?
> 
> Is that good enough? What if I am using 2 kernels, one with DIAGNOSTIC,
> DEBUG, LOCKDEBUG, and one without?

The ABI is insensitive to DIAGNOSTIC and LOCKDEBUG these days. Actually,
there is a problem with KASSERT(mutex_owned(...)) on a spin mutex but that
is a bug and I had forgotten about it. I'll file a PR.

I don't know of problem areas with DEBUG but it's possible that some exist.

Andrew