kldxref: Be more conservative about what we reject.

[imp]

Summary

kldxref anything that's a .ko. This will skip the kernel, but we don't
need to kldxref the kernel itself.

Sponsored by: Netflix

[jrtc27]

This will skip the kernel, but we don't need to kldxref the kernel itself.

Are we sure about that? Is there any harm in doing dot != NULL && ... instead and preserving the historic behaviour of processing the kernel?

[jrtc27]

Also the manpage needs updating to remove the documentation regarding two dots

[jhb]

Hmm, we certainly have processed the kernel to date. I think it's quite possible we don't need to do the kernel though as we will never try to load a kernel as a dependency.

[jrtc27]

Perhaps do it as two changes, and only MFC the fix, not dropping the kernel?

[imp]

I'll update the man pages..

But we've not been doing the kernel for months now and there's been no harm from that (the harm is from foo.bar.ko being ignored).

[jrtc27]

In D43507#991894, @imp wrote:

I'll update the man pages..

But we've not been doing the kernel for months now and there's been no harm from that (the harm is from foo.bar.ko being ignored).

But we do still do the kernel?

[imp]

In D43507#991895, @jrtc27 wrote:

In D43507#991894, @imp wrote:

I'll update the man pages..

But we've not been doing the kernel for months now and there's been no harm from that (the harm is from foo.bar.ko being ignored).

But we do still do the kernel?

Ah, yes, we are. I thought we'd only done one . but we do 0 or 1 dot.

So I can add kernel to the allowed list, but it should be only kernel. Though this is at best a heuristic and wouldn't work if you changed the default name of 'kernel'...

But I just updated with a fix to not skip files w/o a dot in them.

[imp]

commit message also tweaked, but git arc doesn't update that automatically.