[Box Backup-dev] NetBSD patches?

[Martin Ebourne]

On Wed, 2006-02-22 at 15:19 -0800, Per Thomsen wrote:
> Missing items:
> 
>     * disabling of HAVE_KQUEUE on NetBSD 2.0.2? I saw no difference in
>       the test results on 3.0 with or without it. Or should it be for
>       all NetBSDs?
>     * The EV_SET cast issues. I'll test doing it in a different way, to
>       avoid having boat-loads of ifdefs, for every EV_SET call. Maybe a
>       macro of some kind. I know you were concerned about 64-bit
>       safeness as well, with the current patch casting to an int.

The disabling of KQUEUE is the same thing as the EV_SET problems. If you
remove HAVE_KQUEUE then the EV_SET stuff won't be compiled.

Casting a pointer to an int is completely non-portable, and just not an
option for us. Ben's suggestion of indexes into an array is the most
obvious workaround, but too much for now. If we get the disable KQUEUE
in then that is a perfectly acceptable solution.

> I'm unsure of how to put the HAVE_KQUEUE stuff into configure.ac.  Any
> help with that would be appreciated.

I sent an email yesterday (well, the day before technically) with an
example autoconf call to do generic compile tests, and even included a
possible few lines of code to put inside it to finish the kqueue test
off. A cut&paste with a search&replace should do the trick. If you're
still having trouble I can try and code up a patch, but I can't test it
obviously.

> I sent all the other patches I made to the dev-list a few days ago, and
> Martin said he'd commit the statvfs stuff.

Yes, I didn't want to commit that yet because I didn't think netbsd
support was going to make it, so didn't see the point in taking that
risk. I'll do that now, after all, we can always branch from the rev.
back if needed.

Cheers,

Martin.