> > Yup. -O0 lets it compile. So we'll need something like this?
> 
> No. The right thing to do is to fix the bug. We should isolate it and
> kill it. It could hurt lots of other programs.

Okay, but if that is not going to happen for many days, why do we have to
leave the tree broken? Anyone who decides to sync up with -current could
get fried, we'd see redundant bug reports on the mailing lists, and the
guy who tried to sync up with -current will most likely have to choose
between trying again later, or manually discovering and bringing in the
set of known workarounds and then maintaining them as local changes.

As far as I am concerned, a conditional -O0 in a makefile is the way to go,
but only for -current; such things should not be tolerated in a release
branch. It's obviously delaying the inevitable, but it is easy to spot and
is a common class of workaround, so you would get meaningful results from:

	grep -w -- -O0 `find /usr/src -type f -name 'Makefile*'`

... any time someone wants to go bogon hunting.

In my mind -current is the correct place for such hacks to be checked in.
Do you want more people testing it? KEEP IT BUILDING! Limping is vastly
better than broken.

I would love it if there were a documented list of filters that every
release branch must pass. That grants any workaround detected by the
filters permission to be committed to -current, without any worries
that it will survive indefinitely by hiding in the tree.

(Yes, this might be related to what I've been going through with arm32 lately)

Todd Whitesel
toddpw @ best.com