[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Clang



Hi Steven,

2013/6/23 Steven Chamberlain <steven@pyro.eu.org>:
> so it may be more advantageous to instead try to make
> 9.1 available through wheezy-backports.

You mean the kernel only? In my experience this is not as easy as it
looks. A new kernel often drags in the rest of FreeBSD userland due to
ABI changes.

Then you get a new set of kernel headers, but you can't use those
since it would cause build breakage over the whole archive.

As a consequence, you might be unable to backport freebsd-libs, etc.
So perhaps you need to live with some of the kernel<->user ABI
disparity (things like #658617 are a good example of what I mean). Or
try to revert kernel subsystems to their older versions.

Then there may be some packages elsewhere also affected. GRUB and
Glibc typically are. Last time I recall sysvinit was too (because of
the cons25->term transition).

If you want 9.1 in wheezy-backports, you can always try of course. But
I wouldn't make plans based on the assumption that it will be
possible.

> If you use GCC for this, you must probably use gcc-4.8 as other versions
> will go away soon.
>
> If you want to use Clang, a couple of extra CFLAGS may be needed:
> http://bugs.debian.org/706490

TBH, I'm not sure what's best. I know a fraction of upstream
developers have tested 9.x with Clang during the whole release cycle.
However, despite that an overwhelming majority of FreeBSD developers
wanted this transition to happen, they weren't confident enough
themselves to use it as default for their release yet.

OTOH, GCC diverges more and more from the old version they were using.
Sounds like a potential source of trouble, and we've had serious
problems in the past because of this.

I guess we have to pick our poison then? :-)

--
Robert Millan


Reply to: