llvm/clang once more

[Alex Hornung]

I know there is a rewrite of the driver underway, but nonetheless it was
just about no effort to add a dfbsd target and it allowed me to
experiment with different toolchain settings.
I also have the feeling that we are/we will be running into a lot of
similar problems, so a joint effort is of interest. Beware that we still
haven't even started on the userland. Actually I just tried it this
morning, and ld has some problem with errno and TLB:
/usr/libexec/binutils217/elf/ld: errno: TLS definition
in /usr/lib/libc.a(errlst.o) section .tbss mismatches non-TLS reference
in cat.o
/usr/lib/libc.a: could not read symbols: Bad value

Did you happen to run across a similar problem? How do you guys compile
userland with clang?

Sincerely,
Alex



On Tue, 2009-03-17 at 09:19 +0100, Roman Divacky wrote:
> On Mon, Mar 16, 2009 at 10:56:57PM +0000, Alex Hornung wrote:
> > Hi,
> > 
> > I've managed to get our GENERIC kernel to compile with llvm/clang. There
> > is still one temporary fix (at least until clang guys fix the related
> > bug), but the kernel definitely compiles, boots and works stable.
> > As a stress test I've done a buildworld, which it survived without
> > problems.
> > I've put everything about DragonFly and llvm/clang so far together on
> > one page, http://leaf.dragonflybsd.org/~alexh/clang/clang.html in case
> > you are interested in details or even want to try it yourself.
> 
> there's not much point in patching the python ccc as a new driver in C++
> is being finished these days... and the python one will be thrown anyway
> 
> anyway, great to see this progress! maybe freebsd and dragonfly can join
> forces in this effort? our code is still a lot similar and problems are
> usually generic enough.
> 
> roman