GCC 3.3.2 kernel

[Craig Dooley]

Some of the stuff has been taken from FreeBSD.  Im currently trying to work my 
way through makefiles to find out why things would be different on gcc 2.95 
and gcc 3.3.2.  Could we take binutils and gcc from their tree instead of a 
GNU import when the time comes?  I dont know if they have changed the abi at 
all, but if it's already worked out from them, it's probably a much better 
place to start.  I was working on buildworld too, and I know amd has a couple 
bugs.  I've been thinking of a new package manager and would like to get a 
working version of that in perl with static programs in /usr/lib/dfpkg just 
as a test platform, but moving the base system to build packages instead of 
installworld would be a nice test.

-Craig

On Wednesday 29 October 2003 07:50, David Rhodus wrote:
> On Tuesday, October 28, 2003, at 11:32 PM, Craig Dooley wrote:
> > Heres a patch that gets the kernel closer to being built with gcc
> > 3.3.2 from
> > ports.  The kernel compiles and runs with gcc 2.95 also as of now, but
> > there
> > are changes to drivers I cannot test (if_ray and if_wx that i can
> > think of)
> > but I think all the changes deal with either string tokenization or
> > multiline
> > strings, so I dont think there should be any problems.  The kernel
> > still
> > cannot be built with gcc 3.3.2 though.  When using buildkernel, it
> > gets to
> > linking kernel.debug and dies with MANY undefined references to puts
> > and
> > putchar.  When using config; make depend; make it dies in make depend.
> >  The
> > first pass in aic7xxx/aicasm goes fine, but the second pass dies with
> > missing
> > headers and sure enough the headers it wants are not in
> > compile/TRAGEDY/usr/
> > src/sys/...  I'm looking into those.
>
> Thanks Craig. I'm wondering are most of these changes in FreeBSD ? I'm
> asking from the perspective to see if they have already done most of
> the initial testing for us. Though this is something well definitely
> need to
> get working as preparation for the amd64 port. Last time I check our
> user
> land build was dyeing in openssl, which is a vendor branch we need to
> do a import of to get the fresh bits. Though I wonder if it was
> crapping out
> because of some deceleration from our include files.
>
> -DR

-- 
Craig Dooley											cd5697 at xxxxxxxxxx


Attachment:
pgp00007.pgp
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgp00007.pgp
Type: application/octet-stream
Size: 187 bytes
Desc: "Description: signature"
URL: <http://lists.dragonflybsd.org/pipermail/submit/attachments/20031029/fffd090c/attachment-0014.obj>