code review 4527098: gc: enable building under clang/2.9

gc: enable building under clang/2.9

To build under clang, pass the path to clang in CC when
calling ./make.bash

CC=/opt/llvm/llvm-2.9/bin/clang ./make.bash

Credit goes to jmhodges for suggestions.

[jmhodges, 2011-06-06 07:16:40]

http://codereview.appspot.com/4527098/diff/4/src/cmd/gc/subr.c
File src/cmd/gc/subr.c (left):

http://codereview.appspot.com/4527098/diff/4/src/cmd/gc/subr.c#oldcode108
src/cmd/gc/subr.c:108: *(int*)0 = 0;
Alternatively, this could be done with `*(volatile int*)0 = 0;` if a gcc builtin
isn't acceptable.

[dave_cheney.net, 2011-06-06 07:20:25]

Hello jeff@somethingsimilar.com (cc: golang-dev@googlegroups.com),

I'd like you to review this change to
https://go.googlecode.com/hg/

[r, 2011-06-06 07:55:51]

http://codereview.appspot.com/4527098/diff/3001/src/cmd/gc/subr.c
File src/cmd/gc/subr.c (right):

http://codereview.appspot.com/4527098/diff/3001/src/cmd/gc/subr.c#newcode108
src/cmd/gc/subr.c:108: __builtin_trap();
is there a reason not to call abort()?

[ality, 2011-06-06 08:03:19]

dave@cheney.net once said:
> -		*(int*)0 = 0;
> +		__builtin_trap();

I'm suprised clang won't let you compile this.

Can you do this another way?  The only other use of
__builtin goo is in 8g/gsubr.c:/^regalloc and I've
been meaning to replace it with getcallerpc (once I
import the code from plan9port's lib9).

I know there are a few others, aside from myself, who
would like to compile the Go toolchain on Plan 9.

Thanks,
  Anthony

[jmhodges, 2011-06-06 08:18:04]

On 2011/06/06 07:55:51, r wrote:
> http://codereview.appspot.com/4527098/diff/3001/src/cmd/gc/subr.c
> File src/cmd/gc/subr.c (right):
> 
> http://codereview.appspot.com/4527098/diff/3001/src/cmd/gc/subr.c#newcode108
> src/cmd/gc/subr.c:108: __builtin_trap();
> is there a reason not to call abort()?

Ah, my current read of this would say that it's okay. 

This may be a sleep-deprived mind but I think we found something interesting
here. My read is that the SIGSEGV that was meant to be raised here would be
caught by the signal(SIGSEGV, fault) call in lex.c's main(). But if that is
true, we eventually wind up back in hcrash() like so: lex.c's fault(int) -> to
subr.c's fatal(char *...) -> subr.c's hcrash(). So, by fixing this attempted
SIGSEGV, we could be introducing a stack blow up.

In any case, abort() does seem okay (and probably preferable).

[dave_cheney.net, 2011-06-06 08:53:32]

Thanks for the feedback. PTAL.

[jmhodges, 2011-06-06 08:57:58]

LGTM

[r, 2011-06-06 09:51:08]

LGTM

[r, 2011-06-06 09:53:48]

*** Submitted as http://code.google.com/p/go/source/detail?r=6414e4f6b736 ***

gc: enable building under clang/2.9

To build under clang, pass the path to clang in CC when
calling ./make.bash

CC=/opt/llvm/llvm-2.9/bin/clang ./make.bash

Credit goes to jmhodges for suggestions.

R=jeff, r, ality
CC=golang-dev
http://codereview.appspot.com/4527098

Committer: Rob Pike <r@golang.org>

[dave_cheney.net, 2011-06-06 09:54:20]

Thanks Rob.

[rsc, 2011-06-07 19:19:55]

I'm sad to see abort instead of *(int*)0 = 0.
The former deposits you in the bowels of glibc
while the latter leaves you in the actual program
you want to debug.

Does clang let you do *(volatile int*)0 = 0 ?

Russ

[jmhodges, 2011-06-07 22:04:25]

Totally does, as my first comment pointed out!

On Tue, Jun 7, 2011 at 12:19 PM, Russ Cox <rsc@golang.org> wrote:
> I'm sad to see abort instead of *(int*)0 = 0.
> The former deposits you in the bowels of glibc
> while the latter leaves you in the actual program
> you want to debug.
>
> Does clang let you do *(volatile int*)0 = 0 ?
>
> Russ
>

[dave_cheney.net, 2011-06-08 00:28:57]

Hi Russ,

Please try http://codereview.appspot.com/4517143

Cheers

Dave

On Wed, Jun 8, 2011 at 5:19 AM, Russ Cox <rsc@golang.org> wrote:
> I'm sad to see abort instead of *(int*)0 = 0.
> The former deposits you in the bowels of glibc
> while the latter leaves you in the actual program
> you want to debug.
>
> Does clang let you do *(volatile int*)0 = 0 ?
>
> Russ
>