[cfe-commits] [PATCH] PR14306: Move -fbounds-checking to-fsanitize=bounds

[Joey Gouly]

Hi Richard & Nuno,

> > What I'm not sure is if 'bounds' should be included in the ubsan group.
> > While it also checks for undefined behaviour, it won't produce any nice
> > diagnostics like the other ubsan checkers. This bound checker just
crashes
> > the program when it detects a violation.
>
> I agree that it would be preferable to produce a call into the ubsan
> runtime to produce a diagnostic for this, but I don't think that needs
> to block this patch. We should decide one way or the other, though,
> and not link in the ubsan runtime for -fsanitize=bounds if we're not
> going to use it.

I was planning on looking into using ubsan for diagnostics after this
initial patch.
I updated the patch to not link with the ubsan-rt if -fsanitize=bounds is
passed.
Ok to commit?

> Presumably there will be a followup patch to LLVM, to remove the
> vestigial 'Penalty' argument; we could at that time add an argument to
> the createBoundsCheckingPass function to specify how to handle a
> failure.

There is a patch to remove the penalty arg, it has been approved for
committing,
but I'm waiting for the Clang patch to land before I commit it. I have also
started to look into adding a new parameter to select between traps and
runtime calls.

Thanks,
Joey

> FWIW, I would prefer if 'bounds' was included in the ubsan group. I'm just
> raising the concern that not everyone may agree.
>
> Nuno
>
> ----- Original Message ----- From: "Joey Gouly" <joey.gouly at arm.com>
> To: <cfe-commits at cs.uiuc.edu>; <richard at metafoo.co.uk>
> Sent: Wednesday, November 21, 2012 11:52 AM
> Subject: [cfe-commits] [PATCH] PR14306: Move -fbounds-checking
> to-fsanitize=bounds
>
>
>
> Hi all,
>
> Attached is the patch to change the -fbounds-checking flag to
> -fsanitize=bounds, and also put it under the ubsan flag as well.
>
> Note: I removed the bounds checking penalty parameter, but that is in a
> separate patch.
>
> Please review!
>
> Thanks,
> Joey