Subject: Re: [ub] signed left shifts again
From: Lawrence Crowl (Lawrence_at_[hidden])
Date: 2013-11-13 16:10:32
On 11/13/13, John Regehr <regehr_at_[hidden]> wrote:
> Ok, I've spent some time looking at open source crypto libraries and
> wrote up some results here:
> A good number of these problems will go away once shifting into the sign
> bit is no longer UB, and most of the rest go away once the maintainers
> of 5 packages fix their rotate functions.
It seems that this code also relies on a two's-complement representation.
We already have examples of rare machines that have different behavior.
I have not seen the examples, but is it not the case that casting to unsigned
before the shifts would have created defined behavior?
-- Lawrence Crowl
SG12 list run by firstname.lastname@example.org