Date: Thu, 24 Oct 2013 16:15:05 -0700
On 10/24/13, John Regehr <regehr_at_[hidden]> wrote:
> We hacked Clang to return a random number for any undefined integer
> operation. Stuff breaks. Most of SPEC CINT breaks.
I had an earlier proposal to define the signed integer representation to be
two's complement. It died because (1) I wasn't proposing to make overflow
defined, which confused people, and (2) I wasn't aware of the two's
complement definition of the intN_t typedefs, which then made the
proposal less pressing.
> There's no way Unisys's compiler implements non-two's-complement if they
> run any substantial amount of C/C++ from the wild.
Those machines have historically been programmed in Algol. I really have
no idea how much C code they run.
> We hacked Clang to return a random number for any undefined integer
> operation. Stuff breaks. Most of SPEC CINT breaks.
I had an earlier proposal to define the signed integer representation to be
two's complement. It died because (1) I wasn't proposing to make overflow
defined, which confused people, and (2) I wasn't aware of the two's
complement definition of the intN_t typedefs, which then made the
proposal less pressing.
> There's no way Unisys's compiler implements non-two's-complement if they
> run any substantial amount of C/C++ from the wild.
Those machines have historically been programmed in Algol. I really have
no idea how much C code they run.
-- Lawrence Crowl
Received on 2013-10-25 01:15:07