Re: 32-bit archs, want64=true, and gas integers

[Jan Beulich <jbeulich@suse.com>]

On 06.05.2022 11:00, Pedro Alves wrote:
> Looking at the "Numbers" chapter in the manual:
> 
>  https://sourceware.org/binutils/docs/as/Numbers.html
> 
> we have:
> 
>  "Integers are numbers that would fit into an int in the C language. Bignums are integers, but they are stored in more than 32 bits."
> 
> I note it says "more than 32 bits", not "32 bits or more".
> 
> By my reading of the "Integers" and "Bignums" subsections, these integers are always signed.  Maybe
> that should be explicitly said in the manual, instead of introducing signed vs unsigned numbers?
> 
> And then, couldn't we make gas use int32_t for integers, and int64_t for Bignums (and clarify in the manual
> that "more than 32 bits" is exactly "64 bits"?  IOW, use int64_t instead of bfd_vma in the
> expression evaluation stuff.  

Bignums are quite a bit wider than 64 bits, and on 64-bit architectures
using just int32_t for integers is definitely insufficient (there are
pretty limited operations one can do on bignums). I'm afraid the doc is
simply outdated in talking about "int in C language"; it's more like
"long", and even then only in Unix-like environments, so it would likely
be even better to talk about {,u}intptr_t.

Jan

> That way, gas would be following what it documents, and 32-bit or 64-bit bfd would make no difference,
> and we'd get rid of a subtle host / --enable-64-bit-bfd -dependent behavior change for 32-bit ports.
> 
> Pedro Alves
>