Re: A few proposals, this time from the C++ standards committee

[Linus Torvalds <torvalds@linux-foundation.org>]

On Wed, 5 Jun 2024 at 11:08, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> I would suggest that people look at improving 'restrict' and making it
> more useful, and just admit that the type-based thing was a mistake.

Note that I did see the other proposal on 'restrict', but I think that
one was a pretty small improvement.

I think people should work on making it work better in general. Real
compilers already effectively do that thin in much more interesting
ways, as part of finding the origin of a pointer.

For example, both clang and gcc have a notion of "alloc-like" functions:

   __attribute__((__malloc__))

which is a function attribute that basically says "the returned
pointer is a 'restricted' pointer". Except it is much better than the
'restrict' keyword, in that it actually works on real loads.

So I think the real answer to type-based aliasing is to throw the
garbage out, and instead help extend on existing notions of
"provenance of where the pointer came from".

Because compilers already do a *lot* of that kind of alias analysis,
and I think the proper approach is to strive to help compilers do
better on something reliable, instead of working around the fact that
some rodent-like creature got dropped on its head a few too many
times, and came up with the notion of type-based aliasing.

                     Linus