[WIP] Add lint for int to ptr cast

[mattico]

Fixes #42915

First commit adds the lint, the second commit fixes the lint warnings in the compiler on Windows. More sure to follow for other platforms.

[WIP] because we should wait for the decision about #43291 to determine what message we use. Specifically, if it's decided that those strange casts can't/won't be fixed, the message should warn that these casts don't sign-extend as expected unless you go through {u,i}size.

Edit: Also we should consider the interaction this lint can/should have with the portability lint RFC (unimplemented right now), and if it requires an RFC of its own like rust-lang/rfcs#1782. I hadn't considered that adding a lint would require an RFC.

If #43291 does get fixed, we may want this lint to only apply to {i,u}8 as those cases are much more likely to be bugs than false positives compared to the other integer types.

I should also probably add a NOTE: try {expr} as usize as *const

[rust-highfive]

r? @arielb1

(rust_highfive has picked a reviewer for you, use r? to override)

[zackmdavis]

if let Some(&CastKind::AddrPtrCast) = ctx.tables.cast_kinds.get(&source.id) {
?

[mattico]

certainly

[aidanhs]

Thanks for the PR @mattico! We'll check in now and again to make sure @arielb1 or another reviewer gets to this soon.

[carols10cents]

friendly ping @arielb1!

[arielb1]

This LGTM. Let's wait for the lang team meeting to see what we want for the message to be.

[arielb1]

The message is a placeholder, but I think "casting from a potentially non-pointer-sized integer to a pointer" would be better English.

[ollie27]

While you're at it, how about linting the other direction as well (rust-lang/rfcs#1782)?

[mattico]

@ollie27 I hadn't seen that RFC before, thanks for the link.

It seems like these lints could interact with the portability lint (rust-lang/rfcs#1868) to have a different warning when the integer size matches the pointer. e.g:

#[cfg(target_pointer_width = "32")]
fn bar(foo: *const u8) {
    let x = foo as u32; // No warning
}

fn baz(foo: *const u8) {
    let x = foo as u32; // warning: pointer cast is not portable ...
}

This is becoming more complicated than I expected...

[retep998]

I'd personally rather have a more broad lint now that doesn't take into account cfg portability stuff, and later it can be amended to take that stuff into account.

[alexcrichton]

According to #43339 (comment) I've tagged as T-lang and waiting on team

[mattico]

Note to self: rebase on #43522

[mattico]

Lang team decided (#43291 (comment)) to lint first, maybe fix later. Will update after #43522.

[RalfJung]

This looks like it's doing something very similar to #43641? (as noted in #43641 (review))

[retep998]

#43641 is specifically for casts from signed integer types smaller than isize to raw pointers, which is a future compatibility hazard if we ever change that cast to perform sign extension instead of zero extension. This PR meanwhile is for a more general lint on all casts from any integer types other than isize/usize to raw pointers.

[aidanhs]

#43522 has landed, looks like this didn't need a rebase. Just for clarity, @mattico can you explain where this PR is up to/what it's waiting on?

[mattico]

There's some overlap between this and #43641. I think I'm just going to let that PR take over. If we change our minds about what cases we want to cover, we can edit that lint later on.