simplify const params diagnostic on stable

[OliverMD]

Resolves #95150

[rust-highfive]

r? @wesleywiser

(rust-highfive has picked a reviewer for you, use r? to override)

[OliverMD]

@lcnr You added the E-Mentor tag on #95150, is this sort of what you had in mind? Would you like to review this?

[lcnr]

Thanks 👍

While your approach works, I think that this code already was a bit more complex than necessary and should get cleaned up a bit while we're already here.

[lcnr]

we already have a check for the adt_const_params feature a few lines above. I would prefer to only have 1 branch for that. So pretty much

if tcx.features().adt_const_params {
   // complex check¹
} else {
   // simple check, completely ignoring structural equality
}

¹ It might be interesting to check whether function pointers are considered a structural match violation. If they are, you can completely avoid checking the ty in that branch.

[OliverMD]

Ah gotcha, thanks! Unfortunately it looks like function pointers (and raw pointers) aren't structural match violations, but just to check, my (simplistic) reasoning is based on removing the ty check in the complex check case and then observing that the following test case successfully compiled, which I believe it shouldn't:

// revisions: full min
#![cfg_attr(full, feature(adt_const_params))]
#![cfg_attr(full, allow(incomplete_features))]

struct Test();

fn pass() {
    println!("Hello, world!");
}

impl Test {
    pub fn call_me(&self) {
        self.test::<pass>();
    }

    fn test<const FN: fn()>(&self) {
        //~^ ERROR: using function pointers as const generic parameters is forbidden
        FN();
    }
}

fn main() {
    let x = Test();
    x.call_me()
}

[lcnr]

r? @lcnr

[lcnr]

thanks

@bors r+ rollup

[bors]

📌 Commit 3b4589a has been approved by lcnr