rustdoc: Implementors of a trait is not const-evaled, thus showing `[T; 32 - 1 - 1 - 1 - 1 - 1 - 1 - 1]`

[kennytm]

Repro steps:

1. Go to https://doc.rust-lang.org/1.22.1/std/default/trait.Default.html or https://doc.rust-lang.org/nightly/std/default/trait.Default.html
2. Scroll down.
3. Notice the amusement.

The array lengths are normal in 1.21.0, but it is not clear whether this is the effect of performing const evaluation or the macro generating these impl changed in the new version.

[KamilaBorowska]

The array lengths are normal in 1.21.0, but it is not clear whether this is the effect of performing const evaluation or the macro generating these impl changed in the new version.

The macro in question is the same for 2 years, as seen with blame. Here is a commit that introduced it: 975a8ed.

[kennytm]

Thanks for checking @xfix! Considering it's not due to change of generating macro, untagging T-doc and tagging regression-from-stable-to-stable.

[hanna-kruppe]

Likely because array types now contain a constant expression instead of a fully evaluated usize. cc @eddyb

[hanna-kruppe]

Offending rustdoc code:

https://github.com/rust-lang/rust/blob/master/src/librustdoc/clean/mod.rs#L2032

[eddyb]

We attempt to const-evaluate when we get the type from HIR:

rust/src/librustdoc/clean/mod.rs

Line 2029 in 75a02a9

let n = cx.tcx.const_eval(param_env.and((def_id, substs))).unwrap();

But not from Ty:

rust/src/librustdoc/clean/mod.rs

Lines 2155 to 2156 in 75a02a9

	ty::TyArray(ty, n) => {
	let n = if let ConstVal::Integral(ConstInt::Usize(n)) = n.val {

So a few lines from the former should be copied to the latter.

[GuillaumeGomez]

Is anyone picking it up or should I wait a bit more before fixing it?

[detrumi]

This doesn't look too difficult, so I'll have a go at it.