Rustdoc does not substitute const parameters

[fmease]

I ran cargo doc on this code (src/lib.rs):

// #![feature(min_const_generics)] // stable by now

type Array<T, const N: usize> = [T; N];

pub type A0 = [i32; 3];           // printed correctly
pub type A1 = Array<i32, 3>;      // printed INCORRECTLY as `[i32; N]` (should print `[i32; 3]`)
pub struct B0(pub [i32; 3]);      // printed correctly
pub struct B1(pub Array<i32, 3>); // printed INCORRECTLY as `B1(pub [i32; N])` (should print `[i32; 3]`)

I expected to see the definitions type A1 = [i32; 3]; and pub struct B1(pub [i32; 3]); in the generated documentation.
Instead, I found type A1 = [i32; N]; and pub struct B1(pub [i32; N]);.

That is, the definitions wrongly contain a stray identifier N that should have been substituted with the literal 3 just like the T was successfully substituted with i32.

Meta

rustc --version --verbose:

rustc 1.52.0-nightly (51748a8fc 2021-03-05)
binary: rustc
commit-hash: 51748a8fc77283914d4135f31b5966a407208187
commit-date: 2021-03-05
host: x86_64-unknown-linux-gnu
release: 1.52.0-nightly
LLVM version: 12.0.0

@rustbot modify labels: +T-rustdoc +A-const-generics +F-min_const_generics +C-bug

[fmease]

I hope this is not a duplicate or an already known issue that is part of some bigger tracking issue or some agenda. I thoroughly searched beforehand but couldn't find anything related.

[jyn514]

This is a duplicate of #14072 I think.

[fmease]

Hmm, maybe it is. Hard to tell as that old issue is really vague and updates to it are scattered.

In my eyes, this simple case of const parameters (literals only) looks manageable to implement. Not so sure about the general case which requires const evaluation.

@ jyn514, do you think writing the code for the substitution of const parameters is a matter of "copying"/mirroring the one for the substitution of type parameters (which in some parts has to exist as seen by T -> i32)? Thinking about a potential first-time contribution.

[jyn514]

In my eyes, this simple case of const parameters (literals only) looks manageable to implement. Not so sure about the general case which requires const evaluation.

You can evaluate constants with print_evaluated_const:

rust/src/librustdoc/clean/mod.rs

Line 1957 in 27885a9

value: print_evaluated_const(cx, def_id),

@ jyn514, do you think writing the code for the substitution of const parameters is a matter of "copying"/mirroring the one for the substitution of type parameters (which in some parts has to exist as seen by T -> i32)? Thinking about a potential first-time contribution.

I'm not sure how this part of the code works, sorry. Happy to look into it if you get stuck though :)

[fmease]

Happy to look into it if you get stuck though :)

Thank you :) I appreciate that. I opened a Zulip stream to discuss the work on the implementation.

[clarfonthey]

Provided another example in a (now closed) duplicate issue:

pub trait Test {}
pub struct Array<T, const N: usize>([T; N]);

impl Test for [u8; 1 + 1] {}
impl Test for Array<u8, {1 + 1}> {}

[camelid]

I think the issue here is that DocContext.ct_substs is actually never read from. I'm looking into this a bit.

[fmease]

@camelid Yes, I've made the same observation and noted down some more findings in the Zulip stream linked above (although some information there might be outdated by now).