add AnonConst as potential parent_item

[lcnr]

This prevents some help messages in const contexts, due to

rust/src/librustc_typeck/check/demand.rs

Lines 597 to 601 in 413a129

	if self.tcx.hir().is_const_context(expr.hir_id) {
	// Shouldn't suggest `.into()` on `const`s.
	// FIXME(estebank): modify once we decide to suggest `as` casts
	return false;
	}

considering that the check in is_const_context was previously unreachable,
this is actually a bugfix 😆

rust/src/librustc_middle/hir/map/mod.rs

Line 563 in 413a129

| Node::AnonConst(_) => true,

r? @eddyb cc @estebank

[lcnr]

This help message was incorrect.

[lcnr]

https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=3e0d446869bd89792a113dbd8f140f15

[estebank]

It'd be nice if we changed check_for_cast to still emit a suggestion for literal suffix changes. Would you mind doing that in a separate PR or filing a ticket after this is merged?

[lcnr]

Can implement it in a separate PR 👍

[eddyb]

I don't think this is the right approach, get_parent_item should barely exist itself, what's the actual ICE being fixed here? What's calling get_parent_item on something inside AnonConst and hoping to see the body owner? Because that's what it should be looking for, not "parent item".

[lcnr]

This PR is needed for https://github.com/rust-lang/rust/blob/4b575996168edbc5b7ad46229f5ed3bfee35484c/src/test/ui/const-generics/type-dependent/const-arg-in-const-arg.rs #70 (#71154 uses it to get the correct typeck_tables)

get_parent_item is also used in is_const_context to get the body owner of an anon cost.

rust/src/librustc_middle/hir/map/mod.rs

Line 564 in bf45975

let parent_id = self.get_parent_item(hir_id);

which currently results in wrong diagnostics in

rust/src/librustc_typeck/check/demand.rs

Lines 597 to 601 in 413a129

	if self.tcx.hir().is_const_context(expr.hir_id) {
	// Shouldn't suggest `.into()` on `const`s.
	// FIXME(estebank): modify once we decide to suggest `as` casts
	return false;
	}

to get the body owner of an anon cost.

What do you want to use instead?

[eddyb]

You want to loop through the parents with get_parent_node and stop when maybe_body_owned_by returns Some. That is the enclosing body owner, which is what you want.

If you want to be more efficient you could probably add an enclosing_body_owner(_def_id) which is like body_owner(_def_id) but does the loop I described above, stopping at a node for which is_body_owner returns true.

cc @ecstatic-morse @oli-obk @wesleywiser on is_const_context, I'm suspicious of its correctness and uses (it probably wants to be unified with whatever MIR const-checking uses)

[lcnr]

I now use the following in #71154:

    pub fn enclosing_body_owner(&self, hir_id: HirId) -> HirId {
        for (parent, _) in self.parent_iter(hir_id) {
            if let Some(body) = self.maybe_body_owned_by(parent) {
                return self.body_owner(body);
            }
        }

        bug!("no `enclosing_body_owner` for hir_id `{}`", hir_id);
    }

This means that this PR now only fixes incorrect suggestions afaict. Don't fully understand when get_parent_node is appropriate yet, so I don't know how to proceed here.

[lcnr]

Closing this for now as this is not the correct approach.

Will try and make a more general fix for is_const_context in the near future.