dropck: do not add types with escaped-regions into the breadcrumbs set.

[pnkfelix]

dropck: do not add types with escaped-regions into the breadcrumbs set.

In particular, when we traverse a type like:

(for <'r> Fn(&'r u8), for <'r> Fn(&'r u8))

we should not add the first type Fn(&'r u8) to the bread-crumbs set,
because that type ends up being structurally equivalent to the
second Fn(&'r u8) type, even though they each should be considered
distinct since they occur under distinct for <'r> binders.

Fix #25750.

[rust-highfive]

r? @Aatch

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

[pnkfelix]

r? @nikomatsakis

[arielb1]

What's the issue here again (test?)? I think adding a skip_current_subtree before the continue; when we meet a breadcrumb would be a wiser choice (actually, it is required, see #27240) - if there is an issue with that, it would be nice to at least comment on it.

Anyway, I think ty_walk is a stupid choice here, given that we don't actually treat anything nominally here (except trivially for primitives, and semi-accidentally for tuples).

[pnkfelix]

@arielb1 you're right I should have included a regression test.

[pnkfelix]

@arielb1 also I agree that it would probably be good to add a skip_current_subtree before the continue; ... its possible that doing so would resolve #25750 as well, ... I will think about it.

[pnkfelix]

(closing, i suspect @arielb1 's approach of #27261 is going to be a more realistic PR to land.)