`use super::*` importing private members from parent module

[sgrif]

Reproduction script: https://is.gd/qvbF0b

This code compiles on stable and beta but fails to compile on nightly. use super::* previously did not include private members. On nightly it is reported that "Foo is ambiguous"

[bluss]

It's an intentional change, but everything that makes stable code stop compile is classed as a regression initially.

[sgrif]

Figured as much. I believe technically this is a "minor breaking change" since code can be forwards compatibly changed to handle this by not using glob imports, right? Is it still a regression in that case? Either way a release where this warns instead of erroring would definitely be nice.

[jseyfried]

@sgrif right, there can only be a regression when these conditions are met.

We landed this change #37127 without a warning cycle because Crater found no breakage. It would be feasible to implement a warning cycle as detailed in #35120 (comment).

cc @nrc

[sgrif]

As an aside, I don't think that Crater finding no breakage should be used as justification for skipping warning cycles. It does not examine all available Rust code, or even all code on crates.io

[nikomatsakis]

Discussed in @rust-lang/compiler meeting. Indeed some fallout from these changes was expected but a warning cycle is still appropriate. The only reason to skip a warning cycle is supposed to be if a warning cycle is technically infeasible -- in which case we are to use crater to make a best effort to have all affected crates get updated beforehand.

[nikomatsakis]

triage: P-high

Specifically, high priority to add a warning cycle.