resolve: Recover from indeterminate macro resolutions more agressively

[petrochenkov]

If we are in "forced resolution" mode and in-module resolution is indeterminate, don't give up and continue searching in outer scopes.

Fixes#53481

[rust-highfive]

r? @nikomatsakis

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

[petrochenkov]

r? @alexcrichton

[alexcrichton]

@bors: r+

[bors]

📌 Commit c9c79b7 has been approved by alexcrichton

[petrochenkov]

Nah, this is not a good solution, easy way to get compiling code with potentially inconsistent resolutions.
Need to try something more conservative first.
@bors r-

[petrochenkov]

The root issue here is that derive helper attributes like #[serde] in #53481 or #[my_attr] in the test require going through forced resolution, this is something that needs to be avoided.

Ideally, forced resolution should be an error recovery measure and should never result in successful compilation, except for the feature(custom_attribute) case, but that feature is probably going to be retired eventually.

I'll return to this until beta/stable, but for now it's blocked by some other work from #50911 (comment).

[norru]

Has consensus been reached on this resolution yet? I have been unable to upgrade Nightly past 2018-08-15 (build failures on dependencies).

[petrochenkov]

@norru
This PR as written will result in accepting code that needs to be rejected and thus poses a backward-compatibility risk.
The proper fix is less trivial, but I hope to implement it in the next few weeks.

In the meantime, I'd recommend to implement a workaround in log4rs (if that is the dependency you are talking about) and publish a minor version, or simply to stay on the old nightly.

[norru]

Hi @petrochenkov, thanks for the extensive explanation. I am not willing/keen on using a patched version of https://github.com/sfackler/log4rs so I'll stick to the old Nightly. How long do you think is going to take, for the sake of my own planning? @sfackler FYI

[nikomatsakis]

@petrochenkov Is this PR still blocked?

[petrochenkov]

This PR can be closed actually,
The only thing that can be reused from here is the test.
I hope to return to this issue this weekend or next week.
It would be nice to land refactorings from #53778 before that, but not necessary.

[Boscop]

@petrochenkov

or simply to stay on the old nightly.

I'm running into this issue with both nightly-2018-08-25 and nightly-2018-09-13.
Which old nightly would work with log4rs?
I need to use at least nightly-2018-08-25 because of rocket..

[petrochenkov]

@Boscop
The issue was fixed in #54086 (merged into rust-lang:master 10 hours ago), so the fix should appear in today's or tomorrow's nightly.

[ghost]

@petrochenkov thank you for being so responsive.

Everyone working with rocket and log4rs owes you one 🍺