Rescope temp lifetime in let-chain into IfElse

[dingxiangfei2009]

This is a proposal born out of #103108 as a potential answer to #104843.

Three tests are failing from this.

• tests/ui/drop/drop_order.rs
• tests/ui/mir/mir_let_chains_drop_order.rs
• tests/ui/nll/issue-54556-niconii.rs

The first two tests fail because of the change of drop order as intended by this PR. The third test is because the program compiles instead.

[rustbot]

r? @oli-obk

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

[dingxiangfei2009]

cc @est31

I have tested this branch on several of my repositories that I am working on and nothing obvious has broken. Shall we have a crater run?

[est31]

@bors try

[bors]

⌛ Trying commit 003f3cd8d3cecb9b20f5105400499741c936e92a with merge ece70b22820429ebf0c6d388c32356a14c58fc2f...

[bors]

☀️ Try build successful - checks-actions
Build commit: ece70b22820429ebf0c6d388c32356a14c58fc2f (ece70b22820429ebf0c6d388c32356a14c58fc2f)

[dingxiangfei2009]

I have updated the failing tests to show the effect of this change.

[oli-obk]

@craterbot run mode=build-and-test

[craterbot]

👌 Experiment pr-107251 created and queued.
🤖 Automatically detected try build ece70b22820429ebf0c6d388c32356a14c58fc2f
⚠️ Try build based on commit 003f3cd8d3cecb9b20f5105400499741c936e92a, but latest commit is 92550dd. Did you forget to make a new try build?
🔍 You can check out the queue and this experiment's details.

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🚧 Experiment pr-107251 is now running

ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[craterbot]

🎉 Experiment pr-107251 is completed!
📊 128 regressed and 99 fixed (253904 total)
📰 Open the full report.

⚠️ If you notice any spurious failure please add them to the blacklist!
ℹ️ Crater is a tool to run experiments across parts of the Rust ecosystem. Learn more

[traviscross]

@rustbot labels +S-waiting-on-author

[est31]

@nikomatsakis thanks for the response and thanks for the discussion in the lang team.

I will try to make a writeup the next rainy weekend I encounter (I'm sure there will be one soon, this is the season for the rains in my area ☔).

Some preliminary comments: I think it's great to have an RFC about a general effort to limit lifetimes, as I'm sure there is more potential about too long lifetimes. But I don't think we should block let chains on this greater effort. I regularly encounter cases where I'm writing nested if's where let chains would have allowed me to collapse a level.

Given that this change is smaller in scope than the RFC, we might get away way more easily with not doing this in an edition bump.

---

Side note: regarding your blog post, I recently touched Go, and wanted to to specify an *int via a literal. It was weird to me that Go complained about &42 but was fine with v := 42 followed by &v. Keeping &42 working in Rust would be nice.

[bors]

☔ The latest upstream changes (presumably #113970) made this pull request unmergeable. Please resolve the merge conflicts.

[traviscross]

@rustbot labels +I-lang-nominated

Let's discuss to confirm that we do not want to use Rust 2024 for this, as had been discussed above.

[traviscross]

@rustbot labels -I-lang-nominated

We discussed this in the lang meeting today. The consensus was that we'd like this to land in nightly with a feature gate. After that, we'd like to see the result of crater runs and some analysis of the migration story so we can consider landing this in an edition. This may also call for an RFC.

[dingxiangfei2009]

I am going to propose if_let_rescope as the feature gate in front of this change.

[dingxiangfei2009]

I am thinking of running a crater experiment with +rustflags=... or something to force rustc to assume if_let_rescope feature. I am unaware of such an option so far.

[est31]

I don't think a crater experiment would be helpful here if the code is just for edition 2024. There is practically no crate out there with edition 2024. Forcing code onto edition 2024 will give problems as well because 2018 code will then also have the 2021 warnings and even more different for 2015 code.

[ehuss]

I am thinking of running a crater experiment with +rustflags=... or something to force rustc to assume if_let_rescope feature. I am unaware of such an option so far.

If your goal is to see what would fail with the feature enabled, the best thing to do is to temporarily modify your PR to make it always enabled. Essentially change it to something like let mut rvalue_scopes = RvalueScopes::new(true);.

[traviscross]

@rustbot labels +I-lang-nominated

This PR is about shortening, over an edition, the scope of temporary lifetimes in if-let scrutinees so that they end before the else block. The current rules are a particular problem for let-chains which we hope to stabilize.

We last discussed this on 2023-10-25. We had some questions about the migration story here and specific asks for further analysis.

Since then, Ding has done a crater run:

• [Crater run experiment] let-chain rescoping #125151

And on the basis of that, has assembled a report analyzing the results and discussing the migration strategy:

• https://hackmd.io/@dingxf/Hk9PAnmEC

Let's nominate to discuss.