Patterns of empty types

[nikomatsakis]

Metadata
Owner(s)	@Nadrieril
Team(s)	compiler, lang
Goal document	2024h2/Patterns-of-empty-types

Summary

Introduce an RFC for never patterns or other solutions for patterns involving uninhabited types.

Tasks and status

• Stabilize min_exhaustive_patterns (@Nadrieril)
• Defer linting of newly-unreachable arms til later (@Nadrieril)
• Write up "never patterns" RFC (@Nadrieril)
• Submit RFC, discuss and wait for acceptance (lang )
• Implement never patterns (@Nadrieril)
• Amend exhaustive_patterns to only apply to the case of empty types behind references (@Nadrieril)
• Decide on stabilization of exhaustive_patterns (lang )
• Decide on linting of the newly-unreachable arms (lang )

[nikomatsakis]

This issue is intended for status updates only.

For general questions or comments, please contact the owner(s) directly.

[Nadrieril]

Here's the status we're at today, at the start of the goal period:

• The exhaustive_patterns feature has been usable for years but was deemed not-stabilizable-as-is; I aim to deprecate it as part of this goal;
• The min_exhaustive_patterns feature has been proposed, implemented, and approved for stabilization. Stabilization was blocked on improving lints; this is about to be resolved;
• My proposal for never patterns got sufficient approval to enable experimentation. The unstable feature is now in a usable state, though not complete;
• I have an RFC draft for never patterns lying around. It needs to be fleshed out and updated in light of min_exhaustive_patterns being stabilized, after which we can start the RFC process.

[Nadrieril]

Update since last time:

• The min_exhaustive_patterns feature has been stabilized, it will be accessible in 1.82.0;
• The stabilization caused issues because we went from requiring some match arms to warning against them. There was some discussion on the topic, and it was decided to revert the linting behavior for now and find a solution that has a smoother transition.

Next steps:

• Backport the approved linting change to beta;
• Implement whatever linting mitigation we end up deciding on;
• Finish writing up the never patterns RFC.

[Nadrieril]

Update since last time:

• We reverted the linting behavior until an unspecified date

Next steps:

• Decide what we want to do with linting;
• Write the never patterns RFC.

[Nadrieril]

Update since last time:

• The never patterns RFC was posted.
• Feedback on the RFC suggests that the question of "which arms can be omitted" isn't as orthogonal as I hoped, so I'll switch focus to that.

Next steps:

• Amend exhaustive_patterns to only apply to the case of empty types behind references;
• Propose stabilization of exhaustive_patterns, i.e. allow omitting a match arm even when the empty type is behind a reference.

[Nadrieril]

No progress since last time.

[Nadrieril]

Summary of the progress over the semester: I went at my own pace which is a slow pace. I wrote and posted the never patterns RFC as an attempt to make progress without figuring out the whole picture, and the general feedback was "we want to see the whole picture". Next step will be to write up an RFC that includes a clear proposal for which empty patterns can and cannot be omitted. This is 100% bottlenecked on my own writing bandwidth (reach out if you want to do this together!).

I did not end up making much use of the project goal structure, so I decided not to renew this goal for 2025H1. I will keep working on it at my own pace. Stay tuned for future developments!