Tracking Issue for polymorphization

[RalfJung]

This is a tracking issue for polymorphization (-Zpolymorphize).

Current (2024-05) status:

• no one has worked on it for 4 years yes
• It mostly works, but there are some bugs remaining where it incorrectly polymorphizes
• the current implementation is very limited and not really useful
• @lcnr wants to remove the current implementation and replace it with a different approach

Open problems

• see the issues tagged -Zpolymorphize.

Unresolved Questions

• Does this in any way interact with which code gets monomorphized? For fully monomorphic functions, we sometimes monomorphize and codegen them for no good reason (see here or here). This is very hard to avoid, but it would be good to limit this to fully monomorphic functions. So even with polymorphization, ideally we will not generate code (and evaluate constants) for any function that's not already considered by the collector. Currently, it seems like everything is fine here, but we should have an eye on this.

Implementation history

• Initial PR: Polymorphization #69749
• lots of follow-ups

[compiler-errors]

Polymorphization was removed in #133883, for the record.

[jplatte]

Why was this closed? I've seen it brought up multiple places now. It seems like there is no longer an implementation of it, sure, but not because the feature was bad, just because the implementation was bad. We have lots of tracking issues for non-implemented features, right?

[RalfJung]

We don't typically use tracking issues for features without an up-to-date accepted RFC or MCP, no. There's nothing to "track" if there's no actionable implementation plan.

[jplatte]

Right so the difference here is that there hasn't been a formal proposal for this, makes sense.

Though it seems like a new implementation plan exists as per the fourth bullet point in the issue description?

Anyways, thanks for the quick answer :)

[RalfJung]

Though it seems like a new implementation plan exists as per the fourth bullet point in the issue description?

@lcnr has some ideas; if they want to have an issue open to track them they'll create one. :) But this issue here specifically tracked the old implementation, which is gone, and therefore this issue has served its purpose.