Assumptions on Binders

[tomassedovic]

Metadata
Point of contact	@BoxyUwU
Team(s)	types
Goal document	2026/assumptions_on_binders

Summary

Attempt to implement a version of -Zhigher-ranked-assumptions which works for all binders, not just witness types of futures.

Tasks and status

---

Note: we have updated the body to match the 2026 goal. Your original text is preserved below.

Details

Metadata
Point of contact	@BoxyUwU
Team(s)	types
Goal document	2026/assumptions_on_binders

Summary

Attempt to implement a version of -Zhigher-ranked-assumptions which works for all binders, not just witness types of futures.

Tasks and status

[tomassedovic]

This issue is intended for status updates only.

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

[BoxyUwU]

An unstable flag -Zassumptions-on-binders has now landed which depends on the new trait solver and implements some basic support for OR region constraints and uses them to support taking into account implied bounds on binders. I've opened a project-assumptions-on-binders repo/team to track the ongoing work to finish the feature and stuff relating to it: https://github.com/rust-lang/project-assumptions-on-binders

there's also an associated zulip stream: #t-types/assumptions-on-binders

[BoxyUwU]

I've put together a fancy graph to try and represent all the stuff that needs to happen for -Zassumptions-on-binders to be "done": https://raw.githubusercontent.com/rust-lang/project-assumptions-on-binders/refs/heads/main/roadmap.svg. each node is tracked as an issue on the project-assumptions-on-binders repo

right now there's mostly just a bunch of random bug fixes being taken care of now that the initial implementation has landed. thanks @Dnreikronos for some of the PRs for that. I intend to look into what it would take for the standard library or the compiler to compile with -Zassumptions-on-binders and if that's easy get that to work, otherwise/afterwards I intend to work on a custom testing thing for assumption on binders so that writing tests for region handling edge cases is vaguely possible

[BoxyUwU]

I've spent the last few weeks looking into what it would take to get core/std to compile with -Zassumptions-on-binders. So far I've encountered:

• Ambiguity errors caused by differing region constraints between trait solver candidates (@Dnreikronos's trait_solver: normalize next-gen region constraints rust#158404 will fix this)
• Lack of OR region support in regionck, this is a bit tricky but the cases that encounter it seem to be very trivial and we can get away with a fairly basic fix here.
• Lexical region resolution woes. When adding basic OR region support we then get lexical region resolution errors in always applicable checking for specializations. I don't know what's going on here and have just disabled this check for now to progress further xd
• Exponential blowup in region constraints. After fixing the above core now compiles, but compiling object OOMs my computer 😅 It seems to wind up creating absolutely massive region constraints consisting of the exact same constraints repeated many many many times over. I think fixing this probably isn't too bad and just requires us to just be slightly more diligent in not duplicating region constraints all over the place lol. Have yet to properly look into a fix for this

@Dnreikronos has a few open PRs for fixing assumptions on binders problems: rust-lang/rust#157984 rust-lang/rust#158362 rust-lang/rust#158404 rust-lang/rust#158588

[BoxyUwU]

I've spent some more time looking into this and still having some trouble getting object to compile without hanging/OOMing. I implemented splitting region constraints stored in all OR options out of the OR of ANDs in hopes that it would fix the memory usage issues (which it does seem to have) but now it seems to have exponential time problems ✨

draft PR that shows the above WIP stuff for getting std to compile: rust-lang/rust#158872

[BoxyUwU]

Haven't made more progress trying to get std to compile. Instead I've triaged some of the "weird async bug" examples we've got in our issue tracker. Some assortment of issues are fixed from the big "weird async errors" tracking issue #110338:
rust-lang/rust#153638: https://godbolt.org/z/xM7GGP4en
rust-lang/rust#82921: https://godbolt.org/z/vobW3WTbG
rust-lang/rust#60658#issuecomment-1509321859 : https://godbolt.org/z/8Gzfs99Ks
rust-lang/rust#126044#issuecomment-2155830067 : https://godbolt.org/z/1Pno7dM79

I also looked into why rust-lang/rust#153558 isn't currently fixed and the answer turned out to be that it normalizes higher ranked stuff outside of the trait solver so there are no assumptions yet. Unforchies :3

Other than this I've also spent a bunch of time talking with @lcnr about where we're going with assumptions on binders and the plan there which has resulted in us coming up with an idea for a minimal stabilization which only applies to coroutine witness type binders and has a more limited ruleset for how we handle placeholders (making it less scary).

I intend to rewrite this project goal a bit with this new plan in mind.

@Dnreikronos's #158362 has been merged and should make normalization not ICE. Their PR to implement proper diagnostics is also ready to be merged (#158588).
@amirHdev wrote #159964 which fixes a minor issue for normalization goals
@ada4a fixed a lack of derive causing problems for r-a #160164

[BoxyUwU]

I've opened #742 to change this project goal slightly to make clearer the scope and plans we have.

@Dnreikronos has opened rust-lang/rust#160513 to automatically enable -Znext-solver when -Zassumptions-on-binders is enabled which should make the feature nicer to use :3

Other than that not much has happened AFAIK. I've mostly been thinking through the changes we're going to want to make so I can write mentoring instructions for them all.

[BoxyUwU]

@khyperia and I have spent a bunch of time talking about assumptions on binders getting her up to speed on what's going on here. She's since written rust-lang/rust#161443 implementing the start of a testing DSL for the core algorithm behind assumptions on binders. This will make a lot of the future work here a lot easier as we'll actually be able to test the edge cases we encounter directly instead of having to go through five layers of rust-code abstractions 😅

I (@BoxyUwU) have split rust-lang/rust#161306 out of my draft PR from my attempts to make std compile. This is an essential part of replacing old style region constraints with the new ones implemented for -Zassumptions-on-binders. It also generally will make working on this feature a lot easier as it simplifies a bunch of stuff conceptually :3

@Dnreikronos' PR to add proper support for diagnostics to the new style region constrants has landed: rust-lang/rust#158588. His PR rust-lang/rust#159987 has also been merged fixing an ICE in edge cases involving uses of opaque types and binders. Additionally he's opened a a PR relating to non_lifetime_binders ICEs under -Zassumptions-on-binders: rust-lang/rust#160955.

He's also triaged a bunch of issues on the rust-lang/rust issue tracker to figure out whether they're fixed by -Zassumptions-on-binders or not, and if they aren't what it would take to get them to be fixed: rust-lang/project-assumptions-on-binders#15

@ssenthilnathan3 opened rust-lang/rust#160497 fixing an ICE where we weren't resolving inference variables before computing the assumptions for a binder.

[BoxyUwU]

My PR to always store constraints in OR of AND form (rust-lang/rust#161306) has been r+'d thanks @lcnr and @khyperia for the reviewing :3 @khyperia's PR rust-lang/rust#161443 has also landed so we now have a very basic testing DSL which is already really quite handy :3

I've opened a bunch of issues on the rust-lang/project-assumptions-on-binders repo to track stuff that needs to happen.

@Dnreikronos has a PR (rust-lang/rust#161988) to rewrite region outlives constraints to true when leaving their universe if they're in the assumptions. I'd thought we did this as part of the initial impl of assumptions on binders but must have gotten lost at some point :3 This should fix a bunch of test cases: rust-lang/project-assumptions-on-binders#19.

they've also opened rust-lang/rust#161423 to fix an issue where we were dropping new style region constraints on the floor when handling type ops.

they've also also opened a rust-lang/rust#161423 to implement a more minimal version of -Zassumptions-on-binders which only affects binders on coroutine witness types and has simpler rules around handling of region constraints when leaving binders.

@adwinwhite has also started looking into some assumptions on binders stuff with rust-lang/rust#161963