Tracking Issue for -Znext-solver

[lcnr]

This is a tracking issue for the next-generation trait solver in rustc. It can be enabled on nightly by using the -Znext-solver flag. This work is lead by the @rust-lang/initiative-trait-system-refactor.

About tracking issues

Tracking issues are used to record the overall progress of implementation. They are also used as hubs connecting to other relevant issues, e.g., bugs or open design questions. A tracking issue is however not meant for large scale discussion, questions, or bug reports about a feature. Instead, open a dedicated issue for the specific matter and add the relevant feature gate label.

Links

Original Types Team MCP: rust-lang/types-team#58
Types Team Tracking Issue: rust-lang/types-team#76
Types Team announcement and Roadmap: https://blog.rust-lang.org/2023/01/20/types-announcement.html
EOY 2023 Update Blog Post: https://blog.rust-lang.org/inside-rust/2023/12/22/trait-system-refactor-initiative.html
2024 Types Team Update Blog Post: https://blog.rust-lang.org/2024/06/26/types-team-update.html
EOY 2024 Update Blog Post: https://blog.rust-lang.org/inside-rust/2024/12/04/trait-system-refactor-initiative.html
-Znext-solver=globally project board: -Znext-solver=globally

Milestones and future work

For a complete list of changes, see the list of merged PRs. We also track the decisions we've made - and decisions we still have to make - in https://github.com/rust-lang/trait-system-refactor-initiative/issues. We mostly discuss these issues on zulip and in Types Team meetings.

Major milestones

• implement the skeleton of the updated trait solver #105661 initial skeleton of the trait solver
• -Ztrait-solver=next: stop depending on old solver #113317 full independence of the old trait solver implementation
• [experiment] -Ztrait-solver=next-coherence but hacks that make things not fail anymore #113991 initial crater run with -Znext-solver=coherence
• Rework negative coherence to properly consider impls that only partly overlap #112875 always use next-gen trait solver for feature(with_negative_coherence), which is used while compiling the standard library
• Implement T-types suggested logic for perfect non-local impl detection #122747 always use next-gen trait solver in the non_local_definitions lint
• Uplift next trait solver to rustc_next_trait_solver #126614 uplift the next-generation trait solver into a separate library
• stabilize -Znext-solver=coherence #121848 (reverted) stabilize -Znext-solver=coherence again #130654 stabilize -Znext-solver=coherence
• [DO NOT MERGE] bootstrap with -Znext-solver=globally  #133502 bootstrap with -Znext-solver=globally
• [DO NOT MERGE] bootstrap with -Znext-solver=globally  #133502 (comment) initial perf run with -Znext-solver=globally
• fix remaining known issues -Znext-solver=globally
• initial crater run with -Znext-solver=globally
• stabilize -Znext-solver=globally

[alex]

Is it useful to file bugs with crates that don't compile under -Ztrait-solver=next? (Specifically with an error of not implemented: overflowed on pending obligations)

[lcnr]

not yet as there are still a lot of failing crates with the new solver. will probably make an announcement blog post asking for testing once we're able to successfully compile our own tests and did the first crater run.

[ItsRedly]

Is the new solver SLG or DFS?

[lcnr]

The new solver uses DFS, same as the current stable implementation. I think the main issue is that SLG has different performance characteristics and causes hangs in cases where DFS succeeds (and the other way round, but unlike for DFS, there doesn't currently exist code which depends on SLG :3)

Please ask questions about the trait solver in #t-types/trait-system-refactor zulip channel.