[Tracking] Incremental Systems Rethought

This issue lists all current list points in order to complete the Incremental Systems Rethought project goal. These should be done in order-of-appearance, and they should be self-contained enough that anyone can pick them up.

If you pick one of these up, please ping me in the resulting PR or contact me through Zulip.

Reading materials

• Rustc's docs.rs
• Rustc-dev-guide
  • Specifically, the "Queries: demand-driven compilation" section and all sub-sections.
• RFC PR#3118
• The project goal itself and linked documents.

Progress

• Merge RFC.
• Build-over-check
  • We should be able to run cargo check && cargo build, and have cargo build be able to re-use cached information by cargo check.
  • Being that cargo check monomorphizes, I'm tracking monomorphization information and use it in cargo build.
  • Other approach: Track name resolution, with very primitive type information (limits us in many aspects, but lower memory fingeprrint).
• Per-owner querying.
  • We currently have a lot of queries that work on the whole crate (e.g. codegen_crate, collect_and_partition_mono_items), we should fragment these, trying to keep overhead as low as possible.
• Cargo: Go top-down instead of bottom-up.
  • Currently we completely codegen the lowest dependencies in a crate tree. We should start by doing nameres in the root, and then compiling the dependency judging by that name resolution information (So, don't take into account unused items)

Experiments

Workflows

I (@blyxyas) am currently the only person actively working on this. I've received a couple of offers for help, so it might evolve into a working group or other kind of organized effort.

I'm holding bi-weekly meetings with Jack Huey @jackh726, discussing the design.