1. What cost the most. Designing the host interface. The brief named parse_unit but not how check_imports and declared_once would reach a Parsed that Rust holds by index. I read about 8 files to settle this (project_members hosts, sysroot_cache, BtHost, ParsedView, fs.rs, block_on). The second biggest cost was a full clippy + it round (about 8 minutes) lost because another session reset lowered/ while I worked.

2. What I looked up that the brief could have given me:
   • fs::read_to_string is async, so the collect functions need block_on on the Rust side.
   • How to build an fs::Path from text (let p: fs::Path = ".").
   • Hosts that already existed: ProjectMembersForeignHost has sha256_hex and read_beside, but its Rust implementations are private to project.rs, which I was told not to touch.
   • Noted and Checked are Rust types, so the inference callback could not move.
   • The Unreadable { what, why } pattern that keeps “cannot read” byte for byte.
   • What IoError prints, to check the cause text no test reads.
3. Checks I ran:
   • Useful: the lower-std cycle with a fixpoint check, tests/compiler once with my copy of the binary, the it filters plus the guard set together with clippy, pre-push-check, and rs2nika --score.
   • Wasted: the first clippy + it run (on the reset tree), and --lib modules::, which matched 0 tests because the module has no unit tests.
4. Mistakes in my report:
   • Listing the tree reset under Found: was questionable: it is a process problem, not a code defect, yet it cuts the batch.
   • Disk freed: 0 left out the binary copies I deleted from the scratchpad.
5. The change that would have saved the most: a builder.md rule to git status the lowered/ files and the index (lowered.rs, which sits outside lowered/) right before tests. Second: a brief that states each host method’s signature, not just its name, including how the per-file checks reach Parsed, so no reading is needed to design the interface.

This site is open source. Improve this page.