[kixelated]

Problem

The overnight PRs (#13, #15, #16, #18, #26, #27, #28) left follow-ups and decisions, which the maintainer settled on 2026-10-24 (prompts below). They need to be quests.

Approach

New quests, each ranked in its milestone:

• m2 MIR lowering spike [L]: port the whole lowering to MIR on the codec and task fixtures; its report decides whether the translator moves. Added to the Required sections of Drop and Async, to rewrite them to its report before starting.
• m2 Stable names [S]: always print a generic instance's suffix and a trait-impl method's trait, in today's format.
• m2 Diverging calls through closures and generics [S], from fix(ts): type-check values of type ! where a value goes #28.
• m2 Chrome timeout [XS], A/B routes [S], and error-path workload [S] for just ts bench.
• m2 Wide-integer fast paths [S], from feat(ts): wide integers are numbers while safe, and each integer type is branded #26, each kept only if measured faster.
• m2 Delete the C++ generator's dead modes [S]: scaffolding mode and exception-style async.
• m3 No-C check [XS] for the moq-pure-crypto corpus.
• m3 Early cutoff for edit rebuilds [M]: build Relink don't rebuild: improve public api hashing using an rmeta reachability graph rust-lang/rust#156940's -Z public-api-hash on a scratch toolchain and replay moq-net's real edits through it, counting unchanged hashes, span-only breaks, and async fn breaks, with incremental on and off, plus an rmeta determinism check. The report picks: carry #156940, help upstream, or stop. Relink without rebuilding notes it may change its plan.
• m3 ra-jit in the build benchmark [S]: a variant on the moq-net edit, recording what it refuses or hands back to cargo.

Build-speed research behind them (2026-10-24): rustc's SVH covers every body, so a reused dependent fails with E0460, and no build tool gets early cutoff for Rust today. A hash-patching wrapper is dropped: tinymold ran that spike (tinyhumansai/tinymold#13, ADR 0007) and hit E0460, and mbx rejected it as unsound (jdx/mr-boxington#645). mbx's numbers put dependents at about 1.1 s of a 1.7 s edit with incremental on, so the spike measures what incremental leaves and classifies edits by mbx's list of interface-relevant bodies. headstart (early interface-only metadata for clean builds) is not chosen for now; it also found rmeta def-id order unstable across incremental sessions, hence the determinism check. The OpenAI-funded cg_clif performance goal speeds up the backend, 6% of moq's edit, so it is not relevant. ra-jit hot-patches changed functions through Cranelift into the linked binary, claiming 12-24x, but treats async fn bodies as signatures.

Output goals, recorded in ts/README.md's Design and the m2 Plan: correctness first, speed second, and the consumer's final tree-shaken, minified JS third; readable output is not a goal; output stays .ts; types are solid at the API boundary (root items and printable types are branded), while internal casts stay since they compile away. The MIR spike now weighs its 1.4-2.6x speed ahead of its roughly 18% larger bundle. The README's opening line drops "readable".

• m2 Tree-shaking spike [M]: classes everywhere, free functions everywhere, or free functions behind root-type class facades, measured on speed, a consumer's bun build --minify bundle, and caller code, with a survey of RxJS, Zod 4 mini, Valibot, Effect, date-fns, and wasm-bindgen.
• m2 Tree-shakable runtime [M]: rt becomes individually exported functions, kept only if no slower, with a tree-shake test.

Into existing quests: Async gains ? in Poll-returning functions (about 224 ready!(..)? sites in moq-net) and the tsc error on a () future awaited as a statement; Drop gains owning a temporary held across an .await instead of refusing it.

Merged with main after #13, #21, #26, and #30 landed. #16 is still open: it deletes ts-async-spike.md, which these link to, and rewrites Drop and Async. Whichever merges second aligns the quests.

Impact

No public API change. ts/README.md records the output goals.

Alternatives

None beyond the prompts below.

Follow-ups

moq's side (the pure-crypto epic, the C name clash, the crate split rerun) is in moq-dev/moq#6659.

Decisions

• Async feat(ts): async fns as generators and drops from THIR scopes, from the async spike #16: ✅ Merge feat(ts): async fns as generators and drops from THIR scopes, from the async spike #16 + MIR spike / Go to MIR now / Merge, stay on THIR
• MIR spike scope: ✅ Whole lowering, 2 fixtures / Async and drops only
• Wide feat(ts): wide integers are numbers while safe, and each integer type is branded #26: ✅ plain numbers, but brand every integer type with Rust type names / bigint / limbs everywhere
• Naming: ✅ Always suffix / Leave as is
• Suffix format: ✅ Rust-like suffix / Hash suffix
• Crypto feat(bench): measure moq-relay on pure-Rust crypto #13: ✅ Adopt, all 7 recs / Keep measuring / Drop it
• moq crypto split: ✅ Epic, 3 quests / One quest
• TS gaps: ✅ ? in Poll fns, ✅ Await/drop gaps, ✅ ! edge cases / Unused self param
• TS perf: ✅ Chrome timeout, ✅ A/B routes, ✅ Wide-int fast paths, ✅ Error-path workload
• UniFFI: ✅ Drop C++ dead modes, ✅ C name clash, ✅ moq crate split rerun / Namespace edge cases
• Build: ✅ No-C check / Early cutoff upstream / moq-net split
• Ranking: Correct > size > speed / ✅ Correct > speed > size
• Output: ✅ .ts, brands at boundary / .js + .d.ts / Both
• Shaking: No classes, free functions / Classes, prove unused / Runtime only / ✅ (free text) "Spike both and measure, see what other libraries do; we want libraries that don't force callers into ugly code; maybe make it optional"
• Variants: ✅ Classes everywhere, ✅ Free functions everywhere, ✅ Free inside, classes at roots / ts.toml option
• Runtime: ✅ Runtime always free / Same as types
• Boundary: ✅ Root items only / Every exported item. Clarified: casts are free in JS, so internal casts stay; no quest.
• Bundler for size: ✅ Bun build, tree-shaken and minified (what the bench already uses) / Vite (considered; the maintainer said Bun is fine)
• Build plan after the thread research: ✅ Re-scope cutoff spike, ✅ ra-jit bench variant / headstart later / None
• Early cutoff goal: ✅ Spike, research first / Cargo replacement / Not now
• First step: ✅ Measure real edits first / Carry #156940 / Wrapper prototype / Upstream only

(written by Opus 5.5)

🤖 Generated with Claude Code

https://claude.ai/code/session_0115AT2ue6x9cT5e5J9SZtCn

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-24T18:10:54.522562Z	eae358a	New commits
🔒 Security Review	✅ Completed 2026-10-24T18:11:40.750400Z	eae358a	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[kixelated]

Merging. Summary:

• New m2 quests: the MIR lowering spike, stable names, ! edge cases, Chrome timeout, A/B routes, the error-path workload, wide-integer fast paths, the tree-shaking spike, the tree-shakable runtime, and deleting the C++ generator's dead modes.
• New m3 quests: the no-C check, early cutoff (rust#156940 replayed over real moq-net edits), and an ra-jit benchmark variant.
• Async and Drop gain the spike gate and the gaps found since. ts/README.md records the output goals.
• Checks: just check and quest check --base origin/main --branch plan/overnight-followups pass, CI passes, and Codex had no findings on eae358a. just test was not run, since only quest files and README prose changed.
• moq's side: quest: plan pure-Rust crypto, the C name clash, and the crate split rerun moq#6659 (open, not merged).

(written by Opus 5.5)