chore: merge train 241 (v0.5.1620)

[proggeramlug]

Merge train 241 — ethers compiles, released as v0.5.1620.

Contents

PR	Change
#10801	fix(hir): memoize class_mutates_capture recursion to stop exponential HIR lowering — fixes #10757

The diagnosis is the valuable part

Compiling ethers 6.17.0 from real source never finished: HIR lowering of @noble/curves' weierstrass.js spun 8+ minutes on one core before being killed. A probe during this session's audit measured 19 min 22 s at 99.1% on one core, still not past module collection, where node loads the same entry in 0.19 s.

It is a bounded superlinear blowup, not a hang — and establishing that is what made it fixable. Instrumented call counts to class_mutates_capture grow 252 → 35,770 → 688,870 as the Point class gains one arithmetic method at a time, fitting (Bⁿ-1)/(B-1) for n=9 (the hardcoded MAX_NESTED_CLASS_DEPTH) at branching factors B=1, 2, 3. It terminates; just not in any practical time for a realistic method count.

Root cause. for_each_nested_capture was written to find a class genuinely nested inside another class's method body — class Outer { make() { return class Inner {} } } — by scanning member bodies for capture-forwarding constructions. A self-referential new Self(...) matches that same scan, so Point is misread as nested inside Point, and the recursion re-enters the class it started from once per self-constructing method, at every depth up to the cap, redoing the identical (class, id) subproblem each time. Point's double, add, fromAffine, multiplyUnsafe all construct new Point(...) forwarding the outer context — completely ordinary self-referential-class code.

Replacing a cap with an invariant is worth more than the speedup

The fix memoizes by (class_name, id) with an in-progress set to break cycles, replacing MAX_NESTED_CLASS_DEPTH = 9. The new termination argument is "finitely many distinct (class, id) pairs" — a property of the algorithm — where the old one was a bound that happened to hold, and which as the PR notes could also have under-covered a legitimately deep but acyclic chain.

Measured, --no-auto-optimize --no-link to isolate lowering:

fixture	pristine	fixed
8 self-constructing methods	2.34 B	2.21 B
12 self-constructing methods	26.2 B (11.2× its own 8-method run)	2.26 B (flat, +2%)

With the fix, the whole ethers@6.17.0 tree — ethers + @noble/curves + @noble/hashes + @adraffy/ens-normalize + aes-js, 153 modules — lowers and codegens natively with 0 JS fallback in about a minute.

What this does NOT do, stated plainly

ethers compiles; it does not yet link. The final link fails on undefined references from ethers/src.ts/crypto/crypto.ts's export { createHash, createHmac, pbkdf2Sync, randomBytes } from "crypto" re-export shape. That is tracked as #10802, and it is the same defect as #10432, which has had a six-line reproducer since 2026-09-06:

export { createHash } from "crypto";   // dep.ts — undefined perry_fn_dep_ts__createHash at link

Confirmed during this session's audit with a two-file, no-ethers reproduction. So #10757 stays open: "compiles" is not "links", and closing it would sever the thread from the original never-finishes report to the remaining blocker.

An incidental gap closed

The parity harness gains PERRY_COMPILE_TIMEOUT (default 300 s). It previously timed only the executed binary (PERRY_RUN_TIMEOUT), never the compile — so a compiler hang on any one fixture wedged the whole harness instead of failing that fixture. That is the same shape as three other things fixed on main today: a check that cannot distinguish "still working" from "stuck". It deserves more than the parenthetical the PR gave it.

Validation

Assembled on c1d9f73e58; source head asserted fresh; PR fully represented — zero missing insertions; no attribution trailers. Ten cheap gates, cargo check --workspace --all-targets under -D warnings, all five pinned artifacts byte-identical before and after, six unit suites with an empty failing set, and lint complete at 6-of-6 with nothing outside the known-red public-baseline step.

Both compiler-output suites at failed_workloads=[] and repsel_census rc=0 wasted_promotion=False.

The PR verifies HIR output byte-identical via --print-hir only at fixture sizes the unfixed pass could finish — the honest limit, and why the sweep is weighted to this change's blast radius. Gap sweep at PERRY_RUN_TIMEOUT=30, seven areas, 212 fixtures, every area asserted live, zero unexplained regressions:

class 84   gc_ 54   import 20   property 18   closure 14   module 14   extends 8

[coderabbitai]

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 05294a99-3945-43e7-a19f-31471d435c70

📥 Commits

Reviewing files that changed from the base of the PR and between c1d9f73 and 3bd2796.

⛔ Files ignored due to path filters (1)

• Cargo.lock is excluded by !**/*.lock

📒 Files selected for processing (6)

• CLAUDE.md
• Cargo.toml
• changelog.d/10801-exponential-class-capture-lowering.md
• crates/perry-hir/src/lower/shared_mutable_capture.rs
• run_parity_tests.sh
• test-files/test_gap_10757_self_referential_class_capture.ts

 _____________________________________________
< My `#FF570A` highlighter marks all the TODOs. >
 ---------------------------------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}