Node-compat library: zod normal and jitless validation corpus

[pmatos]

Part of #227. Split from #234 because the validation/codegen cluster requires one PR per library and #234's autonomous run implemented the AJV slice.

Goal

Wire Zod's pinned upstream suite into the generalized Node-compat library harness and exercise it once normally and once with {jitless: true}.

Expected coverage

• roughly 1,900 upstream cases
• zero runtime dependencies
• regex, BigInt, Proxy, Reflect, and Symbol behavior
• the identical bundled corpus cross-checked against Node with an expected-count lock

PR bar

The Zod suite is green, or remaining failures are reduced and triaged as engine bugs with follow-up issues; test262 remains unaffected.

[pmatos]

Autonomous design decision for this run:

• Pin Zod v4.4.3, the latest stable release when Node-compat library: zod normal and jitless validation corpus #279 was assigned. A native Vitest run (typechecking disabled) is green at exactly 1,092 runtime tests across 104 v4 classic files.
• Run that same classic corpus in normal and {jitless: true} modes from one generated, filesystem-free bundle. The harness will lock exactly 2,184 registered cases on both jsse and Node.
• I considered separate zod / zod-jitless configs and a generalized multi-variant library runner. A generated Vitest adapter is narrower: one command, one pinned corpus, and no shared runner abstraction that another library has not needed.
• Typecheck-only Vitest projects, Zod v3, and v4-mini are excluded because they do not exercise the v4 classic runtime path named by this issue. Native TypeScript checks are not a JavaScript-engine corpus.
• The generated adapter will reuse published matcher/serializer code and the shared TAP runner on both engines. Any Node-only test imports will get symmetric bundle-time portability seams; cases requiring a materially new host API will be explicit rather than silently counted.

The unattended-run contract forbids waiting for design approval. The checked-in design is docs/specs/2026-07-17-zod-library-harness-design.md; I am proceeding with this vertical slice.

[pmatos]

Implementation update: native Zod v4.4.3 has 1,092 runtime tests across 79 classic files, so the normal+jitless lock is 2,184 (correcting my earlier exploratory comment that said 104 files). The final design concatenates two independently scoped copies of the same IIFE so module caches do not leak across modes.

One upstream case generates two 10 MiB buffers solely to stress base64/base64url throughput. That single regex case occupied JSSE at full CPU for several minutes. I chose a symmetric portability patch that scales both inputs to 64 KiB on Node and JSSE: it preserves large-input validation and the identical-bundle oracle while keeping the repeated corpus practical. The separate async-continuation blocker is tracked in #309 and remains visibly registered as one # SKIP per mode.

[pmatos]

Full test262 validation note: the default run completed 99,546/99,568 with 22 baseline-listed Intl date/time scenarios failing and 323 passes not yet present in origin/main:test262-pass.txt. To determine whether #279 caused the 22 reported regressions, I built untouched origin/main (176116b) in a detached temporary worktree and ran the exact 11 affected files against both binaries under the same host environment. Both binaries produced the identical result: 0/22, with the same 22 baseline regressions. The branch only changes optional-chain parsing, and its targeted upstream suite is 76/76 plus the new reserved-IdentifierName test at 2/2. I am therefore treating the full-suite result as pre-existing baseline/environment drift, leaving README.md and test262-pass.txt unchanged, and carrying the exact comparison in the PR validation notes.

[pmatos]

Implementation is ready for review in #320. The pinned Zod v4.4.3 corpus locks 2,184 normal+jitless cases against Node; Node reports 2,184/2,184, while JSSE reports 2,176 passing (including one visible jitless-only skip) and eight failing assertions tracked in #313-#315. The async-continuation and timer-portability decisions remain recorded in #309-#310. All focused quality gates are green; the full test262 baseline anomaly and direct origin/main comparison are documented in the preceding comment and PR body.

[github-actions]

🎉 This issue has been resolved in version 0.2.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀