expose ctx.version to workers with the enable_version_api compat flag

[sohpeach]

When a worker with the experimental enable_version_api compat flag has a versionInfo passed while its entrypoint is being fetched, ctx.version will be populated.

ctx.version is currently optional even when the enable_version_api compat flag is enabled because we have some cases where it might not be populated (dynamic workers, the ctx object created for the nonclass_entrypoint_reuses_ctx_across_invocations compat flag, and the single enable_version_api test in server-test.c++).

Similar to #6111, there's not much testing here. We haven't exposed a way to hit all these codepaths directly from workerd so we're relying on tests from the internal codebase right now. This can change once we expose VersionInfo to WorkerdDebugPort.

[ask-bonk]

I'm Bonk, and I've done a quick review of your PR.

This PR threads Worker::VersionInfo through the entrypoint/handler/ExecutionContext pipeline and exposes it as ctx.version behind the experimental enable_version_api compat flag.

Findings (highest severity first):

1. Missing test coverage: No test exercises ctx.version being populated with real data. The existing server-test (loopback binding calls accept version property) passes kj::none for versionInfo at the server level and tests the loopback binding, not ctx.version content. Consider adding a test that provides a VersionInfo with real id/cohort/key values and asserts they appear correctly on ctx.version.
2. Mutable version object: toJs() returns a plain unfrozen JS object — user code can mutate ctx.version.id etc. Consider freezing it.
3. JSG_TS_OVERRIDE combinatorial explosion: 4 branches with duplicated version type. The TODO acknowledges this but duplicating the 4-field version type is fragile.
4. Pre-existing: exports not traced in visitForGc: Not introduced by this PR but you're modifying this class — exports is a JsRef that should be visited.

This review was generated by an AI assistant and may contain inaccuracies.

[ask-bonk]

Review posted successfully on PR #6283. Here's a summary of the findings:

1. Missing test coverage — No test exercises ctx.version being populated with real data. The existing server-test tests the loopback binding, not the actual ctx.version content.
2. Mutable version object — toJs() returns an unfrozen plain JS object, so user code can mutate version metadata.
3. JSG_TS_OVERRIDE combinatorial explosion — 4 branches with duplicated version type definition, acknowledged by the TODO.
4. Pre-existing exports not traced in visitForGc — Suggested a one-line fix since the author is already modifying the method.

github run

[github-actions]

The generated output of @cloudflare/workers-types matches the snapshot in types/generated-snapshot 🎉

[codspeed]

Merging this PR will not alter performance

✅ 70 untouched benchmarks
⏩ 129 skipped benchmarks¹

---

_{Comparing swallace/ctx_version (6cf8ef5) with main (a0cfac4)}

Footnotes

1. 129 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩