fix(vscode): keep Rslint healthy when rstack.config has no define.lint()

[fi3ework]

Summary

A root rstack.config.* with only define.fmt() lights the lint stack (by design: the extension never reads the Rstack config itself). rstack's lint shim then calls process.exit(1) because it finds no define.lint(), and since the editor-shipped lint worker evaluates that shim in-process, the worker died mid-rslint/loadConfigs and took the Go server and the language-client transport with it. That is the "No lint configuration found" → connection-disposed → server-restart cascade in #93.

 lint worker (bridged)
   ConfigModuleHost
-    loadFresh(shim)              # shim calls process.exit(1) → worker dies
+    withProcessExitAsThrow(loadFresh)(shim)
+      process.exit → throws { code: RSTACK_CONFIG_PROCESS_EXIT }
+      → failed load result, worker and Go stay up
   ConfigTransactionAdapter.loadConfigs
     missing dependency → `missing`
+    RSTACK_CONFIG_PROCESS_EXIT → `unconfigured`
 editor (Rslint.ts)
   rstack/rslintConfigDependency
     missing → disabled (poll)
     plugin  → disabled (poll)
+    unconfigured → running, detail "no define.lint() in rstack.config.ts", one info line
     error   → crashed

Behaviour for a formatting-only project: Rslint still starts (the shim is the only thing that can decide), reports running with a detail, logs one info line, and produces no diagnostics, no toast, no crashed, and no 60-second dependency poll. Adding define.lint() later is picked up by the same worker through the existing bridged rstack.config.* watcher, without a restart. Native runtimes build their ConfigModuleHost exactly as before.

Why not "don't start Rslint at all": the extension cannot know whether define.lint() exists without evaluating the config, and a text probe is forbidden (AGENTS.md) and wrong under shared config layers. The shim's refusal is currently expressed only as process.exit(1); once rstack exports a marker or throws a typed error instead, the wrapper can go and the editor can classify by contract.

Known trade-offs, recorded in packages/vscode/AGENTS.md:

• Any process.exit during a bridged config load (including one from the user's own config) is classified unconfigured. With rstack 0.8.2 the shim's refusal is the only exit in its import graph. A tighter check (exit code 1 + caller is the shim path) is possible as a follow-up.
• After a catalog with define.lint() has committed, removing it makes Go keep the last-good catalog while the status says no define.lint(); rs lint would refuse.
• The shim's own No lint configuration found… line and Go's Skipped config … line still reach the Rslint Output channel beside our info line.

Evidence

• Before: the new Rstack fmt-only config E2E suite, run against dist/ built from the pre-fix source, fails with the issue's error:

  Rslint became crashed while waiting for the unconfigured running detail:
  Pending response rejected since connection got disposed (…/@rslint/core)
  

  After: pnpm lint ✓ · pnpm test:unit 365 passed, 0 failed · lint E2E (Rstack fmt-only config, Missing config dependency, Rstack lint bridge, JS config tests) all Exit code: 0 on macOS with VSCODE_CLI=1. The fmt-only suite asserts no crashed state, no diagnostics, no warnings, then adds define.lint() and asserts no-debugger appears from the same lint-worker pid.
• Manual check in a clean VS Code profile with the packaged VSIX on a define.fmt()-only project (rstack 0.8.2): status bar hover shows running — no define.lint() in rstack.config.ts, the Output channel has the single info line and no connection-disposed or restart errors; uncommenting define.lint(...) and saving produces the diagnostic without a restart.

Merge Danger

Door: two-way

Revert restores the previous behaviour; no settings, protocol versions or published surfaces change. The new unconfigured verdict is editor-internal (worker ↔ extension host).

Blast Radius: lint-bridged

Only bridged lint runtimes (a root rstack.config.* without a tool-native rslint.config.*) take the new path, and only when the shim refuses. Native Rslint runtimes, Rstest and fmt are untouched.

Related Links

• Closes Rslint starts and fails for a formatting-only Rstack config #93
• rstack-cli: the shim's process.exit(1) on a missing lint configuration (fix(lint): report missing lint configuration rstack-cli#493, discussed in rs test / rs lint run with an empty config when define.test() / define.lint() is missing: intended or an oversight? rstack-cli#490); a follow-up there to export a marker or throw a typed error instead would let the editor drop the wrapper.

Checklist

• Tests updated (or not required).
• Documentation updated (or not required).

[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 2026-10-10T03:58:16.955448Z	9119895	PR opened

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.