fix(provider): remove model refresh during deactivation

[Vizards]

Related issues

Closes #268

Problem

Reloading a window or restarting the extension host logs Failed to refresh DeepSeek models during deactivate Canceled: Canceled during otherwise normal shutdown. The provider tries to clear model-picker entries through vscode.lm.selectChatModels() after VS Code has disposed the extension-host RPC connection, so the refresh cannot complete.

Changes and scope

Remove the deactivation-time refresh, its inactive-provider guard and global provider reference, and the unused English/Chinese deactivation-error strings. Deactivation now releases local logging resources synchronously. Startup, API-key, and configuration-change model refreshes remain intact.

The shutdown workaround was introduced in ced0242 — chore(release): prepare 0.2.0, alongside the original model-visibility fixes. Its comment intended the refresh to make the host re-query an inactive provider returning an empty list and remove stale picker entries. PR #104 (788b1d7) later moved the deactivation hook into src/runtime/lifecycle.ts while retaining that behavior.

Local VS Code source review from 1.116 onward confirms that extension-host RPC is disposed before deactivation, so this shutdown refresh cannot achieve its intended effect. The original model-visibility fixes remain useful. The separate startup refresh added in PR #13 — fix: thinking effort dropdown missing on first launch (27deac1) also remains intact.

Validation:

• npm run lint, npm run format:check, npm run compile, VSIX packaging, and git diff --check passed.
• Computer Use testing in existing Stable 1.136.2 and Insiders 1.138.0-insider environments reproduced the warning with a VSIX built from main.
• With the fixed VSIX, window reload and extension-host restart produced no new deactivation-refresh warning. DeepSeek models and all four thinking-effort choices remained available after reload.
• Disabling the extension and restarting extensions removed DeepSeek from the selectable model list; re-enabling restored the models in both builds.
• Insiders retained the previously selected model label in the input while disabled, although its picker contained no DeepSeek entries. That label behavior was not separately compared against the original build. No inference requests were sent in these lifecycle tests.

Final checks

• I reviewed and can explain every change
• I reviewed any user-facing text changes in both English and Chinese
• This pull request contains no API keys, authorization headers, private prompts, private code, or unreviewed sensitive logs

[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 2026-09-11T06:58:22.970717Z	98d893b	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.

[Copilot]

🟢 Approval recommended

The focused changes resolve the shutdown warning while preserving active model-refresh paths.

Pull request overview

Removes an ineffective shutdown-time model refresh that caused cancellation warnings.

Changes:

• Simplifies deactivation to local logger cleanup.
• Removes obsolete provider state and refresh logic.
• Deletes unused deactivation error translations.

File summaries

File	Description
src/runtime/lifecycle.ts	Simplifies shutdown cleanup.
src/provider/index.ts	Removes deactivation refresh machinery.
src/i18n.ts	Removes unused translations.

Review details

• Files reviewed: 3/3 changed files
• Comments generated: 0
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.