Harden Copilot completion context caching

[Sean McManus (sean-mcmanus)]

Summary

Ensures Copilot completion context cache entries are reused only when the caret distance and request options still match. It also invalidates shared cache state on C/C++ document and configuration changes and prevents in-flight computations from repopulating invalidated entries.

This PR was investigated and created by Copilot with GPT-5.6 Sol (in VS Code). Any message starting with ✨Copilot: was sent by Copilot.

Details

• Associate cached results with the feature, count, length, and aggregation policy used to compute them.
• Correct stale-distance evaluation and reject late results from an invalidated cache generation.
• Clear shared cache and parameter state when relevant documents or configurations change.
• Dispose registrations that complete after the provider has already been disposed.

Validation

• TypeScript compilation
• ESLint on the changed source and test files
• All 219 extension unit tests

[Copilot]

🟢 Approval recommended

The changes are focused, internally consistent, and cover the identified cache and disposal scenarios with unit tests.

Pull request overview

Hardens Copilot completion-context caching against stale entries and late asynchronous results.

Changes:

• Validates cache entries against caret distance and request policy.
• Invalidates shared cache state on document/configuration changes.
• Safely handles late provider registrations after disposal.

File summaries

File	Description
Extension/src/LanguageServer/copilotCompletionContextCache.ts	Adds generation-aware cache and disposable storage.
Extension/src/LanguageServer/copilotCompletionContextProvider.ts	Integrates validation, invalidation, and safe disposal.
Extension/src/LanguageServer/client.ts	Shares and clears the provider across clients.
Extension/test/unit/copilotCompletionContextCache.test.ts	Tests cache validation, invalidation, and disposal.

Review details

• Files reviewed: 4/4 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.