test(backends): add unit tests for OpenCode stream event handling

[aaight]

Summary

• Adds tests/unit/backends/opencode-stream.test.ts with 53 unit tests covering all stream event helper functions in src/backends/opencode/stream.ts
• Tests cover appendPartialOutput, getPartialOutput, reportToolPart, handleSessionTerminalEvent, handleMessagePartUpdated, and handlePermissionEvent
• Follows the existing mock patterns from opencode.test.ts and opencode-permissions.test.ts

What was tested

• appendPartialOutput / getPartialOutput: text trimming, empty/whitespace rejection, chunk accumulation and joining
• reportToolPart: deduplication via reportedToolCalls Set, pending-state skip, distinct call tracking
• handleSessionTerminalEvent: session.error (with/without sessionID, error message extraction, fallback to name), session.idle (sessionID match/mismatch), session.status idle detection
• handleMessagePartUpdated: session-ID mismatch guard, step-start iteration counting, step-finish cost accumulation and LLM call logging, tool-call deduplication, text delta streaming, non-streaming text deduplication, synthetic/ignored part skipping, logWriter truncation at 300 chars
• handlePermissionEvent: session-ID mismatch returns false, allow/deny decision routing through resolvePermissionDecision + normalizePermissionDecision, SDK client called with correct path/body shape

Technical notes

• Mocked @opencode-ai/sdk/client, node:fs, runsRepository, nativeToolRetry, and logging at the module boundary — no full engine orchestration
• nativeToolRetry is mocked as a pass-through to avoid timer complexity
• clearMocks: true in global vitest config requires a beforeEach resetting mockStoreLlmCall to mockResolvedValue(undefined) in the step-finish suite so logLlmCall's .catch() does not throw

Trello card

https://trello.com/c/69c3e0fc5b8788ac3c9fe573

🤖 Generated with Claude Code

🕵️ claude-code · claude-sonnet-4-6 · run details

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[nhopeatall]

Summary

LGTM — Well-structured unit tests that accurately cover the exported stream event handling helpers.

I traced each test suite against the source implementation in src/backends/opencode/stream.ts and verified:

• Mock strategy is correct: Mocks are applied at the leaf dependency boundary (runsRepository, node:fs, logging, nativeToolRetry) rather than at intermediate modules, which means the real logLlmCall and appendEngineLog logic is exercised through their actual code paths.
• Session ID matching logic is properly tested: The casing asymmetry (sessionID in events vs sessionId in state) matches the source and is consistently used across all test suites.
• Text deduplication logic is accurately verified: The tests correctly distinguish between streaming (delta-based, no dedup) and non-streaming (part.id-based dedup) paths through getPartDelta + the seenTextPartIds guard.
• Permission normalization chain is tested end-to-end: resolvePermissionDecision → normalizePermissionDecision mapping ('allow' → 'always', 'deny' → 'reject') is verified for each permission type.
• Fire-and-forget logLlmCall is handled correctly: The mockStoreLlmCall.mockResolvedValue(undefined) reset in beforeEach prevents the .catch() chain in logLlmCall from throwing after clearMocks resets return values — the PR description explains this well.
• retryNativeToolOperation pass-through mock is an appropriate simplification that avoids timer complexity while still exercising the actual permission response path.

53 tests covering all 6 exported functions with good edge case coverage (whitespace, undefined state, synthetic/ignored parts, session mismatches, deduplication). All CI checks pass.

🕵️ claude-code · claude-opus-4-6 · run details