perf(server): stop retaining unused OpenCode tool history

[t3dotgg]

OpenCode keeps every version of tool output in a session history array that nothing reads. The audit fixture retained 5,171,200 logical output bytes for a tool whose final output was 102,400 bytes.

Remove that array and its append helpers. Thread reads and rollback still fetch upstream messages. Keep the live part map and event delivery unchanged.

Verified with 116 existing adapter and provider tests, server typecheck, and targeted lint. No browser, device, or live state was used.

Part of #9661. Other session-map retention remains separate work.

Created with GPT-6 Astra (preview) in Codex.

---

Note

Low Risk
Deletion-only memory optimization with no callers of the removed snapshot path; thread read/rollback still use upstream APIs.

Overview
Removes the in-memory turns array on OpenCode session context and the resolveTurnSnapshot / appendTurnItem helpers that pushed every tool Part update into it during message.part.updated handling.

That history was never read in the adapter—only grown—so long tool streams could retain huge redundant output while live behavior stays on partById, text delta maps, and runtime event emission.

readThread and rollbackThread are unchanged: they still load conversation shape from session.messages / session.revert, not from session context.

^{Reviewed by Cursor Bugbot for commit ecc5669. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Remove unused OpenCode turn-snapshot retention from OpenCodeAdapter

• Removes the turn-snapshot collection from OpenCodeSessionContext and the resolveTurnSnapshot and appendTurnItem helpers, as this history was retained but never read.
• The makeOpenCodeAdapter session-context initialization no longer allocates snapshot storage, and the event-part handler stops recording processed parts to per-turn snapshots before emitting the runtime event.
• Behavioral Change: OpenCode session contexts no longer hold in-memory turn snapshots; runtime event emission is unchanged.

^{Macroscope summarized ecc5669.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.6 KiB	13.8 KiB	+194 B (+1.4%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+5 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.7 KiB	+189 B (+2.8%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.1 KiB	58.5 KiB	+1.4 KiB (+2.5%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.8 KiB	13.7 KiB	−27 B (−0.2%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.7 KiB	−27 B (−0.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	59.3 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: 7cf5b28 · PR result: ecc5669 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Approved at ecc5669

Macroscope's review found this PR approvable — This single-file deletion removes unused OpenCode tool-history retention while preserving live part tracking, event emission, thread reads, and rollback behavior. The resulting runtime impact is limited to lower memory usage with no schema, deployment, security, or product-default changes.

^{You can add or adjust custom eligibility rules. Learn more.}

[francoisrob]

how is your experience with gpt6 @t3dotgg ?