mirror of
https://github.com/QwenLM/qwen-code.git
synced 2026-11-01 12:12:24 +00:00
b9840886b8
Tool timing frames in paged transcript replay carried only a duration: the scheduler logs a batch's tool calls in one loop after every call in it has settled, so a record's event timestamp is the batch's end and no honest per-call start could be derived from it. The start was always measured, though; it just was not kept. Record it as an optional `started_at_ms` on the tool_call event, from the scheduler's terminal states (which now keep `startTime`) and from both places the ACP session logs a tool call, and pass it through unchanged: replay emits it as the tool frame's `startedAt`, the SDK timing reader keeps it, and the trajectory projection puts it on the tool row. Nothing derives it. Records written before this carry none and their frames stay start-less, as before. The trajectory fixture's two tool starts came from a pre-merge build that did derive them, and are removed so the fixture matches what a released daemon sends. |
||
|---|---|---|
| .. | ||
| _meta.ts | ||
| deployment.md | ||
| integration-tests.md | ||
| issue-and-pr-automation.md | ||
| npm.md | ||
| telemetry.md | ||