Repository navigation
Conversation
|
The following comment was made by an LLM, it may be inaccurate: |
![opencode-agent[bot]](https://avatars.githubusercontent.com/in/1549082?s=60&v=4){kind=small}
There was a problem hiding this comment.
Looks good. I reproduced #56281 on the base with the issue's script, using a simulated model: a reply plus a tool call, then a step with only reasoning. /copy showed "No text content found…". With this change it copies the reply. The new tests pass here, and 5 of the 8 fail on the base. The TUI typecheck passes.
Public API impact: none. This only changes how the TUI /copy command picks the message to copy.
I left two non-blocking comments: the error toast wording now that the selection only finds messages with text, and the size of the new test file.
Related: #55924
| (msg.type === "compaction" && msg.status === "completed" && msg.summary.trim() !== ""), | ||
| ) | ||
| if (!lastAssistantMessage) { | ||
| toast.show({ message: "No assistant messages found", variant: "error" }) |
There was a problem hiding this comment.
Now that the selection only finds messages with non-blank text, this toast also shows when the current turn has assistant messages but none with text (e.g. a reasoning-only turn), and "No assistant messages found" is misleading there. Also, the if (!text) "No text content found in last assistant message" branch below can no longer run. Could you merge the two into one accurate message (e.g. "No reply text found in the last turn")?
There was a problem hiding this comment.
Updated in be343af: the missing-selection toast is now "No reply text found in the last turn", and the unreachable empty-text branch is removed. The retained empty-turn case asserts that message and that no clipboard write occurs. Selection, revert filtering and compaction handling are unchanged.
There was a problem hiding this comment.
Thanks, I checked be343af and it handles both points. The toast now says "No reply text found in the last turn", the unreachable empty-text branch is gone, and the test is down to two cases (reasoning-only last message, empty current turn). The selection logic hasn't changed since the version I verified. Nothing more from me.
| @@ -0,0 +1,158 @@ | |||
| import { expect, spyOn, test } from "bun:test" | |||
There was a problem hiding this comment.
This is 158 lines of tests for a 12-line fix, and each of the 8 cases starts the whole TUI app. Could you keep only the core cases, e.g. a reasoning- or tool-only tail falls back to the earlier text, and an empty current turn does not copy the previous turn? Dropping the file would also be fine; the fix is small enough not to need it.
There was a problem hiding this comment.
Reduced in be343af from eight full-app cases to the two core cases you suggested: fallback past a reasoning-only tail and no copy from a previous turn. The file is now 83 lines, with two app fixtures instead of eight. The original broader matrix is retained in my local evidence only.
The five selected TUI modules now pass 13 tests, and the retained pair passes ten repetitions (20 passes). TUI typecheck, scoped lint and test formatting pass. No full repository suite or upstream CI pass is claimed.
There was a problem hiding this comment.
Thanks, the trimmed test looks good. Its two cases (falling back past a reasoning-only last message, and no copy from an earlier turn) cover the fix, and each starts the app only once. Nothing more from me on the tests.
Issue for this PR
Closes #56281
Type of change
What does this PR do?
/copyfails when an answer ends with an assistant message containing only reasoning, tools or blank text, even though its earlier message contains the reply.Select the last text-bearing assistant message after the latest user prompt, respecting the existing revert cutoff. Completed compaction summaries remain eligible. This follows @fantisst's answer-block diagnosis; no argument features or plugin code are included. Synthetic-input turn boundaries retain their existing behavior. A current turn without copyable text now reports that directly.
How did you verify your code works?
AI-assisted implementation and independent local review on native Windows/Bun 1.4.2 with frozen dependencies:
Tests use fixture history and inspect the OSC52 writer payload. No live model, system clipboard, full repository suite or upstream CI validation was run.
reviewed and tested locally.
Screenshots / recordings
Command selection logic only; the tests drive the production TUI and assert the copied text.
Checklist
Added by the opencode bot (AI-generated).
Related: #55924