fix(openai): make tool_call arguments optional and fix silent stream termination #6309
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Some LLM providers (e.g., vLLM with QwQ models) send tool_call chunks where the first chunk contains only the function name without arguments, and subsequent chunks contain the arguments. This caused deserialization to fail with 'missing field arguments'. The deserialization error was silently treated as end-of-stream because 'while let Some(Ok(...))' doesn't match 'Some(Err(...))' - it just exits the loop as if the stream had completed normally. Changes: - Added #[serde(default)] to DeltaToolCallFunction.arguments in openai.rs to handle missing arguments by defaulting to empty string - Fixed stream consumption pattern: changed 'while let Some(Ok(...))' to 'while let Some(result)' with explicit error propagation via '?' - Added test for stream error propagation Signed-off-by: rabi <[email protected]>
rabi
changed the title
ThanhNguyxn
pushed a commit
to ThanhNguyxn/goose
that referenced
this pull request
Jan 14, 2026
…termination (block#6309) Signed-off-by: rabi <[email protected]> Signed-off-by: ThanhNguyxn <[email protected]>
zanesq
added a commit
that referenced
this pull request
Jan 14, 2026
…ased * 'main' of github.com:block/goose: fix(code_execution): serialize record_result output as JSON (#6495) perf(google): avoid accumulating thoughtSignatures across conversation history (#6462) fix(openai): make tool_call arguments optional and fix silent stream termination (#6309) fix: Improve error messages for invalid tool calls (#6483) fix: require auth when running goose on non loopback address (#6478) chore(deps): bump hono from 4.11.3 to 4.11.4 in /ui/desktop (#6485) feat(cli): graceful fallback for keyring failures (#5808) fix: support global .gooseignore and negation patterns (#6157) docs: manual config for jetbrains (#6490) fix: Recipe slash command doesn't work with single optional parameter (#6235) fix(openrouter): Handle Gemini thoughtSignature for tool calls (#6370) docs: fix extensions page (#6484) Allow customizing the new line keybinding in the CLI (#5956) Ask for permission in the CLI (#6475) docs: add Ralph Loop tutorial for multi-model iterative development (#6455) Remove gitignore fallback from gooseignore docs (#6480) fix: clean up result recording for code mode (#6343) fix(code_execution): handle model quirks with tool calls (#6352) feat(ui): support prefersBorder option for MCP Apps (#6465) fixed line breaks (#6459)
lifeizhou-ap
added a commit
that referenced
this pull request
Jan 15, 2026
* main: fixed 0 token in openrouter steaming (#6493) feat(goose-acp): enable parallel sessions with isolated agent state (#6392) copilot instruction to flag prelease docs (#6504) docs: acp mcp support (#6491) feat: add flatpak support for linux (#6387) fix(code_execution): serialize record_result output as JSON (#6495) perf(google): avoid accumulating thoughtSignatures across conversation history (#6462) fix(openai): make tool_call arguments optional and fix silent stream termination (#6309) fix: Improve error messages for invalid tool calls (#6483)
zanesq
added a commit
that referenced
this pull request
Jan 16, 2026
* 'main' of github.com:block/goose: (28 commits) chore(deps): bump aiohttp from 3.13.0 to 3.13.3 in /scripts/provider-error-proxy (#6539) chore(deps): bump brotli from 1.1.0 to 1.2.0 in /scripts/provider-error-proxy (#6538) docs: temp correction for agent directory (#6544) chore: upgrade rmcp (#6516) docs: clarify directory in /documentation readme (#6541) Release 1.20.0 Standalone mcp apps (#6458) don't add escaping to the command field (#6519) Fix popular topics not starting chat when clicked (#6508) fix[desktop]: deeplink ui repeat on refresh (#6469) fixed test compilation on main branch (#6512) fix: correctly parse extension name from tool call for MCP apps (#6482) fixed 0 token in openrouter steaming (#6493) feat(goose-acp): enable parallel sessions with isolated agent state (#6392) copilot instruction to flag prelease docs (#6504) docs: acp mcp support (#6491) feat: add flatpak support for linux (#6387) fix(code_execution): serialize record_result output as JSON (#6495) perf(google): avoid accumulating thoughtSignatures across conversation history (#6462) fix(openai): make tool_call arguments optional and fix silent stream termination (#6309) ...
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Some LLM providers (e.g., vLLM with QwQ models) send tool_call chunks where the first chunk contains only the function name without arguments, and subsequent chunks contain the arguments. This caused deserialization to fail with 'missing field arguments'.
The deserialization error was silently treated as end-of-stream because 'while let Some(Ok(...))' doesn't match 'Some(Err(...))' - it just exits the loop as if the stream had completed normally.
Changes:
Type of Change
AI Assistance
Testing
Unit and manual testing with vLLM.