agentjido / req_llm / 73d05073c910ee310af5714438784f28bdea4de6

49%

Build Type

push

github

Committed by web-flow

Commit Message

Refactor: use ReqLLM.ToolCall directly, remove normalization layer (#105)

* chore: clean up whitespace in stream_server.ex

Remove trailing whitespace from documentation and code comments.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* refactor: use ReqLLM.ToolCall directly, remove normalization

Remove the normalization layer that converted tool calls to plain maps.
Since `ReqLLM.ToolCall` (based on `Req.ToolCall`) already exists as a
proper struct representation, there's no reason to introduce another
intermediate representation.

Changes:
- Remove `normalize_tool_call/1` from ReqLLM.Response
- Remove fallback logic for ContentPart-based tool calls (`:tool_result`
  content part type no longer exists in ContentPart)
- Update all tests to expect `ReqLLM.ToolCall` structs with
  `%{function: %{name: ..., arguments: ...}, id: ...}` structure
- Remove redundant pattern match in XAI provider

This simplifies the API and aligns with existing Req patterns.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* Enhance ReqLLM.ToolCall functionality and simplify tool call extraction

- Added new functions to `ReqLLM.ToolCall` for extracting function names, JSON arguments, and decoding arguments into maps.
- Implemented `find_args/2` to streamline the process of locating and decoding arguments from tool calls.
- Refactored `extract_from_tool_calls` in the provider modules to utilize the new `find_args/2` function, improving code clarity and reducing redundancy.
- Introduced comprehensive tests for the new `ToolCall` functionalities, ensuring robust validation of argument extraction and decoding.

This update enhances the usability of tool calls within the ReqLLM library and aligns with the existing API structure.

---------

Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Mike Hostetler <84222+mikehostetler@users.noreply.github.com>

Run Details

11 of 15 new or added lines in 6 files covered. (73.33%)

2 existing lines in 1 file now uncovered.

3283 of 6514 relevant lines covered (50.4%)

204.94 hits per line