language_models: Fix Mistral tool use erroring out (#54058)

Smit Barmase created 2 weeks ago

Closes #52717

Mistral's streaming API sometimes sends `"id": "null"` (the literal
string, not JSON null) in continuation chunks for tool calls. Our stream
handler treated this as a valid ID, overwriting the real tool call ID
from the first chunk. When the corrupted ID was sent back in the next
request, Mistral's API rejected it with "Tool call id has to be defined
in serving mode."

Self-Review Checklist:

- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content is consistent with the [UI/UX
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable

Release Notes:

- Fixed issue where Mistral models erroring out with "Tool call id has
to be defined" when using tools.

Change summary

crates/language_models/src/provider/mistral.rs | 64 ++++++++++++++++++++
1 file changed, 64 insertions(+)

Detailed changes

crates/language_models/src/provider/mistral.rs

@@ -652,6 +652,7 @@ impl MistralEventMapper {
 
                 if let Some(tool_id) = tool_call.id.clone()
                     && !tool_id.is_empty()
+                    && tool_id != "null"
                 {
                     entry.id = tool_id;
                 }
@@ -905,6 +906,69 @@ mod tests {
     use super::*;
     use language_model::{LanguageModelImage, LanguageModelRequestMessage, MessageContent};
 
+    fn tool_call_chunk(
+        id: Option<&str>,
+        name: Option<&str>,
+        arguments: Option<&str>,
+        finish_reason: Option<&str>,
+    ) -> mistral::StreamResponse {
+        mistral::StreamResponse {
+            id: "resp".into(),
+            object: "chat.completion.chunk".into(),
+            created: 0,
+            model: "test".into(),
+            choices: vec![mistral::StreamChoice {
+                index: 0,
+                delta: mistral::StreamDelta {
+                    role: None,
+                    content: None,
+                    tool_calls: if finish_reason.is_some() {
+                        None
+                    } else {
+                        Some(vec![mistral::ToolCallChunk {
+                            index: 0,
+                            id: id.map(Into::into),
+                            function: Some(mistral::FunctionChunk {
+                                name: name.map(Into::into),
+                                arguments: arguments.map(Into::into),
+                            }),
+                        }])
+                    },
+                },
+                finish_reason: finish_reason.map(Into::into),
+            }],
+            usage: None,
+        }
+    }
+
+    #[test]
+    fn test_streaming_tool_call_ignores_null_id() {
+        // Mistral's streaming API sometimes sends `"id": "null"` in continuation chunks.
+        let mut mapper = MistralEventMapper::new();
+
+        mapper.map_event(tool_call_chunk(
+            Some("real_id_123"),
+            Some("read_file"),
+            Some("{\"path\":"),
+            None,
+        ));
+        mapper.map_event(tool_call_chunk(
+            Some("null"),
+            None,
+            Some("\"a.txt\"}"),
+            None,
+        ));
+        let events = mapper.map_event(tool_call_chunk(None, None, None, Some("tool_calls")));
+
+        let Ok(LanguageModelCompletionEvent::ToolUse(tool_use)) = &events[0] else {
+            panic!("Expected first event to be ToolUse, got: {:?}", events[0]);
+        };
+
+        assert_eq!(tool_use.id.to_string(), "real_id_123");
+        assert_eq!(tool_use.name.as_ref(), "read_file");
+        assert_eq!(tool_use.input, serde_json::json!({"path": "a.txt"}));
+    }
+
     #[test]
     fn test_into_mistral_basic_conversion() {
         let request = LanguageModelRequest {