fix(qoder): unique completion id per request

changelog/unreleased/qoder-unique-completion-id.md

@@ -0,0 +1,7 @@
+### English
+
+- Qoder streaming responses now carry a unique completion `id` per request, shared by every chunk of that request, so clients no longer merge replies from different requests.
+
+### 中文
+
+- Qoder 流式响应现在每个请求使用唯一的 completion `id`，同一请求的所有分片保持一致，客户端不再把不同请求的回复合并在一起。

internal/providers/qoder/completion_id_test.go

@@ -0,0 +1,62 @@
+package qoder
+
+import (
+	"bytes"
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+func completionIDs(t *testing.T, stream string) []string {
+	t.Helper()
+	var output bytes.Buffer
+	if err := writeOpenAIChunks(&output, strings.NewReader(stream), "synthetic-model"); err != nil {
+		t.Fatal(err)
+	}
+	var ids []string
+	for _, line := range strings.Split(output.String(), "\n") {
+		data, ok := strings.CutPrefix(line, "data: ")
+		if !ok || data == "[DONE]" {
+			continue
+		}
+		var chunk struct {
+			ID string `json:"id"`
+		}
+		if err := json.Unmarshal([]byte(data), &chunk); err != nil {
+			t.Fatalf("decode %q: %v", data, err)
+		}
+		ids = append(ids, chunk.ID)
+	}
+	return ids
+}
+
+// Regression for #292: every request gets its own completion id, shared by
+// all of its chunks including the synthesized terminal and usage chunks.
+func TestWriteOpenAIChunksUsesOneUniqueIDPerRequest(t *testing.T) {
+	cases := map[string]string{
+		// No upstream finish: a terminal chunk is synthesized.
+		"synthesized_finish": "data: " + `{"choices":[{"delta":{"content":"hi"}}]}` + "\n\n",
+		// Finish without usage, then usage: a trailing usage chunk is synthesized.
+		"synthesized_usage": "data: " + `{"choices":[{"delta":{"content":"hi"}}],"usage":{"prompt_tokens":1,"completion_tokens":1,"total_tokens":2}}` + "\n\n" +
+			"data: " + `{"choices":[{"delta":{"content":"!"},"finish_reason":"stop"}]}` + "\n\n",
+	}
+	for name, stream := range cases {
+		t.Run(name, func(t *testing.T) {
+			first := completionIDs(t, stream)
+			second := completionIDs(t, stream)
+			if len(first) < 2 {
+				t.Fatalf("expected synthesized chunk, got ids %v", first)
+			}
+			for _, ids := range [][]string{first, second} {
+				for _, id := range ids {
+					if id != ids[0] || !strings.HasPrefix(id, "chatcmpl-") || id == "chatcmpl-qoder" {
+						t.Fatalf("inconsistent or fixed ids %v", ids)
+					}
+				}
+			}
+			if first[0] == second[0] {
+				t.Fatalf("two requests share completion id %q", first[0])
+			}
+		})
+	}
+}

internal/providers/qoder/direct.go

@@ -819,7 +819,18 @@ func (b *convertedBody) Close() error {
 	return b.closeUpstream()
 }
 
+// newCompletionID returns one OpenAI completion id per request. Clients that
+// merge assistant chunks by id must not see two requests share one (#292).
+func newCompletionID() string {
+	id, err := randomUUID()
+	if err != nil {
+		return fmt.Sprintf("chatcmpl-qoder-%d", time.Now().UnixNano())
+	}
+	return "chatcmpl-" + strings.ReplaceAll(id, "-", "")
+}
+
 func writeOpenAIChunks(w io.Writer, src io.Reader, model string) error {
+	completionID := newCompletionID()
 	var sawFinish, sawTools, sawPayload, sawError bool
 	var lastUsage any
 	var lastChunkEmittedHadUsage bool
@@ -841,7 +852,7 @@ func writeOpenAIChunks(w io.Writer, src io.Reader, model string) error {
 			return err
 		}
 		chunk := map[string]any{
-			"id": "chatcmpl-qoder", "object": "chat.completion.chunk", "model": model,
+			"id": completionID, "object": "chat.completion.chunk", "model": model,
 			"choices": []any{map[string]any{"index": 0, "delta": deltaOf(frame), "finish_reason": emptyOrNil(frame.finish)}},
 		}
 		if len(frame.usage) > 0 {
@@ -881,7 +892,7 @@ func writeOpenAIChunks(w io.Writer, src io.Reader, model string) error {
 				finish = "tool_calls"
 			}
 			chunk := map[string]any{
-				"id": "chatcmpl-qoder", "object": "chat.completion.chunk", "model": model,
+				"id": completionID, "object": "chat.completion.chunk", "model": model,
 				"choices": []any{map[string]any{"index": 0, "delta": map[string]any{}, "finish_reason": finish}},
 			}
 			if lastUsage != nil {
@@ -906,7 +917,7 @@ func writeOpenAIChunks(w io.Writer, src io.Reader, model string) error {
 		}
 	} else if !lastChunkEmittedHadUsage && lastUsage != nil {
 		chunk := map[string]any{
-			"id": "chatcmpl-qoder", "object": "chat.completion.chunk", "model": model,
+			"id": completionID, "object": "chat.completion.chunk", "model": model,
 			"choices": []any{},
 			"usage":   lastUsage,
 		}