fix/openai: non-streaming LLM response (#64473)

Fixes CODY-3194

Previously, using `/chat/completions` with OpenAI models always returned
an empty completion because we were reading a non-existent `"text"`
property instead of the nested `"message": { "content": ...}` property.

This PR fixes the bug and adds a test case to demonstrate how we parse a
real-world OpenAI response.

<!-- PR description tips:
https://www.notion.so/sourcegraph/Write-a-good-pull-request-description-610a7fd3e613496eb76f450db5a49b6e
-->

## Test plan

See new test cases.

I'm failing to get a locally running setup to manually test this e2e,
see https://sourcegraph.slack.com/archives/C04MYFW01NV/p1723664513193939
<!-- REQUIRED; info at
https://docs-legacy.sourcegraph.com/dev/background-information/testing_principles
-->

## Changelog

* Fix bug where requests to `/.api/completions/stream` for OpenAI models
returned an empty completion when using `stream: false`.
<!-- OPTIONAL; info at
https://www.notion.so/sourcegraph/Writing-a-changelog-entry-dd997f411d524caabf0d8d38a24a878c
-->

Author: olafurpg

internal/completions/client/openai/openai.go

@@ -74,7 +74,7 @@ func (c *openAIChatCompletionStreamClient) Complete(
 		logger.Warn("Failed to count tokens with the token manager %w ", log.Error(err))
 	}
 	return &types.CompletionResponse{
-		Completion: response.Choices[0].Text,
+		Completion: response.Choices[0].Message.Content,
 		StopReason: response.Choices[0].FinishReason,
 	}, nil
 }
@@ -138,7 +138,7 @@ func (c *openAIChatCompletionStreamClient) Stream(
 
 		if len(event.Choices) > 0 {
 			if request.Feature == types.CompletionsFeatureCode {
-				content += event.Choices[0].Text
+				content += event.Choices[0].Message.Content
 			} else {
 				content += event.Choices[0].Delta.Content
 			}

internal/completions/client/openai/openai_test.go

@@ -25,32 +25,35 @@ func (c *mockDoer) Do(r *http.Request) (*http.Response, error) {
 	return c.do(r)
 }
 
-func TestErrStatusNotOK(t *testing.T) {
-	tokenManager := tokenusage.NewManager()
-	mockClient := NewClient(&mockDoer{
+var compRequest = types.CompletionRequest{
+	Feature: types.CompletionsFeatureChat,
+	Version: types.CompletionsVersionLegacy,
+	ModelConfigInfo: types.ModelConfigInfo{
+		Provider: modelconfigSDK.Provider{
+			ID: modelconfigSDK.ProviderID("xxx-provider-id-xxx"),
+		},
+		Model: modelconfigSDK.Model{
+			ModelRef: modelconfigSDK.ModelRef("provider::apiversion::test-model"),
+		},
+	},
+	Parameters: types.CompletionRequestParameters{
+		RequestedModel: "xxx-requested-model-xxx",
+	},
+}
+
+func NewMockClient(statusCode int, response string) types.CompletionsClient {
+	return NewClient(&mockDoer{
 		func(r *http.Request) (*http.Response, error) {
 			return &http.Response{
-				StatusCode: http.StatusTooManyRequests,
-				Body:       io.NopCloser(bytes.NewReader([]byte("oh no, please slow down!"))),
+				StatusCode: statusCode,
+				Body:       io.NopCloser(bytes.NewReader([]byte(response))),
 			}, nil
 		},
-	}, "", "", *tokenManager)
+	}, "", "", *tokenusage.NewManager())
+}
 
-	compRequest := types.CompletionRequest{
-		Feature: types.CompletionsFeatureChat,
-		Version: types.CompletionsVersionLegacy,
-		ModelConfigInfo: types.ModelConfigInfo{
-			Provider: modelconfigSDK.Provider{
-				ID: modelconfigSDK.ProviderID("xxx-provider-id-xxx"),
-			},
-			Model: modelconfigSDK.Model{
-				ModelRef: modelconfigSDK.ModelRef("provider::apiversion::test-model"),
-			},
-		},
-		Parameters: types.CompletionRequestParameters{
-			RequestedModel: "xxx-requested-model-xxx",
-		},
-	}
+func TestErrStatusNotOK(t *testing.T) {
+	mockClient := NewMockClient(http.StatusTooManyRequests, "oh no, please slow down!")
 
 	t.Run("Complete", func(t *testing.T) {
 		logger := log.Scoped("completions")
@@ -74,3 +77,36 @@ func TestErrStatusNotOK(t *testing.T) {
 		assert.True(t, ok)
 	})
 }
+
+func TestNonStreamingResponseParsing(t *testing.T) {
+	mockClient := NewMockClient(http.StatusOK, `{
+  "id": "chatcmpl-9wEJ9hnLdPcCLrfdZLrRPGOz48Pmo",
+  "object": "chat.completion",
+  "created": 1723665051,
+  "model": "gpt-4o-mini-2024-07-18",
+  "choices": [
+    {
+      "index": 0,
+      "message": {
+        "role": "assistant",
+        "content": "yes",
+        "refusal": null
+      },
+      "logprobs": null,
+      "finish_reason": "stop"
+    }
+  ],
+  "usage": {
+    "prompt_tokens": 15,
+    "completion_tokens": 1,
+    "total_tokens": 16
+  },
+  "system_fingerprint": "fp_48196bc67a"
+}`)
+	logger := log.Scoped("completions")
+	resp, err := mockClient.Complete(context.Background(), logger, compRequest)
+	require.NoError(t, err)
+	assert.NotNil(t, resp)
+	autogold.Expect(&types.CompletionResponse{Completion: "yes", StopReason: "stop"}).Equal(t, resp)
+
+}

internal/completions/client/openai/types.go

@@ -50,10 +50,14 @@ type openaiChoiceDelta struct {
 	Content string `json:"content"`
 }
 
+type openaiMessage struct {
+	Content string `json:"content"`
+}
+
 type openaiChoice struct {
 	Delta        openaiChoiceDelta `json:"delta"`
+	Message      openaiMessage     `json:"message"`
 	Role         string            `json:"role"`
-	Text         string            `json:"text"`
 	FinishReason string            `json:"finish_reason"`
 }