fix(chat): restore image generation blocks (#1402)

* fix(chat): restore image generation blocks

* test(chat): scope image accumulator test

Author: zerob13

src/main/presenter/deepchatAgentPresenter/accumulator.ts

@@ -119,6 +119,21 @@ export function accumulate(state: StreamState, event: LLMCoreStreamEvent): void
       }
       break
     }
+    case 'image_data': {
+      if (state.firstTokenTime === null) state.firstTokenTime = Date.now()
+      const block: AssistantMessageBlock = {
+        type: 'image',
+        status: 'pending',
+        timestamp: Date.now(),
+        image_data: {
+          data: event.image_data.data,
+          mimeType: event.image_data.mimeType
+        }
+      }
+      state.blocks.push(block)
+      state.dirty = true
+      break
+    }
     case 'usage': {
       state.metadata.inputTokens = event.usage.prompt_tokens
       state.metadata.outputTokens = event.usage.completion_tokens

src/main/presenter/llmProviderPresenter/providers/openAICompatibleProvider.ts

@@ -66,6 +66,19 @@ const SUPPORTED_IMAGE_SIZES = {
 // Add list of models with configurable sizes
 const SIZE_CONFIGURABLE_MODELS = ['gpt-image-1', 'gpt-4o-image', 'gpt-4o-all']
 
+export function normalizeExtractedImageText(content: string): string {
+  const normalized = content
+    .replace(/\r\n/g, '\n')
+    .replace(/\n\s*\n/g, '\n')
+    .trim()
+  if (!normalized) {
+    return ''
+  }
+
+  const semanticText = normalized.replace(/[\`*_~!\[\]\(\)]/g, '').trim()
+  return semanticText.length > 0 ? normalized : ''
+}
+
 function getOpenAIChatCachedTokens(usage: unknown): number | undefined {
   if (!usage || typeof usage !== 'object') {
     return undefined
@@ -1294,7 +1307,7 @@ export class OpenAICompatibleProvider extends BaseLLMProvider {
           // 如果处理了图片，清理多余的空行并记录日志
           if (hasImages) {
             // 清理移除图片后可能留下的多余空行
-            processedCurrentContent = processedCurrentContent.replace(/\n\s*\n/g, '\n').trim()
+            processedCurrentContent = normalizeExtractedImageText(processedCurrentContent)
             console.log(
               `[handleChatCompletion] Processed ${currentContent.length} chars -> ${processedCurrentContent.length} chars (images removed)`
             )

src/shared/types/agent-interface.d.ts

@@ -210,6 +210,7 @@ export type AssistantBlockType =
   | 'error'
   | 'tool_call'
   | 'action'
+  | 'image'
 
 export interface ToolCallBlockData {
   id?: string
@@ -263,6 +264,10 @@ export interface AssistantMessageBlock {
         start: number
         end: number
       }
+  image_data?: {
+    data: string
+    mimeType: string
+  }
   tool_call?: ToolCallBlockData
   extra?: AssistantMessageExtra
   action_type?: 'tool_call_permission' | 'question_request'

test/main/presenter/deepchatAgentPresenter/accumulator.test.ts

@@ -235,4 +235,26 @@ describe('accumulate', () => {
     accumulate(state, { type: 'permission', permission: {} } as any)
     expect(state.blocks.length).toBe(blocksBefore)
   })
+
+  it('creates image blocks for image_data events without empty text blocks', () => {
+    accumulate(state, {
+      type: 'image_data',
+      image_data: {
+        data: 'imgcache://generated/test.png',
+        mimeType: 'deepchat/image-url'
+      }
+    })
+
+    expect(state.blocks).toHaveLength(1)
+    expect(state.blocks[0]).toMatchObject({
+      type: 'image',
+      status: 'pending',
+      image_data: {
+        data: 'imgcache://generated/test.png',
+        mimeType: 'deepchat/image-url'
+      }
+    })
+    expect(state.blocks.some((block) => block.type === 'content')).toBe(false)
+    expect(state.dirty).toBe(true)
+  })
 })

test/main/presenter/llmProviderPresenter/openAICompatibleProvider.test.ts

@@ -7,7 +7,10 @@ import type {
   MCPToolDefinition,
   ModelConfig
 } from '../../../../src/shared/presenter'
-import { OpenAICompatibleProvider } from '../../../../src/main/presenter/llmProviderPresenter/providers/openAICompatibleProvider'
+import {
+  OpenAICompatibleProvider,
+  normalizeExtractedImageText
+} from '../../../../src/main/presenter/llmProviderPresenter/providers/openAICompatibleProvider'
 import { OpenRouterProvider } from '../../../../src/main/presenter/llmProviderPresenter/providers/openRouterProvider'
 import { LLMProviderPresenter } from '../../../../src/main/presenter/llmProviderPresenter'
 
@@ -335,3 +338,16 @@ describe('OpenAICompatibleProvider MCP runtime injection', () => {
     expect(requestParams.tools).toEqual(convertedTools)
   })
 })
+
+describe('normalizeExtractedImageText', () => {
+  it('keeps meaningful text after image markdown cleanup', () => {
+    expect(normalizeExtractedImageText('  Here is the updated image.\n\n')).toBe(
+      'Here is the updated image.'
+    )
+  })
+
+  it('drops markdown residue after image markdown cleanup', () => {
+    expect(normalizeExtractedImageText('`\n')).toBe('')
+    expect(normalizeExtractedImageText('[]()')).toBe('')
+  })
+})