fix(avatar): 仅在 uniapp 容器展示充值入口

Reviewed-on: #22

digital-avatar-app/backend/requirements.txt

@@ -5,6 +5,7 @@ pydantic
 python-multipart
 httpx
 pypdf
+PyMuPDF>=1.24,<2
 python-docx
 openpyxl
 apscheduler>=3.10

digital-avatar-app/backend/services/knowledge_vectorizer.py

@@ -8,7 +8,8 @@ import threading
 from datetime import datetime, timezone
 
 from database import SessionLocal
-from models import KnowledgeChunk, KnowledgeDoc
+from models import Avatar, KnowledgeChunk, KnowledgeDoc
+from services.pdf_ocr_service import extract_scanned_pdf_text
 import embeddings
 
 logger = logging.getLogger(__name__)
@@ -76,7 +77,23 @@ class KnowledgeVectorizer:
 
             self._set_progress(db, doc, "extracting", 8)
             text = embeddings.extract_text(path, f".{doc.file_type}")
-            self._set_progress(db, doc, "chunking", 22)
+            if doc.file_type == "pdf" and not text.strip():
+                avatar = db.get(Avatar, doc.avatar_id)
+                if not avatar:
+                    raise ValueError("文档所属分身不存在")
+
+                def ocr_progress(done: int, total: int):
+                    percent = 8 + int((done / max(1, total)) * 20)
+                    self._set_progress(db, doc, "ocr", min(percent, 28))
+
+                self._set_progress(db, doc, "ocr", 8)
+                text = extract_scanned_pdf_text(
+                    db,
+                    avatar,
+                    path,
+                    on_progress=ocr_progress,
+                )
+            self._set_progress(db, doc, "chunking", 29)
             chunks = embeddings.chunk_text(text)
             if not chunks:
                 raise ValueError("文档没有可建立索引的文字内容")

digital-avatar-app/backend/services/pdf_ocr_service.py

@@ -0,0 +1,130 @@
+"""OCR fallback for image-only PDF knowledge documents."""
+
+import logging
+import os
+import time
+from typing import Callable
+
+from sqlalchemy.orm import Session
+
+from models import Avatar
+from services.chat_model_config import get_chat_model_config
+from services.token_billing import (
+    estimate_fallback_usage,
+    release_reservation,
+    reserve_avatar_tokens,
+    settle_reservation,
+)
+from services.vision_service import call_vision_model, prepare_image
+
+
+logger = logging.getLogger(__name__)
+
+PDF_OCR_PROMPT = (
+    "请逐字转录这一页扫描文档中的全部可见文字和表格，只输出转录内容，不要解释，不要使用 Markdown 代码块。"
+    "保留标题、段落、项目编号、数值和自然换行；看不清的内容写作[无法辨认]，不要猜测、纠错或补全。"
+)
+
+
+def _positive_int(name: str, default: int, minimum: int, maximum: int) -> int:
+    try:
+        value = int(os.getenv(name, str(default)))
+    except ValueError:
+        value = default
+    return max(minimum, min(maximum, value))
+
+
+def extract_scanned_pdf_text(
+    db: Session,
+    avatar: Avatar,
+    path: str,
+    *,
+    on_progress: Callable[[int, int], None] | None = None,
+) -> str:
+    """Render and OCR an image-only PDF while preserving page order."""
+    try:
+        import pymupdf
+    except ImportError as exc:
+        raise RuntimeError("扫描型 PDF 识别组件未安装") from exc
+
+    max_pages = _positive_int("KNOWLEDGE_PDF_OCR_MAX_PAGES", 80, 1, 300)
+    render_dpi = _positive_int("KNOWLEDGE_PDF_OCR_DPI", 144, 96, 200)
+    max_attempts = _positive_int("KNOWLEDGE_PDF_OCR_ATTEMPTS", 3, 1, 5)
+    model_config = get_chat_model_config()
+    model = model_config.ocr_model or model_config.vision_model
+    if not model_config.api_key or not model:
+        raise RuntimeError("扫描型 PDF 需要配置视觉 OCR 模型")
+
+    texts: list[str] = []
+    with pymupdf.open(path) as document:
+        total_pages = document.page_count
+        if total_pages <= 0:
+            raise ValueError("PDF 没有可识别页面")
+        if total_pages > max_pages:
+            raise ValueError(
+                f"扫描型 PDF 共 {total_pages} 页，超过单次 OCR 上限 {max_pages} 页，请拆分后上传"
+            )
+
+        scale = render_dpi / 72
+        for page_index in range(total_pages):
+            page = document.load_page(page_index)
+            pixmap = page.get_pixmap(
+                matrix=pymupdf.Matrix(scale, scale),
+                colorspace=pymupdf.csRGB,
+                alpha=False,
+            )
+            prepared = prepare_image(pixmap.tobytes("jpeg", jpg_quality=88))
+            estimate_messages = [{
+                "role": "user",
+                "content": f"[扫描 PDF 第 {page_index + 1}/{total_pages} 页]\n{PDF_OCR_PROMPT}",
+            }]
+            reservation = reserve_avatar_tokens(
+                db,
+                avatar,
+                "knowledge_pdf_ocr",
+                model,
+                estimate_messages,
+                model_config.vision_max_tokens,
+            )
+            try:
+                result = None
+                for attempt in range(1, max_attempts + 1):
+                    try:
+                        result = call_vision_model(
+                            prepared,
+                            model_config,
+                            model=model,
+                            prompt=PDF_OCR_PROMPT,
+                            json_output=False,
+                        )
+                        break
+                    except RuntimeError:
+                        if attempt == max_attempts:
+                            raise
+                        time.sleep(min(4, attempt))
+                content = str((result or {}).get("content") or "").strip()
+                if not content:
+                    raise RuntimeError("扫描型 PDF 页面识别结果为空")
+                settle_reservation(
+                    db,
+                    reservation,
+                    (result or {}).get("usage"),
+                    fallback_total=estimate_fallback_usage(estimate_messages, content),
+                )
+            except Exception as exc:
+                release_reservation(db, reservation, str(exc))
+                raise RuntimeError(
+                    f"扫描型 PDF 第 {page_index + 1}/{total_pages} 页识别失败：{exc}"
+                ) from exc
+
+            texts.append(f"[第 {page_index + 1} 页]\n{content}")
+            if on_progress:
+                on_progress(page_index + 1, total_pages)
+            logger.info(
+                "Scanned PDF OCR completed avatar=%s page=%s/%s",
+                avatar.id,
+                page_index + 1,
+                total_pages,
+            )
+
+    return "\n\n".join(texts).strip()

digital-avatar-app/backend/tests/test_knowledge_storage.py

@@ -238,6 +238,53 @@ def test_background_vectorizer_keeps_failure_reason_for_retry(
         db.close()
 
 
+def test_background_vectorizer_uses_ocr_for_image_only_pdf(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue"),
+    ):
+        response = client.post(
+            f"/api/avatar/{context['avatar'].id}/knowledge/docs",
+            headers=context["owner_headers"],
+            files={"file": ("scanned.pdf", b"image-only-pdf", "application/pdf")},
+        )
+
+    payload = response.json()["data"]
+    progress = []
+    with (
+        patch("services.knowledge_vectorizer.UPLOAD_DIR", str(tmp_path)),
+        patch("services.knowledge_vectorizer.embeddings.extract_text", return_value=""),
+        patch(
+            "services.knowledge_vectorizer.extract_scanned_pdf_text",
+            side_effect=lambda _db, _avatar, _path, on_progress: (
+                on_progress(1, 2), on_progress(2, 2), "扫描页文字"
+            )[-1],
+        ) as ocr,
+        patch("services.knowledge_vectorizer.embeddings.embed", return_value=[[1.0, 0.0]]),
+        patch.object(knowledge_vectorizer, "_set_progress", wraps=knowledge_vectorizer._set_progress) as set_progress,
+    ):
+        knowledge_vectorizer.vectorize_document(payload["id"])
+        progress = [(call.args[2], call.args[3]) for call in set_progress.call_args_list]
+
+    db = SessionLocal()
+    try:
+        stored = db.query(KnowledgeDoc).filter(KnowledgeDoc.id == payload["id"]).one()
+        assert stored.status == "ready"
+        assert stored.chunk_count == 1
+        assert ("ocr", 18) in progress
+        assert ("ocr", 28) in progress
+        ocr.assert_called_once()
+        db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == stored.id).delete()
+        db.delete(stored)
+        db.commit()
+    finally:
+        db.close()
+
+
 def test_retry_queues_a_failed_document_again(
     tmp_path: Path,
     authorization_context,

digital-avatar-app/backend/tests/test_pdf_ocr_service.py

@@ -0,0 +1,104 @@
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from services.pdf_ocr_service import extract_scanned_pdf_text
+
+
+class FakePixmap:
+    def tobytes(self, *_args, **_kwargs):
+        return b"jpeg-page"
+
+
+class FakePage:
+    def get_pixmap(self, **_kwargs):
+        return FakePixmap()
+
+
+class FakeDocument:
+    page_count = 2
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *_args):
+        return None
+
+    def load_page(self, _index):
+        return FakePage()
+
+
+def test_scanned_pdf_ocr_preserves_page_order_and_reports_progress(monkeypatch):
+    fake_pymupdf = SimpleNamespace(
+        open=lambda _path: FakeDocument(),
+        Matrix=lambda x, y: (x, y),
+        csRGB="rgb",
+    )
+    monkeypatch.setitem(__import__("sys").modules, "pymupdf", fake_pymupdf)
+    progress = []
+    reservation = SimpleNamespace()
+    config = SimpleNamespace(
+        api_key="configured",
+        ocr_model="qwen-vl-ocr",
+        vision_model="vision",
+        vision_max_tokens=2048,
+    )
+
+    with (
+        patch("services.pdf_ocr_service.get_chat_model_config", return_value=config),
+        patch("services.pdf_ocr_service.prepare_image", return_value=SimpleNamespace()),
+        patch(
+            "services.pdf_ocr_service.call_vision_model",
+            side_effect=[
+                {"content": "第一页文字", "usage": {"total_tokens": 10}},
+                {"content": "第二页文字", "usage": {"total_tokens": 12}},
+            ],
+        ),
+        patch("services.pdf_ocr_service.reserve_avatar_tokens", return_value=reservation) as reserve,
+        patch("services.pdf_ocr_service.settle_reservation") as settle,
+    ):
+        text = extract_scanned_pdf_text(
+            MagicMock(),
+            SimpleNamespace(id="avatar-1"),
+            "/tmp/scanned.pdf",
+            on_progress=lambda done, total: progress.append((done, total)),
+        )
+
+    assert text == "[第 1 页]\n第一页文字\n\n[第 2 页]\n第二页文字"
+    assert progress == [(1, 2), (2, 2)]
+    assert reserve.call_count == 2
+    assert settle.call_count == 2
+
+
+def test_scanned_pdf_ocr_releases_tokens_after_retries_fail(monkeypatch):
+    fake_document = FakeDocument()
+    fake_document.page_count = 1
+    fake_pymupdf = SimpleNamespace(
+        open=lambda _path: fake_document,
+        Matrix=lambda x, y: (x, y),
+        csRGB="rgb",
+    )
+    monkeypatch.setitem(__import__("sys").modules, "pymupdf", fake_pymupdf)
+    monkeypatch.setenv("KNOWLEDGE_PDF_OCR_ATTEMPTS", "2")
+    reservation = SimpleNamespace()
+    config = SimpleNamespace(
+        api_key="configured",
+        ocr_model="qwen-vl-ocr",
+        vision_model="vision",
+        vision_max_tokens=2048,
+    )
+
+    with (
+        patch("services.pdf_ocr_service.get_chat_model_config", return_value=config),
+        patch("services.pdf_ocr_service.prepare_image", return_value=SimpleNamespace()),
+        patch("services.pdf_ocr_service.call_vision_model", side_effect=RuntimeError("timeout")) as call,
+        patch("services.pdf_ocr_service.reserve_avatar_tokens", return_value=reservation),
+        patch("services.pdf_ocr_service.release_reservation") as release,
+        patch("services.pdf_ocr_service.time.sleep"),
+    ):
+        with pytest.raises(RuntimeError, match="第 1/1 页识别失败"):
+            extract_scanned_pdf_text(MagicMock(), SimpleNamespace(id="avatar-1"), "/tmp/scanned.pdf")
+
+    assert call.call_count == 2
+    release.assert_called_once()

digital-avatar-app/src/router/index.ts

@@ -1,6 +1,7 @@
 import { createRouter, createWebHashHistory } from 'vue-router'
 import type { RouteRecordRaw } from 'vue-router'
 import { getAuthToken } from '@/api'
+import { isInUniWebView } from '@/utils/uniapp-bridge'
 
 const routes: RouteRecordRaw[] = [
   {
@@ -55,7 +56,7 @@ const routes: RouteRecordRaw[] = [
     path: '/token/charge',
     name: 'TokenCharge',
     component: () => import('@/views/TokenCharge.vue'),
-    meta: { title: '积分充值', requiresAuth: true }
+    meta: { title: '积分充值', requiresAuth: true, requiresUniWebView: true }
   },
   {
     path: '/avatar/card',
@@ -133,6 +134,10 @@ const router = createRouter({
 
 router.beforeEach((to, from, next) => {
   document.title = to.meta.title as string || '会会数字分身'
+  if (to.meta.requiresUniWebView && !isInUniWebView()) {
+    next({ path: '/avatar/manage' })
+    return
+  }
   const hasLocalSession = Boolean(localStorage.getItem('hh_app_token'))
   const hasInjectedSession = Boolean(getAuthToken())
   if (to.meta.requiresAuth && !hasLocalSession && !hasInjectedSession) {

digital-avatar-app/src/views/AvatarManage.vue

@@ -20,7 +20,7 @@
       </div>
     </section>
 
-    <!-- 积分余额条：暂时隐藏，保留完整实现便于后续恢复。 -->
+    <!-- 积分余额条：仅在 uni-app 原生壳内开放充值购买。 -->
     <section v-if="SHOW_POINTS_BALANCE_CARD" class="token-section">
       <div class="token-card">
         <div class="token-info">
@@ -28,7 +28,7 @@
           <span class="token-amount">{{ tokenBalance.toLocaleString() }}</span>
           <span class="token-used">累计使用 {{ tokenConsumed.toLocaleString() }}</span>
         </div>
-        <button class="recharge-btn" @click="goToRecharge">充值</button>
+        <button class="recharge-btn" @click="goToRecharge">充值购买</button>
       </div>
     </section>
 
@@ -89,14 +89,15 @@ import { useAvatarStore } from '@/store/avatar'
 import { useUserStore } from '@/store/user'
 import { createAvatarShareLink } from '@/api'
 import { isHuihuiEmbeddedMode } from '@/utils/embed-mode'
+import { isInUniWebView } from '@/utils/uniapp-bridge'
 
 const router = useRouter()
 const avatarStore = useAvatarStore()
 const userStore = useUserStore()
 const isEmbedded = isHuihuiEmbeddedMode()
 
-// 临时产品开关：余额卡片代码保留，后续改为 true 即可恢复展示。
+// 充值购买只在 uni-app 原生壳内提供，避免普通 H5 进入支付链路。
-const SHOW_POINTS_BALANCE_CARD = false
+const SHOW_POINTS_BALANCE_CARD = isInUniWebView()
 
 // 当前登录会会用户的资料（头像 / 昵称）
 const me = computed(() => userStore.user)

digital-avatar-app/src/views/KnowledgeManage.vue

@@ -147,7 +147,7 @@ const documentState = (doc: any) => {
   if (['uploaded', 'parsing'].includes(String(doc.status || '').toLowerCase())) {
     const stage = String(doc.indexStage || 'queued').toLowerCase()
     const labels: Record<string, string> = {
-      queued: '等待处理', extracting: '解析文档', chunking: '切分文本', embedding: '向量化中'
+      queued: '等待处理', extracting: '解析文档', ocr: '扫描件识别', chunking: '切分文本', embedding: '向量化中'
     }
     const progress = Math.max(0, Math.min(99, Number(doc.indexProgress || 0)))
     return { tone: 'pending', label: labels[stage] || '处理中', detail: `${labels[stage] || '正在建立知识索引'} ${progress}%`, progress }