deepdoc-lib 0.2.2__tar.gz → 0.2.4__tar.gz

deepdoc_lib-0.2.4/.github/workflows/ci.yml

@@ -0,0 +1,92 @@
+name: Tests
+
+on:
+    push:
+    pull_request:
+    workflow_dispatch:
+
+jobs:
+    pytest:
+        name: Pytest (${{ matrix.python-version }})
+        runs-on: ubuntu-latest
+        strategy:
+            matrix:
+                python-version: ["3.10", "3.11", "3.12", "3.13"]
+
+        steps:
+            - uses: actions/checkout@v4
+
+            - name: Set up Python
+              uses: actions/setup-python@v5
+              with:
+                  python-version: ${{ matrix.python-version }}
+                  cache: "pip"
+
+            - name: Install dependencies
+              run: |
+                  python -m pip install --upgrade pip
+                  python -m pip install -e . pytest
+
+            - name: Resolve the built wheel with current PyJWT
+              run: |
+                  python -m pip install build
+                  python -m build --wheel
+                  python -m pip install --dry-run --ignore-installed dist/*.whl 'PyJWT[crypto]>=2.15.1,<3'
+
+            # Preprovision immutable official data instead of asking the NLTK
+            # downloader to traverse the runner's proxy. Do not disable pathsec.
+            - name: Check out pinned NLTK test resources
+              uses: actions/checkout@v4
+              with:
+                  repository: nltk/nltk_data
+                  ref: 550b6625bcef1f2abff2ff770a5a0d272c9c6b2a
+                  path: .ci-nltk-data
+                  persist-credentials: false
+                  sparse-checkout-cone-mode: false
+                  sparse-checkout: |
+                      packages/tokenizers/punkt.zip
+                      packages/tokenizers/punkt_tab.zip
+                      packages/corpora/wordnet.zip
+                      packages/taggers/averaged_perceptron_tagger.zip
+                      packages/taggers/averaged_perceptron_tagger_eng.zip
+
+            - name: Prepare and validate offline NLTK test resources
+              env:
+                  DEEPDOC_NLTK_DATA_DIR: ${{ runner.temp }}/deepdoc-nltk-data
+                  NLTK_DATA: ${{ runner.temp }}/deepdoc-nltk-data
+              run: |
+                  for package in tokenizers/punkt tokenizers/punkt_tab corpora/wordnet \
+                      taggers/averaged_perceptron_tagger taggers/averaged_perceptron_tagger_eng; do
+                      python -m zipfile -e ".ci-nltk-data/packages/$package.zip" "$NLTK_DATA/$(dirname "$package")"
+                  done
+                  python -c 'from deepdoc.depend.nltk_manager import ensure_nltk_data; ensure_nltk_data(offline=True)'
+
+            - name: Run tests
+              env:
+                  DEEPDOC_NLTK_DATA_DIR: ${{ runner.temp }}/deepdoc-nltk-data
+                  NLTK_DATA: ${{ runner.temp }}/deepdoc-nltk-data
+              run: python -m pytest tests
+
+    minimum-sdk:
+        name: Minimum SDK contract (3.10)
+        runs-on: ubuntu-latest
+        steps:
+            - uses: actions/checkout@v4
+
+            - name: Set up Python
+              uses: actions/setup-python@v5
+              with:
+                  python-version: "3.10"
+                  cache: "pip"
+
+            - name: Build and resolve the wheel with the minimum SDK and current PyJWT
+              run: |
+                  python -m pip install --upgrade pip build
+                  python -m build --wheel
+                  python -m pip install --dry-run --ignore-installed dist/*.whl 'zai-sdk==0.0.4' 'PyJWT[crypto]>=2.15.1,<3'
+
+            - name: Install and test the minimum SDK contract
+              run: |
+                  python -m pip install dist/*.whl 'zai-sdk==0.0.4' 'PyJWT[crypto]>=2.15.1,<3' pytest
+                  python -m pip check
+                  python -m pytest tests/test_zhipu_sdk_compatibility.py

{deepdoc_lib-0.2.2 → deepdoc_lib-0.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: deepdoc_lib
-Version: 0.2.2
+Version: 0.2.4
 Summary: [RAGFlow](https://ragflow.io/) is an open-source RAG (Retrieval-Augmented Generation) engine based on deep document understanding. It offers a streamlined RAG workflow for businesses of any scale, combining LLM (Large Language Models) to provide truthful question-answering capabilities, backed by well-founded citations from various complex formatted data.
 Author-email: Zhichang Yu <yuzhichang@gmail.com>
 Requires-Python: <3.14,>=3.10
@@ -42,7 +42,7 @@ Requires-Dist: tencentcloud-sdk-python>=3.0.1215
 Requires-Dist: tiktoken>=0.7.0
 Requires-Dist: xgboost<3.0.0,>=2.1.0
 Requires-Dist: xpinyin>=0.7.6
-Requires-Dist: zhipuai>=2.0.1
+Requires-Dist: zai-sdk>=0.0.4
 Requires-Dist: google-generativeai<0.9.0,>=0.8.1
 Requires-Dist: trio>=0.29.0
 Requires-Dist: setuptools<76.0.0,>=75.2.0

deepdoc_lib-0.2.4/UPSTREAM_SYNC.md

@@ -0,0 +1,86 @@
+# Upstream Sync Record
+
+This library is a standalone, pip-installable extraction of the `deepdoc/`
+module from [infiniflow/ragflow](https://github.com/infiniflow/ragflow),
+restructured around explicit configuration injection
+(`PdfModelConfig` / `TokenizerConfig`, `DEEPDOC_*` env vars) and
+ModelScope-based model distribution instead of ragflow's global settings and
+HuggingFace runtime downloads.
+
+Upstream changes are **ported by behavior, never by wholesale file copy**,
+so the library's public API (`from deepdoc import PdfParser, ...`) and
+packaging stay stable.
+
+## Sync history
+
+### 2026-07-23 — synced to ragflow `18ea0fb7` (2026-07-22)
+
+Ported fixes (upstream PR references):
+
+| Area | Fix | Upstream |
+|---|---|---|
+| vision/operators | `StandardizeImag` → `StandardizeImage` rename; the dynamic `getattr(operators, "StandardizeImage")` dispatch silently skipped image standardization before | [#16785](https://github.com/infiniflow/ragflow/pull/16785) |
+| mineru/tcadp | zip extraction hardening: reject symlink / encrypted / absolute-path / `..`-traversal / zip-slip entries; streaming download for TCADP results | [#12527](https://github.com/infiniflow/ragflow/pull/12527) |
+| html_parser | parse bodyless HTML fragments (`soup.body or soup`) | [#16423](https://github.com/infiniflow/ragflow/pull/16423) |
+| html_parser | switch `html5lib` → `html.parser` (aligns with upstream #14486-era behavior the later fixes assume) | [#14486](https://github.com/infiniflow/ragflow/pull/14486) |
+| html_parser | h4 heading mapping `#####` → `####` | [#13833](https://github.com/infiniflow/ragflow/pull/13833) |
+| html_parser | oversized-block splitting preserves original text (CJK-aware atom splitting) | [#16052](https://github.com/infiniflow/ragflow/pull/16052) |
+| excel_parser | keep `0` / `False` cells; skip fully-empty rows | [#16287](https://github.com/infiniflow/ragflow/pull/16287) |
+| excel_parser | no spurious header-only chunk when data rows are an exact multiple of `chunk_rows` | [#15490](https://github.com/infiniflow/ragflow/pull/15490) |
+| excel_parser | binary-search actual row count for sheets with abnormal `max_row` | [#13018](https://github.com/infiniflow/ragflow/pull/13018) |
+| markdown_parser | full sync to upstream state: protected ranges (fenced code / MD tables / HTML tables) never split by delimiters; GFM separators with 1+ dashes; lone-header attachment | #10896 #11018 #11520 #13892 #15630 #15632 #16109 #16319 |
+| ppt_parser | cached, None-safe shape sorting (group shapes with `top=None` no longer crash) | [#13054](https://github.com/infiniflow/ragflow/pull/13054) |
+| parser/utils | `get_text`: `if binary:` → `if binary is not None:` | [#13196](https://github.com/infiniflow/ragflow/pull/13196) |
+| table_structure_recognizer | English caption patterns (`Figure N` / `Table N` / `Fig. N`) | [#15481](https://github.com/infiniflow/ragflow/pull/15481) |
+| vision/ocr | env-configurable ONNX thread counts (`OCR_INTRA_OP_NUM_THREADS` / `OCR_INTER_OP_NUM_THREADS`) and opt-in GPU arena shrinkage (`OCR_GPUMEM_ARENA_SHRINKAGE=1`) | [#12777](https://github.com/infiniflow/ragflow/pull/12777) |
+| layout_recognizer | figure/equation placeholder boxes get per-type layoutno namespaces (`figure-N` / `equation-N`), preventing unrelated figure and equation crops from merging on scientific PDFs | [#15873](https://github.com/infiniflow/ragflow/pull/15873) |
+| layout_recognizer | re-enable CID-pattern garbage filter for text boxes | [#13404](https://github.com/infiniflow/ragflow/pull/13404) |
+| pdf_parser | garbled-text detection → OCR fallback (PUA/CID chars, subset-font encoding garbling, OCR-alphabet coverage check) | [#13404](https://github.com/infiniflow/ragflow/pull/13404) |
+| pdf_parser | geometry-based word-space recovery, CJK-aware (`_insert_word_spaces`) | [#16958](https://github.com/infiniflow/ragflow/pull/16958) |
+| pdf_parser | remove 299-page hardcoded limit (`MAXIMUM_PAGE_NUMBER = 100000`) | [#14382](https://github.com/infiniflow/ragflow/pull/14382) |
+| pdf_parser | chunked `parse_into_bboxes` for large PDFs (`PDF_PARSER_PAGE_BATCH_SIZE`, default 50) with window→global box remapping; lazy `img_np` in `__ocr` | [#14385](https://github.com/infiniflow/ragflow/pull/14385) |
+| pdf_parser | out-of-range page-index guards in `cropout` / table-figure insertion | [#12938](https://github.com/infiniflow/ragflow/pull/12938) [#12848](https://github.com/infiniflow/ragflow/pull/12848) |
+| pdf_parser (VisionParser) | `pdf_page_num = from_page + idx` (page numbering was wrong when `from_page > 0`); fixed broken relative import of `llm_adapter` | [#12938](https://github.com/infiniflow/ragflow/pull/12938) + fork-local fix |
+| parser (new) | EPUB parser (spine-order XHTML extraction, delegates to HtmlParser), exported as `deepdoc.EpubParser` | [#13650](https://github.com/infiniflow/ragflow/pull/13650) |
+
+Already present before this sync (no action needed): `ast.literal_eval`
+security fix (#12236), XLS/calamine loading (#10660), chartsheet guards
+(#10819), KMeans multi-column detection (#11415 + #12534 cleanup),
+`is_english` regex fix (#11432), OCR GPU memory limit env (#10407),
+`vision_figure_parser_docx_wrapper` guard (#12104).
+
+**Deliberately NOT ported** (decisions):
+
+- **Go + HTTP microservice pipeline** (#16323, `deepdoc/server/`, `DEEPDOC_URL`
+  DLA client): conflicts with this library's pure-Python, in-process design.
+- **Automatic table orientation detection/rotation** (#12719 feature +
+  #12981 / #17016 follow-ups): feature not in our baseline; runs 4× OCR per
+  table; adopt separately if ever needed.
+- **`rag.utils.lazy_image` memory refactors** (docx/excel lazy image loading,
+  #13329 / #13558): optimization only; deferred until vendoring lazy_image is
+  justified.
+- **3-tuple outlines** (`extract_pdf_outlines` returning
+  `(title, depth, page)`, #10456 TOC feature): our `outlines` stays
+  `(title, depth)` to avoid breaking downstream consumers.
+- **`model_speciess` → `model_species` rename** (#13929): our attribute name
+  is part of the public surface; renaming would break existing callers.
+- **paddleocr / opendataloader / somark parsers**: cloud-API clients, not
+  currently needed; revisit on demand.
+- **MinerU feature evolution** (API-only backends, VLM descriptions, etc.):
+  only the security fix was taken; feature sync deferred until MinerU usage
+  is confirmed.
+- **Pure refactors** (moving functions between ragflow modules, lefthook
+  reformatting): not applicable to our layout.
+
+Regression tests: `tests/test_upstream_sync_regressions.py`.
+
+## How to run the next sync
+
+```bash
+# in a ragflow clone
+git log --reverse --format="%h %ad %s" --date=short 18ea0fb7..origin/main -- deepdoc
+```
+
+Then for each commit: understand the behavior change, find the corresponding
+code in this repo, and re-implement it in this library's style (config
+injection, relative imports, vendored `common/` + `depend/`). Update this file.

{deepdoc_lib-0.2.2 → deepdoc_lib-0.2.4}/deepdoc/__init__.py

@@ -27,6 +27,7 @@ __all__ = [
     "PlainParser",
     "DocxParser",
     "DoclingParser",
+    "EpubParser",
     "ExcelParser",
     "PptParser",
     "HtmlParser",

{deepdoc_lib-0.2.2 → deepdoc_lib-0.2.4}/deepdoc/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.2.2'
-__version_tuple__ = version_tuple = (0, 2, 2)
+__version__ = version = '0.2.4'
+__version_tuple__ = version_tuple = (0, 2, 4)
 
-__commit_id__ = commit_id = 'g2c122018a'
+__commit_id__ = commit_id = 'g7403fae36'

{deepdoc_lib-0.2.2 → deepdoc_lib-0.2.4}/deepdoc/depend/simple_cv_model.py

@@ -11,7 +11,7 @@ import requests
 from openai import OpenAI
 from openai.lib.azure import AzureOpenAI
 from PIL import Image
-from zhipuai import ZhipuAI
+from zai import ZhipuAiClient
 
 # 可选导入 ollama
 try:
@@ -162,7 +162,12 @@ class Zhipu4V(Base):
 
     def __init__(self, key, model_name="glm-4v", lang="Chinese", **kwargs):
         super().__init__(key, model_name, lang)
-        self.client = ZhipuAI(api_key=key)
+        # Preserve the legacy override and China default instead of allowing
+        # the maintained SDK's ZAI_BASE_URL to redirect this provider adapter.
+        self.client = ZhipuAiClient(
+            api_key=key,
+            base_url=kwargs.get("base_url") or os.getenv("ZHIPUAI_BASE_URL") or "https://open.bigmodel.cn/api/paas/v4/",
+        )
 
     def describe_with_prompt(self, image, prompt=None):
         """使用智谱AI视觉模型描述图片"""
@@ -466,4 +471,4 @@ def create_vision_model(config=None):
         return VisionModelFactory.create_model(config)
     
     else:
-        raise ValueError(f"不支持的配置类型: {type(config)}")
+        raise ValueError(f"不支持的配置类型: {type(config)}")

{deepdoc_lib-0.2.2 → deepdoc_lib-0.2.4}/deepdoc/parser/__init__.py

@@ -16,6 +16,7 @@
 
 from .docx_parser import RAGFlowDocxParser as DocxParser
 from .docling_parser import DoclingParser
+from .epub_parser import RAGFlowEpubParser as EpubParser
 from .excel_parser import RAGFlowExcelParser as ExcelParser
 from .html_parser import RAGFlowHtmlParser as HtmlParser
 from .json_parser import RAGFlowJsonParser as JsonParser
@@ -31,6 +32,7 @@ __all__ = [
     "PlainParser",
     "DocxParser",
     "DoclingParser",
+    "EpubParser",
     "ExcelParser",
     "PptParser",
     "HtmlParser",

deepdoc_lib-0.2.4/deepdoc/parser/epub_parser.py

@@ -0,0 +1,139 @@
+#
+#  Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+
+import logging
+import warnings
+import zipfile
+from io import BytesIO
+from xml.etree import ElementTree
+
+from ..config import TokenizerConfig
+from .html_parser import RAGFlowHtmlParser
+
+# OPF XML namespaces
+_OPF_NS = "http://www.idpf.org/2007/opf"
+_CONTAINER_NS = "urn:oasis:names:tc:opendocument:xmlns:container"
+
+# Media types that contain readable XHTML content
+_XHTML_MEDIA_TYPES = {"application/xhtml+xml", "text/html", "text/xml"}
+
+logger = logging.getLogger(__name__)
+
+
+class RAGFlowEpubParser:
+    """Parse EPUB files by extracting XHTML content in spine (reading) order
+    and delegating to RAGFlowHtmlParser for chunking."""
+
+    def __init__(self, tokenizer_cfg: TokenizerConfig | None = None):
+        self.html_parser = RAGFlowHtmlParser(tokenizer_cfg=tokenizer_cfg)
+
+    def __call__(self, fnm, binary=None, chunk_token_num=512):
+        if binary is not None:
+            if not binary:
+                logger.warning(
+                    "RAGFlowEpubParser received an empty EPUB binary payload for %r",
+                    fnm,
+                )
+                raise ValueError("Empty EPUB binary payload")
+            zf = zipfile.ZipFile(BytesIO(binary))
+        else:
+            zf = zipfile.ZipFile(fnm)
+
+        try:
+            content_items = self._get_spine_items(zf)
+            all_sections = []
+
+            for item_path in content_items:
+                try:
+                    html_bytes = zf.read(item_path)
+                except KeyError:
+                    continue
+                if not html_bytes:
+                    logger.debug("Skipping empty EPUB content item: %s", item_path)
+                    continue
+                with warnings.catch_warnings():
+                    warnings.filterwarnings("ignore", category=UserWarning)
+                    sections = self.html_parser(item_path, binary=html_bytes, chunk_token_num=chunk_token_num)
+                all_sections.extend(sections)
+
+            return all_sections
+        finally:
+            zf.close()
+
+    @staticmethod
+    def _get_spine_items(zf):
+        """Return content file paths in spine (reading) order."""
+        # 1. Find the OPF file path from META-INF/container.xml
+        try:
+            container_xml = zf.read("META-INF/container.xml")
+        except KeyError:
+            return RAGFlowEpubParser._fallback_xhtml_order(zf)
+
+        try:
+            container_root = ElementTree.fromstring(container_xml)
+        except ElementTree.ParseError:
+            logger.warning("Failed to parse META-INF/container.xml; falling back to XHTML order.")
+            return RAGFlowEpubParser._fallback_xhtml_order(zf)
+
+        rootfile_el = container_root.find(f".//{{{_CONTAINER_NS}}}rootfile")
+        if rootfile_el is None:
+            return RAGFlowEpubParser._fallback_xhtml_order(zf)
+
+        opf_path = rootfile_el.get("full-path", "")
+        if not opf_path:
+            return RAGFlowEpubParser._fallback_xhtml_order(zf)
+
+        # Base directory of the OPF file (content paths are relative to it)
+        opf_dir = opf_path.rsplit("/", 1)[0] + "/" if "/" in opf_path else ""
+
+        # 2. Parse the OPF file
+        try:
+            opf_xml = zf.read(opf_path)
+        except KeyError:
+            return RAGFlowEpubParser._fallback_xhtml_order(zf)
+
+        try:
+            opf_root = ElementTree.fromstring(opf_xml)
+        except ElementTree.ParseError:
+            logger.warning("Failed to parse OPF file '%s'; falling back to XHTML order.", opf_path)
+            return RAGFlowEpubParser._fallback_xhtml_order(zf)
+
+        # 3. Build id->href+mediatype map from <manifest>
+        manifest = {}
+        for item in opf_root.findall(f".//{{{_OPF_NS}}}item"):
+            item_id = item.get("id", "")
+            href = item.get("href", "")
+            media_type = item.get("media-type", "")
+            if item_id and href:
+                manifest[item_id] = (href, media_type)
+
+        # 4. Walk <spine> to get reading order
+        spine_items = []
+        for itemref in opf_root.findall(f".//{{{_OPF_NS}}}itemref"):
+            idref = itemref.get("idref", "")
+            if idref not in manifest:
+                continue
+            href, media_type = manifest[idref]
+            if media_type not in _XHTML_MEDIA_TYPES:
+                continue
+            spine_items.append(opf_dir + href)
+
+        return spine_items if spine_items else RAGFlowEpubParser._fallback_xhtml_order(zf)
+
+    @staticmethod
+    def _fallback_xhtml_order(zf):
+        """Fallback: return all .xhtml/.html files sorted alphabetically."""
+        return sorted(n for n in zf.namelist() if n.lower().endswith((".xhtml", ".html", ".htm")) and not n.startswith("META-INF/"))

{deepdoc_lib-0.2.2 → deepdoc_lib-0.2.4}/deepdoc/parser/excel_parser.py

@@ -74,6 +74,14 @@ class RAGFlowExcelParser:
 
         return df.apply(lambda col: col.map(clean_string))
 
+    @staticmethod
+    def _fill_worksheet_from_dataframe(ws, df: pd.DataFrame):
+        for col_num, column_name in enumerate(df.columns, 1):
+            ws.cell(row=1, column=col_num, value=column_name)
+        for row_num, row in enumerate(df.values, 2):
+            for col_num, value in enumerate(row, 1):
+                ws.cell(row=row_num, column=col_num, value=value)
+
     @staticmethod
     def _dataframe_to_workbook(df):
         # if contains multiple sheets use _dataframes_to_workbook
@@ -84,30 +92,19 @@ class RAGFlowExcelParser:
         wb = Workbook()
         ws = wb.active
         ws.title = "Data"
-
-        for col_num, column_name in enumerate(df.columns, 1):
-            ws.cell(row=1, column=col_num, value=column_name)
-
-        for row_num, row in enumerate(df.values, 2):
-            for col_num, value in enumerate(row, 1):
-                ws.cell(row=row_num, column=col_num, value=value)
-
+        RAGFlowExcelParser._fill_worksheet_from_dataframe(ws, df)
         return wb
-    
+
     @staticmethod
     def _dataframes_to_workbook(dfs: dict):
         wb = Workbook()
         default_sheet = wb.active
         wb.remove(default_sheet)
-        
+
         for sheet_name, df in dfs.items():
             df = RAGFlowExcelParser._clean_dataframe(df)
             ws = wb.create_sheet(title=sheet_name)
-            for col_num, column_name in enumerate(df.columns, 1):
-                ws.cell(row=1, column=col_num, value=column_name)
-            for row_num, row in enumerate(df.values, 2):
-                for col_num, value in enumerate(row, 1):
-                    ws.cell(row=row_num, column=col_num, value=value)
+            RAGFlowExcelParser._fill_worksheet_from_dataframe(ws, df)
         return wb
 
     @staticmethod
@@ -156,6 +153,55 @@ class RAGFlowExcelParser:
                 continue
         return raw_items
 
+    @staticmethod
+    def _get_actual_row_count(ws):
+        max_row = ws.max_row
+        if not max_row:
+            return 0
+        if max_row <= 10000:
+            return max_row
+
+        max_col = min(ws.max_column or 1, 50)
+
+        def row_has_data(row_idx):
+            for col_idx in range(1, max_col + 1):
+                cell = ws.cell(row=row_idx, column=col_idx)
+                if cell.value is not None and str(cell.value).strip():
+                    return True
+            return False
+
+        if not any(row_has_data(i) for i in range(1, min(101, max_row + 1))):
+            return 0
+
+        left, right = 1, max_row
+        last_data_row = 1
+
+        while left <= right:
+            mid = (left + right) // 2
+            found = False
+            for r in range(mid, min(mid + 10, max_row + 1)):
+                if row_has_data(r):
+                    found = True
+                    last_data_row = max(last_data_row, r)
+                    break
+            if found:
+                left = mid + 1
+            else:
+                right = mid - 1
+
+        for r in range(last_data_row, min(last_data_row + 500, max_row + 1)):
+            if row_has_data(r):
+                last_data_row = r
+
+        return last_data_row
+
+    @staticmethod
+    def _get_rows_limited(ws):
+        actual_rows = RAGFlowExcelParser._get_actual_row_count(ws)
+        if actual_rows == 0:
+            return []
+        return list(ws.iter_rows(min_row=1, max_row=actual_rows))
+
     def html(self, fnm, chunk_rows=256):
         from html import escape
 
@@ -171,7 +217,7 @@ class RAGFlowExcelParser:
         for sheetname in wb.sheetnames:
             ws = wb[sheetname]
             try:
-                rows = list(ws.rows)
+                rows = RAGFlowExcelParser._get_rows_limited(ws)
             except Exception as e:
                 logging.warning(f"Skip sheet '{sheetname}' due to rows access error: {e}")
                 continue
@@ -184,7 +230,12 @@ class RAGFlowExcelParser:
                 tb_rows_0 += f"<th>{escape(_fmt(t.value))}</th>"
             tb_rows_0 += "</tr>"
 
-            for chunk_i in range((len(rows) - 1) // chunk_rows + 1):
+            # rows[0] is the header; split the remaining data rows into
+            # ceil(n_data / chunk_rows) chunks. Using +1 here over-counts by one
+            # when the data-row count is an exact multiple of chunk_rows and emits
+            # a spurious header-only chunk.
+            n_data_rows = len(rows) - 1
+            for chunk_i in range((n_data_rows + chunk_rows - 1) // chunk_rows):
                 tb = ""
                 tb += f"<table><caption>{sheetname}</caption>"
                 tb += tb_rows_0
@@ -223,7 +274,7 @@ class RAGFlowExcelParser:
         for sheetname in wb.sheetnames:
             ws = wb[sheetname]
             try:
-                rows = list(ws.rows)
+                rows = RAGFlowExcelParser._get_rows_limited(ws)
             except Exception as e:
                 logging.warning(f"Skip sheet '{sheetname}' due to rows access error: {e}")
                 continue
@@ -233,11 +284,13 @@ class RAGFlowExcelParser:
             for r in list(rows[1:]):
                 fields = []
                 for i, c in enumerate(r):
-                    if not c.value:
+                    if c.value is None or str(c.value).strip() == "":
                         continue
                     t = str(ti[i].value) if i < len(ti) else ""
                     t += ("：" if t else "") + str(c.value)
                     fields.append(t)
+                if not fields:
+                    continue
                 line = "; ".join(fields)
                 if sheetname.lower().find("sheet") < 0:
                     line += " ——" + sheetname
@@ -251,12 +304,12 @@ class RAGFlowExcelParser:
             total = 0
             
             for sheetname in wb.sheetnames:
-               try:
-                   ws = wb[sheetname]
-                   total += len(list(ws.rows))
-               except Exception as e:
-                   logging.warning(f"Skip sheet '{sheetname}' due to rows access error: {e}")
-                   continue
+                try:
+                    ws = wb[sheetname]
+                    total += RAGFlowExcelParser._get_actual_row_count(ws)
+                except Exception as e:
+                    logging.warning(f"Skip sheet '{sheetname}' due to rows access error: {e}")
+                    continue
             return total
 
         if fnm.split(".")[-1].lower() in ["csv", "txt"]: