feat(KnowledgeBase):Add Word97-2003 (.doc) Binary File parsing module (#2544)

.devcontainer.json

@@ -7,9 +7,12 @@
             "USERNAME": "${localEnv:USER}"
         },
         "options": [
-            "--no-cache"
+            "--no-cache",
+            "--network=host"
         ]
     },
+    "updateRemoteUserUID": false,
+    "remoteUser": "${localEnv:USER}",
     "initializeCommand": ".devcontainer/init_env.sh",
     "name": "dbgpt",
     "workspaceFolder": "/app",

.devcontainer/Dockerfile.dev

@@ -1,8 +1,9 @@
-FROM eosphorosai/dbgpt:latest
+FROM eosphorosai/dbgpt-full:latest
 ARG PYTHON_VERSION=3.11
-ARG PIP_INDEX_URL="https://mirrors.aliyun.com/pypi/simple"
+ARG PIP_INDEX_URL="https://pypi.tuna.tsinghua.edu.cn/simple"
 ARG USERNAME
-ARG DEFAULT_VEN=/opt/.uv.venv
+ARG EXTRAS="base,proxy_openai,rag,storage_chromadb, storage_elasticsearch,cuda121,hf,quant_bnb,dbgpts"
+ARG DEFAULT_VENV=/opt/.uv.venv
 WORKDIR /app
 COPY . .
 USER root
@@ -11,30 +12,38 @@ USER root
 # between the container user (root) and the host user,
 # and to resolve the issue of the host user lacking write permissions.
 RUN . .devcontainer/.env && \
-    groupadd -g $USER_GID $USERNAME && \
+    groupadd -g $USER_GID $GROUPNAME && \
     useradd -u $USER_UID -g $USER_GID -m $USERNAME && \
     chown -R $USER_UID:$USER_GID /app
 RUN apt-get update && apt-get install -y \
     git \
     curl \
     wget \
+    python${PYTHON_VERSION}-dev \
+    default-libmysqlclient-dev \
     ssh zsh autojump curl git-flow vim sudo \
     && python${PYTHON_VERSION} -m pip install --upgrade pip \
     && python${PYTHON_VERSION} -m pip install --upgrade pipx \
     && pipx install -i $PIP_INDEX_URL uv --global \
-    && chown -R $USERNAME:$USERNAME $DEFAULT_VEN \
+    && chown -R $USERNAME:$GROUPNAME $DEFAULT_VENV \
     && echo "$USERNAME ALL=(ALL) NOPASSWD:ALL" > /etc/sudoers.d/$USERNAME \
     && chmod 0440 /etc/sudoers.d/$USERNAME
 USER $USERNAME
 ENV UV_LINK_MODE=copy \
     PIP_INDEX_URL=$PIP_INDEX_URL \
-    VIRTUAL_ENV=$DEFAULT_VEN \
-    UV_PROJECT_ENVIRONMENT=$DEFAULT_VEN \
-    UV_PYTHON=$DEFAULT_VEN/bin/python3
+    VIRTUAL_ENV=$DEFAULT_VENV \
+    UV_PROJECT_ENVIRONMENT=$DEFAULT_VENV \
+    UV_PYTHON=$DEFAULT_VENV/bin/python3 \
+    UV_INDEX=$PIP_INDEX_URL \
+    UV_DEFAULT_INDEX=$PIP_INDEX_URL
 
-RUN . $DEFAULT_VEN/bin/activate && \
-    uv pip install --prefix $VIRTUAL_ENV -r pyproject.toml --all-extras --index-url=$PIP_INDEX_URL && \
-    uv pip install --prefix $VIRTUAL_ENV -r requirements/dev-requirements.txt --index-url=$PIP_INDEX_URL && \
-    uv pip install --prefix $VIRTUAL_ENV -r requirements/lint-requirements.txt --index-url=$PIP_INDEX_URL && \
+RUN sed -i "s|/app/\.venv|${FINAL_VENV_NAME}|g" /${DEFAULT_VENV}/bin/activate && \
+    pip config set global.index-url $PIP_INDEX_URL && \
+    pip config set global.trusted-host $(echo "$PIP_INDEX_URL" | sed -E 's|^https?://([^/]+).*|\1|') && \
+    . $DEFAULT_VENV/bin/activate && \
+    extras=$(echo $EXTRAS | tr ',' '\n' | while read extra; do echo "--extra $extra"; done | tr '\n' ' ') && \
+    uv sync -v --active --all-packages $extras --default-index $PIP_INDEX_URL  && \
+    uv pip -v install --prefix $VIRTUAL_ENV -r requirements/dev-requirements.txt && \
+    uv pip -v install --prefix $VIRTUAL_ENV -r requirements/lint-requirements.txt && \
     cp .devcontainer/dbgpt.pth  /opt/.uv.venv/lib/python${PYTHON_VERSION}/site-packages/dbgpt.pth && \
     python -c "import dbgpt; print(dbgpt.__version__)"

.devcontainer/README.md

@@ -0,0 +1,36 @@
+# Developing inside a Container
+Use VS Code's ​Dev Container extension to build a containerized development environment. Leverage the eosphorosai/dbgpt:latest image as the development environment to avoid repeated dependency installations and improve development efficiency.  
+NOTE: **Compatible with Linux and Windows Subsystem for Linux (WSL) environments only.**
+# Setup
+
+- Follow the guide [Developing inside a Container](https://code.visualstudio.com/docs/devcontainers/containers) to set up the Dev Container:  
+  - Install the ​**Dev Containers** extension.   
+
+- Before the first launch, please execute the .devcontainer/init_env.sh script in the project root directory in **host**  
+- Create `models` dir in project root and download text2vec-large-chinese to models/text2vec-large-chinese
+- Use the shortcut `Ctrl+Shift+P` to open the command palette, then enter `Dev Containers: Open Folder in Container`.
+
+# Develop  
+After successfully starting the Dev Container, open the terminal    
+
+- Activate the virtual environment
+```bash
+. /opt/.uv.venv/bin/activate
+```
+
+- Customize the configuration file  
+
+You can copy the configuration file to the `.devcontainer` directory and rename it to `dev.toml` to avoid committing your personal configurations to the repository. 
+```bash
+cp configs/dbgpt-app-config.example.toml .devcontainer/dev.toml
+```
+
+- Start the service
+
+```bash
+dbgpt start webserver --config .devcontainer/dev.toml
+```
+
+# Create A Pull Request
+
+Please refer to [CONTRIBUTING.md](../CONTRIBUTING.md). Before executing the make script or git commit, remember to deactivate the current virtual environment in the development environment.

.devcontainer/init_env.sh

@@ -20,10 +20,19 @@ printf "OS=%s\nUSERNAME=%s\nUSER_UID=%s\nGROUPNAME=%s\nUSER_GID=%s\n" \
 
 # sharing-git-credentials see https://code.visualstudio.com/remote/advancedcontainers/sharing-git-credentials
 init_ssh_agent(){
+  if [[ -z "$SSH_AUTH_SOCK" || ! -S "$SSH_AUTH_SOCK" ]]; then
+    RUNNING_AGENT="$(ps -ax | grep '''ssh-agent -s''' | grep -v grep | wc -l)"
+    if [ "$RUNNING_AGENT" = "0" ]; then
+        ssh-agent -s &> $HOME/.ssh/ssh-agent
+    fi
+    eval $(cat $HOME/.ssh/ssh-agent) > /dev/null
+    ssh-add 2> /dev/null
+    echo $SSH_AUTH_SOCK
+fi
 # Define code block to insert (with unique identifier comment)
 SSH_AGENT_CODE='# SSH Agent Auto Management[ID:ssh_agent_v1]
-if [ -z "$SSH_AUTH_SOCK" ]; then
-    RUNNING_AGENT="$(ps -ax | grep '\''ssh-agent -s'\'' | grep -v grep | wc -l | tr -d '\''[:space:]'\'')"
+if [[ -z "$SSH_AUTH_SOCK" || ! -S "$SSH_AUTH_SOCK" ]]; then
+    RUNNING_AGENT="$(ps -ax | grep '\''ssh-agent -s'\'' | grep -v grep | wc -l)"
     if [ "$RUNNING_AGENT" = "0" ]; then
         ssh-agent -s &> $HOME/.ssh/ssh-agent
     fi
@@ -32,11 +41,7 @@ if [ -z "$SSH_AUTH_SOCK" ]; then
 fi
 # END_SSH_AGENT_CODE'
 
-# Auto detect shell type
 TARGET_FILE="$HOME/.bashrc"
-if [[ "$SHELL" == *"zsh"* ]]; then
-    TARGET_FILE="$HOME/.zshrc"
-fi
 
 # Create .ssh directory if not exists
 mkdir -p "$HOME/.ssh"
@@ -45,6 +50,9 @@ mkdir -p "$HOME/.ssh"
 if ! grep -q 'END_SSH_AGENT_CODE' "$TARGET_FILE"; then
     echo "Adding SSH agent management code to ${TARGET_FILE}..."
     echo "$SSH_AGENT_CODE" >> "$TARGET_FILE"
+    if [[ "$SHELL" == *"zsh"* ]]; then
+    echo "$SSH_AGENT_CODE" >> "$HOME/.zshrc"
+    fi
     echo "Code added successfully. Please run source ${TARGET_FILE} to apply changes immediately"
 else
     echo "Existing SSH agent code detected, no need to add again"

.devcontainer/post-create.sh

@@ -35,8 +35,8 @@ fi
 
 # Configuration section remains the same...
 # Apply custom configuration
-if [ -f /workspace/.devcontainer/zshrc-config ]; then
-  cp /workspace/.devcontainer/zshrc-config ~/.zshrc
+if [ -f /app/.devcontainer/zshrc-config ]; then
+  cp /app/.devcontainer/zshrc-config ~/.zshrc
 else
   # Generate basic .zshrc if no custom configuration exists
   cat << EOF >> ~/.zshrc

.gitignore

@@ -186,4 +186,6 @@ thirdparty
 /examples/**/*.gv.pdf
 /i18n/locales/**/**/*_ai_translated.po
 /i18n/locales/**/**/*~
-configs/my
+configs/my
+.devcontainer/dev.toml
+test_docs

CONTRIBUTING.md

@@ -118,3 +118,6 @@ Write necessary information about your changes and click "Create pull request".
 
 That's it you made it 🐣⭐⭐
 
+# Developing inside a Container
+
+If you are using VS Code as your IDE for development, you can refer to the [configuration here](.devcontainer/README.md) to set up the Dev Containers development environment.

docker/base/Dockerfile

@@ -118,6 +118,8 @@ COPY . .
 # Fix the shebang of the dbgpt script
 RUN sed -i "s|^#\!/app/\.venv/bin/python[0-9.]*|#!/${FINAL_VENV_NAME}/bin/python${PYTHON_VERSION}|" /${FINAL_VENV_NAME}/bin/dbgpt
 RUN sed -i "s|^#\!/app/\.venv/bin/python[0-9.]*|#!/${FINAL_VENV_NAME}/bin/python${PYTHON_VERSION}|" /${FINAL_VENV_NAME}/bin/pip
+RUN sed -i "s|/app/\.venv|${FINAL_VENV_NAME}|g" /${FINAL_VENV_NAME}/bin/activate
+
 ENV PATH="${FINAL_VENV_NAME}/bin:$PATH" \
     VIRTUAL_ENV="${FINAL_VENV_NAME}"
 # Default command

packages/dbgpt-core/src/dbgpt/rag/knowledge/base.py

@@ -27,6 +27,7 @@ class DocumentType(Enum):
     HTML = "html"
     DATASOURCE = "datasource"
     EXCEL = "xlsx"
+    DOC = "doc"
 
 
 class KnowledgeType(Enum):

packages/dbgpt-ext/pyproject.toml

@@ -31,6 +31,7 @@ rag = [
     "bs4",
     "python-pptx",
     "python-docx",
+    "olefile",
     "pypdf",
     "pdfplumber",
 ]

packages/dbgpt-ext/src/dbgpt_ext/rag/knowledge/__init__.py

@@ -19,6 +19,7 @@ def __getattr__(name: str):
         "CSVKnowledge": "csv",
         "DatasourceKnowledge": "datasource",
         "DocxKnowledge": "docx",
+        "Word97DocKnowledge": "doc",
         "HTMLKnowledge": "html",
         "MarkdownKnowledge": "markdown",
         "PDFKnowledge": "pdf",
@@ -43,6 +44,7 @@ __all__ = [
     "CSVKnowledge",
     "DatasourceKnowledge",
     "DocxKnowledge",
+    "Word97DocKnowledge",
     "HTMLKnowledge",
     "MarkdownKnowledge",
     "PDFKnowledge",

packages/dbgpt-ext/src/dbgpt_ext/rag/knowledge/doc.py

@@ -0,0 +1,627 @@
+import struct
+from typing import Any, Dict, List, Optional, Union
+
+import olefile
+
+from dbgpt.core import Document
+from dbgpt.rag.knowledge.base import (
+    ChunkStrategy,
+    DocumentType,
+    Knowledge,
+    KnowledgeType,
+)
+
+
+class Word97DocParser:
+    """Parser for Microsoft Word 97-2003 (.doc) binary files.
+
+    This module implements a parser for the legacy Word 97-2003 binary format (.doc),
+    based on the official Microsoft [MS-DOC] specification.
+
+    Specification Reference:
+        [MS-DOC]: Word (.doc) Binary File Format
+        https://learn.microsoft.com/en-us/openspecs/office_file_formats/ms-doc/ccd7b486-7881-484c-a137-51170af7cc22
+
+    Example:
+        >>> from doc import Word97DocParser
+        >>> with Word97DocParser("example.doc") as parser:
+        ...     paragraphs = parser.extract_text_by_paragraphs()
+        ...     for i, para in enumerate(paragraphs, 1):
+        ...         print(f"\nParagraph {i}:")
+        ...         print(para)
+    """
+
+    # Mapping of special ANSI characters to Unicode
+    ANSI_TO_UNICODE = {
+        0x82: 0x201A,
+        0x83: 0x0192,
+        0x84: 0x201E,
+        0x85: 0x2026,
+        0x86: 0x2020,
+        0x87: 0x2021,
+        0x88: 0x02C6,
+        0x89: 0x2030,
+        0x8A: 0x0160,
+        0x8B: 0x2039,
+        0x8C: 0x0152,
+        0x91: 0x2018,
+        0x92: 0x2019,
+        0x93: 0x201C,
+        0x94: 0x201D,
+        0x95: 0x2022,
+        0x96: 0x2013,
+        0x97: 0x2014,
+        0x98: 0x02DC,
+        0x99: 0x2122,
+        0x9A: 0x0161,
+        0x9B: 0x203A,
+        0x9C: 0x0153,
+        0x9F: 0x0178,
+    }
+
+    # Mapping of nFib values to cbRgFcLcb sizes
+    FIB_VERSIONS = {
+        0x00C1: 0x005D,  # Word 97
+        0x00D9: 0x006C,  # Word 2000
+        0x0101: 0x0088,  # Word 2002
+        0x010C: 0x00A4,  # Word 2003
+        0x0112: 0x00B7,  # Word 2007
+    }
+
+    def __init__(self, doc_path):
+        """Initialize with path to Word document"""
+        self.doc_path = doc_path
+        self.ole = None
+        self.word_doc_stream = None
+        self.table_stream = None
+        self.fib_info = None
+        self.plc_pcd = None
+
+    def __enter__(self):
+        """Context manager entry"""
+        self.open()
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        """Context manager exit"""
+        self.close()
+
+    def open(self):
+        """Open the OLE file and required streams"""
+        self.ole = olefile.OleFileIO(self.doc_path)
+
+        if not self.ole.exists("WordDocument"):
+            raise ValueError("WordDocument stream not found")
+
+        self.word_doc_stream = self.ole.openstream("WordDocument")
+
+        # Determine table stream name
+        table_stream_name = "0Table" if self.ole.exists("0Table") else "1Table"
+        self.table_stream = self.ole.openstream(table_stream_name)
+
+    def close(self):
+        """Close all open resources"""
+        if self.ole:
+            self.ole.close()
+        self.ole = None
+        self.word_doc_stream = None
+        self.table_stream = None
+
+    def read_fib(self):
+        """Read the File Information Block (FIB) from the WordDocument stream (2.5)"""
+        # Read FibBase (32 bytes)
+        fib_base = self.word_doc_stream.read(32)
+
+        # Unpack nFib (offset 2, size 2 bytes)
+        nFib = struct.unpack("<H", fib_base[2:4])[0]
+        if nFib not in self.FIB_VERSIONS:
+            raise ValueError(f"Unsupported nFib version: 0x{nFib:04X}")
+
+        # Skip csw (2 bytes at offset 32)
+        self.word_doc_stream.read(2)
+
+        # Skip FibRgW97 (28 bytes) and cslw (2 bytes)
+        self.word_doc_stream.read(28 + 2)
+
+        # Read FibRgLw97 (88 bytes)
+        fib_rg_lw = self.word_doc_stream.read(88)
+        ccpText = struct.unpack("<I", fib_rg_lw[12:16])[0]  # Total character count
+
+        # Read cbRgFcLcb (2 bytes)
+        cb_rg_fc_lcb = struct.unpack("<H", self.word_doc_stream.read(2))[0]
+
+        # Read FibRgFcLcbBlob (variable size)
+        fib_rg_fc_lcb_blob = self.word_doc_stream.read(cb_rg_fc_lcb * 8)
+
+        # Skip cswNew (2 bytes) and FibRgCswNew (variable size)
+        csw_new = struct.unpack("<H", self.word_doc_stream.read(2))[0]
+        if csw_new > 0:
+            self.word_doc_stream.read(csw_new * 2)
+
+        # Extract fcClx and lcbClx from FibRgFcLcb97
+        fc_clx = struct.unpack("<I", fib_rg_fc_lcb_blob[0x108:0x10C])[0]
+        lcb_clx = struct.unpack("<I", fib_rg_fc_lcb_blob[0x10C:0x110])[0]
+        fc_plcf_bte_papx = struct.unpack("<I", fib_rg_fc_lcb_blob[0x68:0x6C])[0]
+        lcb_plcf_bte_papx = struct.unpack("<I", fib_rg_fc_lcb_blob[0x6C:0x70])[0]
+        self.fib_info = {
+            "nFib": nFib,
+            "fcClx": fc_clx,
+            "lcbClx": lcb_clx,
+            "ccpText": ccpText,
+            "fcPlcfBtePapx": fc_plcf_bte_papx,
+            "lcbPlcfBtePapx": lcb_plcf_bte_papx,
+        }
+        return self.fib_info
+
+    def read_clx(self, fc_clx, lcb_clx):
+        """Read the CLX structure from the Table stream"""
+        self.table_stream.seek(fc_clx)
+        clx_data = self.table_stream.read(lcb_clx)
+
+        # For simplicity, we assume the data starts with Pcdt (0x02)
+        if clx_data[0] != 0x02:
+            raise ValueError("Expected Pcdt structure not found in CLX data")
+
+        return clx_data
+
+    def parse_plc_pcd(self, pcdt_data):
+        """Parse the PLC structure containing PCDs(2.9.177)"""
+        # Get the size of PlcPcd structure
+        lcb = struct.unpack("<I", pcdt_data[1:5])[0]
+        plc_pcd_bytes = pcdt_data[5 : 5 + lcb]
+
+        # Calculate number of PCDs: n = (lcb - 4) // 12
+        n = (lcb - 4) // 12
+
+        # Parse aCP array (n+1 CPs, each 4 bytes)
+        aCP = [
+            struct.unpack("<I", plc_pcd_bytes[i * 4 : (i + 1) * 4])[0]
+            for i in range(n + 1)
+        ]
+
+        # Parse aPcd array (n PCDs, each 8 bytes)
+        aPcd = []
+        for i in range(n):
+            start = (n + 1) * 4 + i * 8
+            pcd_bytes = plc_pcd_bytes[start : start + 8]
+
+            # Extract fc (bytes 2-6) and compression flag
+            fc_bytes = pcd_bytes[2:6]
+            fc = int.from_bytes(fc_bytes, byteorder="little", signed=False)
+            fcompressed = (fc >> 1) & 0x1
+
+            aPcd.append(
+                {
+                    "fc": fc,
+                    "start": aCP[i],
+                    "end": aCP[i + 1],
+                    "f_compressed": fcompressed,
+                }
+            )
+        self.plc_pcd = {"aCP": aCP, "aPcd": aPcd}
+        return self.plc_pcd
+
+    def _find_pcd_index(self, cp):
+        """Find the index of the PCD containing the given character position"""
+        aCP = self.plc_pcd["aCP"]
+        for i in range(len(aCP) - 1):
+            if aCP[i] <= cp < aCP[i + 1]:
+                return i
+        return None
+
+    def get_paragraph_boundaries(self, cp):
+        """Find paragraph boundaries for a given character position (2.4.2)"""
+        if not self.fib_info or not self.plc_pcd:
+            raise RuntimeError("Must read FIB and PLC/PCD first")
+
+        # Find the PCD containing this cp
+        i = self._find_pcd_index(cp)
+        if i is None:
+            return None
+
+        pcd = self.plc_pcd["aPcd"][i]
+        start_cp = self._find_paragraph_start(cp, i, pcd)
+        end_cp = self._find_paragraph_end(cp, i, pcd)
+
+        return (start_cp, end_cp)
+
+    def _find_paragraph_start(self, cp, i, pcd):
+        """Find the start of the paragraph containing cp (algorithm 2.4.2)"""
+        # Step 3: Calculate fc and fc_pcd
+        fc_pcd = pcd["fc"]
+        # Let fcPcd be Pcd.fc.fc. Let fc be fcPcd + 2(cp – PlcPcd.aCp[i]).
+        # If Pcd.fc.fCompressed is one, set fc to fc / 2, and set fcPcd to fcPcd/2.
+        fc = fc_pcd + 2 * (cp - self.plc_pcd["aCP"][i])
+        if pcd["f_compressed"]:
+            fc = fc // 2
+            fc_pcd = fc_pcd // 2
+
+        # Step 4: Read PlcBtePapx
+        self.table_stream.seek(self.fib_info["fcPlcfBtePapx"])
+        plcf_bte_papx_data = self.table_stream.read(self.fib_info["lcbPlcfBtePapx"])
+        a_fc, a_pn = self._parse_plcf_bte_papx(plcf_bte_papx_data)
+
+        # Handle case where a_fc is empty
+        if not a_fc:
+            return None
+
+        fc_last = a_fc[-1]
+
+        # Step 4 continued: Check fcLast
+        if fc_last <= fc:
+            if fc_last < fc_pcd:
+                # Step 8: Check if at beginning of document
+                if self.plc_pcd["aCP"][i] == 0:
+                    return 0
+                # Step 9: Recurse with previous cp
+                return self._find_paragraph_start(
+                    self.plc_pcd["aCP"][i], i - 1, self.plc_pcd["aPcd"][i - 1]
+                )
+            # Adjust fc and fc_last if needed
+            fc = fc_last
+            if pcd["f_compressed"]:
+                fc_last = fc_last // 2
+            fc_first = fc_last
+        else:
+            # Step 5: Find largest j where a_fc[j] <= fc
+            j = self._find_largest_index_le(a_fc, fc)
+            if j is None:
+                return None  # Invalid cp
+
+            # Read PapxFkp
+            papx_fkp = self._read_papx_fkp(a_pn[j])
+            # print(f"papx_fkp:{papx_fkp}, j:{j}")
+            if not papx_fkp or not papx_fkp.get("rgfc"):
+                return None  # Invalid data
+
+            # Step 6: Find largest k where rgfc[k] <= fc
+            k = self._find_largest_index_le(papx_fkp["rgfc"], fc)
+            if k is None:
+                return None  # Invalid cp
+
+            # Check if cp is outside document range
+            if papx_fkp["rgfc"][-1] <= fc:
+                return None
+
+            fc_first = papx_fkp["rgfc"][k]
+
+        # Step 7: Calculate paragraph start
+        if fc_first > fc_pcd:
+            dfc = fc_first - fc_pcd
+            if not pcd["f_compressed"]:
+                dfc = dfc // 2
+            return self.plc_pcd["aCP"][i] + dfc
+
+        # Step 8: Check if at beginning of document
+        if self.plc_pcd["aCP"][i] == 0:
+            return 0
+
+        # Step 9: Recurse with previous cp
+        return self._find_paragraph_start(
+            self.plc_pcd["aCP"][i], i - 1, self.plc_pcd["aPcd"][i - 1]
+        )
+
+    def _find_paragraph_end(self, cp, i, pcd):
+        """Find the end of the paragraph containing cp (algorithm 2.4.2)"""
+        fc_pcd = pcd["fc"]
+        fc = fc_pcd + 2 * (cp - self.plc_pcd["aCP"][i])
+        fc_mac = fc_pcd + 2 * (self.plc_pcd["aCP"][i + 1] - self.plc_pcd["aCP"][i])
+
+        if pcd["f_compressed"]:
+            fc = fc // 2
+            fc_pcd = fc_pcd // 2
+            fc_mac = fc_mac // 2
+
+        # Read PlcBtePapx
+        self.table_stream.seek(self.fib_info["fcPlcfBtePapx"])
+        plcf_bte_papx_data = self.table_stream.read(self.fib_info["lcbPlcfBtePapx"])
+        a_fc, a_pn = self._parse_plcf_bte_papx(plcf_bte_papx_data)
+
+        # Find largest j where a_fc[j] <= fc
+        j = self._find_largest_index_le(a_fc, fc)
+        if j is None or (a_fc and fc >= a_fc[-1]):
+            return self._find_paragraph_end(
+                self.plc_pcd["aCP"][i + 1], i + 1, self.plc_pcd["aPcd"][i + 1]
+            )
+
+        # Read PapxFkp
+        papx_fkp = self._read_papx_fkp(a_pn[j])
+        if not papx_fkp:
+            return None
+
+        # Find largest k where rgfc[k] <= fc
+        k = self._find_largest_index_le(papx_fkp["rgfc"], fc)
+        if k is None or (papx_fkp["rgfc"] and fc >= papx_fkp["rgfc"][-1]):
+            return None
+
+        fc_lim = papx_fkp["rgfc"][k + 1] if k + 1 < len(papx_fkp["rgfc"]) else fc_mac
+
+        if fc_lim <= fc_mac:
+            dfc = fc_lim - fc_pcd
+            if not pcd["f_compressed"]:
+                dfc = dfc // 2
+            return self.plc_pcd["aCP"][i] + dfc - 1
+
+        return self._find_paragraph_end(
+            self.plc_pcd["aCP"][i + 1], i + 1, self.plc_pcd["aPcd"][i + 1]
+        )
+
+    def _parse_plcf_bte_papx(self, data):
+        """Parse PlcBtePapx structure (2.8.6)
+
+        Args:
+            data: Raw bytes of PlcBtePapx structure.
+
+        Returns:
+            (a_fc, a_pn): Tuple of two lists:
+                - a_fc: List of unsigned 4-byte integers (FC offsets).
+                - a_pn: List of unsigned 4-byte integers (PnFkpPapx entries).
+
+        Raises:
+            ValueError: If data is malformed or aFC is not sorted/unique.
+        """
+        if len(data) < 12:  # Minimum: 8 (aFC[0..1]) + 4 (aPnBtePapx[0])
+            return [], []
+
+        # Calculate number of aPnBtePapx entries (n)
+        n = (len(data) - 4) // 8
+        if (2 * n + 1) * 4 != len(data):
+            raise ValueError("Invalid PlcBtePapx size")
+
+        a_fc = []
+        a_pn = []
+
+        # Parse aFC (n+1 entries, each 4 bytes)
+        for i in range(n + 1):
+            offset = i * 4
+            fc = struct.unpack("<I", data[offset : offset + 4])[0]
+            a_fc.append(fc)
+
+        # Parse aPnBtePapx (n entries, each 4 bytes, starting after last aFC)
+        pn_offset = (n + 1) * 4
+        for i in range(n):
+            offset = pn_offset + i * 4
+            pn = struct.unpack("<I", data[offset : offset + 4])[0]
+            a_pn.append(pn)
+
+        # Validate aFC is strictly increasing (sorted and unique)
+        for i in range(len(a_fc) - 1):
+            if a_fc[i] >= a_fc[i + 1]:
+                raise ValueError("aFC must be strictly increasing")
+
+        return a_fc, a_pn
+
+    def _read_papx_fkp(self, pn):
+        """Read PapxFkp structure from WordDocument stream.
+
+        Args:
+            pn: Page number (PnFkpPapx), offset = pn * 512.
+
+        Returns:
+            Dict with keys:
+                - "rgfc": List of FC offsets (4-byte unsigned integers).
+                - "rgbx": List of BxPap (1-byte integers).
+                - "papx_in_fkp": List of PapxInFkp raw bytes.
+
+        Raises:
+            ValueError: If FKP data is invalid.
+        """
+        offset = pn * 512
+        self.word_doc_stream.seek(offset)
+        fkp_data = self.word_doc_stream.read(512)
+
+        if len(fkp_data) != 512:
+            raise ValueError("FKP size must be 512 bytes")
+
+        cpara = fkp_data[511]  # Number of paragraphs (1 ≤ cpara ≤ 0x1D)
+        if not 1 <= cpara <= 0x1D:
+            raise ValueError(f"Invalid cpara: {cpara} (must be 1 ≤ cpara ≤ 29)")
+
+        # Parse rgfc (cpara + 1 entries, each 4 bytes)
+        rgfc = []
+        for i in range(cpara + 1):
+            fc_offset = i * 4
+            fc = struct.unpack("<I", fkp_data[fc_offset : fc_offset + 4])[0]
+            rgfc.append(fc)
+
+        # Parse rgbx (cpara entries, each 1 byte)
+        rgbx_start = (cpara + 1) * 4
+        rgbx = list(fkp_data[rgbx_start : rgbx_start + cpara])
+
+        # Parse PapxInFkp (variable size, located after rgbx)
+        papx_in_fkp_start = rgbx_start + cpara
+        papx_in_fkp_end = 511  # cpara is the last byte
+        papx_in_fkp = fkp_data[papx_in_fkp_start:papx_in_fkp_end]
+
+        return {
+            "rgfc": rgfc,
+            "rgbx": rgbx,
+            "papx_in_fkp": papx_in_fkp,
+        }
+
+    def _find_largest_index_le(self, array, value):
+        """Find largest index where array[index] <= value"""
+        for i in reversed(range(len(array))):
+            if array[i] <= value:
+                return i
+        return None
+
+    def extract_text_by_paragraphs(self):
+        """Extract text organized by paragraphs"""
+        self.word_doc_stream.seek(0)
+        self.table_stream.seek(0)
+        self.read_fib()
+        clx_data = self.read_clx(self.fib_info["fcClx"], self.fib_info["lcbClx"])
+        self.parse_plc_pcd(clx_data)
+
+        paragraphs = []
+        current_cp = 0
+        total_chars = self.fib_info["ccpText"]
+
+        while current_cp < total_chars:
+            boundaries = self.get_paragraph_boundaries(current_cp)
+            if not boundaries:
+                break
+
+            start, end = boundaries
+            if start > end:
+                break
+
+            paragraph_text = self._extract_text_range(start, end)
+            paragraphs.append(paragraph_text)
+            current_cp = end + 1
+
+        return paragraphs
+
+    def _extract_text_range(self, start_cp, end_cp):
+        """Extract text between two character positions"""
+        text_chars = []
+        i = self._find_pcd_index(start_cp)
+
+        while i is not None and start_cp <= end_cp:
+            pcd = self.plc_pcd["aPcd"][i]
+            pcd_start = self.plc_pcd["aCP"][i]
+            pcd_end = self.plc_pcd["aCP"][i + 1]
+
+            # Determine range within this PCD
+            range_start = max(start_cp, pcd_start)
+            range_end = min(end_cp, pcd_end - 1)
+
+            if range_start > range_end:
+                i += 1
+                continue
+
+            fc = pcd["fc"]
+            compressed = pcd["f_compressed"]
+
+            for cp in range(range_start, range_end + 1):
+                if compressed:
+                    offset = fc + (cp - pcd_start)
+                    self.word_doc_stream.seek(offset)
+                    char_byte = self.word_doc_stream.read(1)
+                    char_code = char_byte[0]
+                    char = chr(self.ANSI_TO_UNICODE.get(char_code, char_code))
+                else:
+                    offset = fc + 2 * (cp - pcd_start)
+                    self.word_doc_stream.seek(offset)
+                    char_bytes = self.word_doc_stream.read(2)
+                    char = char_bytes.decode("utf-16-le")
+
+                text_chars.append(char)
+
+            start_cp = range_end + 1
+            i += 1
+
+        return "".join(text_chars)
+
+    def extract_text(self):
+        """Main method to extract text from the document"""
+        self.word_doc_stream.seek(0)
+        self.table_stream.seek(0)
+        fib_info = self.read_fib()
+        clx_data = self.read_clx(fib_info["fcClx"], fib_info["lcbClx"])
+        pcd_array = self.parse_plc_pcd(clx_data)
+
+        full_text = []
+        for pcd in pcd_array["aPcd"]:
+            start_cp, end_cp = pcd["start"], pcd["end"]
+            char_count = end_cp - start_cp
+
+            if char_count == 0:
+                continue
+
+            fc = pcd["fc"]
+            compressed = pcd["f_compressed"]
+            text_chars = []
+
+            for cp in range(start_cp, end_cp):
+                offset = (
+                    fc + (cp - start_cp) if compressed else fc + 2 * (cp - start_cp)
+                )
+                self.word_doc_stream.seek(offset)
+
+                if compressed:
+                    char_byte = self.word_doc_stream.read(1)
+                    char_code = char_byte[0]
+                    char = chr(self.ANSI_TO_UNICODE.get(char_code, char_code))
+                else:
+                    char_bytes = self.word_doc_stream.read(2)
+                    # decode char
+                    char = char_bytes.decode("utf-16-le")
+                text_chars.append(char)
+            full_text.append("".join(text_chars))
+
+        return "".join(full_text)
+
+
+class Word97DocKnowledge(Knowledge):
+    """Microsoft Word 97-2003 (.doc)."""
+
+    def __init__(
+        self,
+        file_path: Optional[str] = None,
+        knowledge_type: Any = KnowledgeType.DOCUMENT,
+        encoding: Optional[str] = "utf-16-le",
+        loader: Optional[Any] = None,
+        metadata: Optional[Dict[str, Union[str, List[str]]]] = None,
+        **kwargs: Any,
+    ) -> None:
+        """Create  Microsoft Word 97-2003 (.doc) Knowledge with Knowledge arguments.
+
+        Args:
+            file_path(str,  optional): file path
+            knowledge_type(KnowledgeType, optional): knowledge type
+            encoding(str, optional): .doc encoding
+            loader(Any, optional): loader
+        """
+        super().__init__(
+            path=file_path,
+            knowledge_type=knowledge_type,
+            data_loader=loader,
+            metadata=metadata,
+            **kwargs,
+        )
+        self._encoding = encoding
+
+    def _load(self) -> List[Document]:
+        """Load doc document from loader."""
+        if self._loader:
+            documents = self._loader.load()
+        else:
+            docs = []
+            content = []
+            with Word97DocParser(self._path) as parser:
+                paragraphs = parser.extract_text_by_paragraphs()
+                for i, para in enumerate(paragraphs):
+                    content.append(para)
+
+            metadata = {"source": self._path}
+            if self._metadata:
+                metadata.update(self._metadata)  # type: ignore
+            docs.append(Document(content="\n".join(content), metadata=metadata))
+            return docs
+        return [Document.langchain2doc(lc_document) for lc_document in documents]
+
+    @classmethod
+    def support_chunk_strategy(cls) -> List[ChunkStrategy]:
+        """Return support chunk strategy."""
+        return [
+            ChunkStrategy.CHUNK_BY_SIZE,
+            ChunkStrategy.CHUNK_BY_PARAGRAPH,
+            ChunkStrategy.CHUNK_BY_SEPARATOR,
+        ]
+
+    @classmethod
+    def default_chunk_strategy(cls) -> ChunkStrategy:
+        """Return default chunk strategy."""
+        return ChunkStrategy.CHUNK_BY_SIZE
+
+    @classmethod
+    def type(cls) -> KnowledgeType:
+        """Return knowledge type."""
+        return KnowledgeType.DOCUMENT
+
+    @classmethod
+    def document_type(cls) -> DocumentType:
+        """Return document type."""
+        return DocumentType.DOC

packages/dbgpt-ext/src/dbgpt_ext/rag/knowledge/factory.py

@@ -171,6 +171,7 @@ class KnowledgeFactory:
 
         from .csv import CSVKnowledge  # noqa: F401
         from .datasource import DatasourceKnowledge  # noqa: F401
+        from .doc import Word97DocKnowledge  # noqa: F401
         from .docx import DocxKnowledge  # noqa: F401
         from .excel import ExcelKnowledge  # noqa: F401
         from .html import HTMLKnowledge  # noqa: F401

packages/dbgpt-ext/src/dbgpt_ext/rag/knowledge/tests/test_doc.py

@@ -0,0 +1,20 @@
+from pathlib import Path
+
+import pytest
+
+from ..doc import Word97DocKnowledge
+
+
+@pytest.fixture
+def mock_file_path():
+    file_path = Path(__file__).parent / "data" / "test_mock.doc"
+    return file_path.as_posix()
+
+
+def test_load_from_docx(mock_file_path):
+    knowledge = Word97DocKnowledge(file_path=mock_file_path)
+    documents = knowledge._load()
+    actual = documents[0].content.replace("\r", "\n")
+    assert len(documents) == 1
+    assert actual == "This is the first paragraph.\n\nThis is the second paragraph.\n"
+    assert documents[0].metadata["source"] == mock_file_path