Python DOCX 手術刀：精準切片與瘦身 (OOXML 實戰); from lxml import etree; doc_xml = zfin.read(‘word/document.xml’) ; doc_tree = etree.fromstring(doc_xml) ; used_rids = set( doc_tree.xpath( “//@r:embed | //@r:link | //@r:id”, namespaces=ns_map)) #獲取 word/document.xml 有使用的used_rids => 讀取 document.xml.rels 建立白名單 keep_files以及黑名單 rels_to_remove ,要移除的Relationship節點 => 從 XML 樹中移除未使用的 Relationship 節點 => 重寫 Zip (過濾孤兒檔案)

import os
from docx import Document
from docx.shared import Inches
from PIL import Image
import io

# 建立測試用的圖片 (紅色與藍色)
def create_dummy_image(color, filename):
    img = Image.new('RGB', (200, 200), color=color)
    img.save(filename)
    return filename

create_dummy_image('red', 'red.png')
create_dummy_image('blue', 'blue.png')

# 建立原始 Word 文件
doc = Document()
doc.add_heading('Chapter 1: Intro', level=1)
doc.add_paragraph('This is the introduction.')

doc.add_heading('Chapter 2: Target Content', level=1)
doc.add_paragraph('We want to KEEP this section and the red image.')
doc.add_picture('red.png', width=Inches(1.0))

doc.add_heading('Chapter 3: Unwanted Content', level=1)
doc.add_paragraph('We want to REMOVE this section and the blue image.')
doc.add_picture('blue.png', width=Inches(1.0))

source_path = 'source_doc.docx'
doc.save(source_path)

print(f"原始檔案已建立: {source_path}")
print(f"原始大小: {os.path.getsize(source_path)} bytes")

import os
from docx import Document
from docx.shared import Inches
from PIL import Image
import io

# 建立測試用的圖片 (紅色與藍色)
def create_dummy_image(color, filename):
    img = Image.new('RGB', (200, 200), color=color)
    img.save(filename)
    return filename

create_dummy_image('red', 'red.png')
create_dummy_image('blue', 'blue.png')

# 建立原始 Word 文件
doc = Document()
doc.add_heading('Chapter 1: Intro', level=1)
doc.add_paragraph('This is the introduction.')

doc.add_heading('Chapter 2: Target Content', level=1)
doc.add_paragraph('We want to KEEP this section and the red image.')
doc.add_picture('red.png', width=Inches(1.0))

doc.add_heading('Chapter 3: Unwanted Content', level=1)
doc.add_paragraph('We want to REMOVE this section and the blue image.')
doc.add_picture('blue.png', width=Inches(1.0))

source_path = 'source_doc.docx'
doc.save(source_path)

print(f"原始檔案已建立: {source_path}")
print(f"原始大小: {os.path.getsize(source_path)} bytes")

# %%
import shutil
from docx import Document

# 設定切片範圍 (模擬堆疊演算法算出的結果)
# Index 0: Chapter 1 Heading
# Index 1: Chapter 1 Text
# Index 2: Chapter 2 Heading (Start)
# Index 3: Chapter 2 Text
# Index 4: Chapter 2 Image
# Index 5: Chapter 3 Heading (End - Exclusive)
start_idx = 2
end_idx = 5

# 1. 複製檔案 (保留樣式)
sliced_path = 'sliced_bloated.docx'
shutil.copy2(source_path, sliced_path)

# 2. 開啟副本進行手術
doc = Document(sliced_path)
body = doc.element.body
all_elements = list(body.iterchildren())

# 3. 倒序刪除不在範圍內的元素
# (倒序是為了避免刪除元素後，後續元素的 index 跑掉)
for i in reversed(range(len(all_elements))):
    if i < start_idx or i >= end_idx:
        # 這是 sectPr (Section Properties) 保護機制，避免刪除最後一個 sectPr 導致格式跑掉
        if all_elements[i].tag.endswith('sectPr'):
            continue
        body.remove(all_elements[i])

doc.save(sliced_path)

print(f"切片完成 (尚未瘦身): {sliced_path}")
print(f"切片後大小: {os.path.getsize(sliced_path)} bytes")
print("觀察：檔案大小幾乎沒有變小，因為藍色圖片還在裡面！")

# %%
import shutil
from docx import Document

# 設定切片範圍 (模擬堆疊演算法算出的結果)
# Index 0: Chapter 1 Heading
# Index 1: Chapter 1 Text
# Index 2: Chapter 2 Heading (Start)
# Index 3: Chapter 2 Text
# Index 4: Chapter 2 Image
# Index 5: Chapter 3 Heading (End - Exclusive)
start_idx = 2
end_idx = 5

# 1. 複製檔案 (保留樣式)
sliced_path = 'sliced_bloated.docx'
shutil.copy2(source_path, sliced_path)

# 2. 開啟副本進行手術
doc = Document(sliced_path)
body = doc.element.body
all_elements = list(body.iterchildren())

# 3. 倒序刪除不在範圍內的元素
# (倒序是為了避免刪除元素後，後續元素的 index 跑掉)
for i in reversed(range(len(all_elements))):
    if i < start_idx or i >= end_idx:
        # 這是 sectPr (Section Properties) 保護機制，避免刪除最後一個 sectPr 導致格式跑掉
        if all_elements[i].tag.endswith('sectPr'):
            continue
        body.remove(all_elements[i])

doc.save(sliced_path)

print(f"切片完成 (尚未瘦身): {sliced_path}")
print(f"切片後大小: {os.path.getsize(sliced_path)} bytes")
print("觀察：檔案大小幾乎沒有變小，因為藍色圖片還在裡面！")

import zipfile
import tempfile
import shutil
from lxml import etree

final_path = 'final_clean.docx'

# 定義 XML Namespace
ns_map = {
    'w': 'http://schemas.openxmlformats.org/wordprocessingml/2006/main',
    'r': 'http://schemas.openxmlformats.org/officeDocument/2006/relationships',
    'pkg': 'http://schemas.openxmlformats.org/package/2006/relationships'
}

# 準備暫存檔
with tempfile.NamedTemporaryFile(delete=False) as tmp:
    tmp_zip_path = tmp.name

with zipfile.ZipFile(sliced_path, 'r') as zfin, \
     zipfile.ZipFile(tmp_zip_path, 'w', zipfile.ZIP_DEFLATED) as zfout:
    
    # A. 讀取 document.xml 找出所有使用中的 rId
    doc_xml = zfin.read('word/document.xml')
    doc_tree = etree.fromstring(doc_xml)
    # 抓取所有 r:embed, r:link, r:id 屬性值
    used_rids = set(doc_tree.xpath("//@r:embed | //@r:link | //@r:id", namespaces=ns_map))
    
    # B. 讀取 document.xml.rels 建立白名單 keep_files 
    # 以及黑名單 rels_to_remove ,要移除的Relationship節點
    rels_xml = zfin.read('word/_rels/document.xml.rels')
    rels_tree = etree.fromstring(rels_xml)
    
    keep_files = set()
    rels_to_remove = []
    
    for rel in rels_tree.xpath("//pkg:Relationship", namespaces=ns_map):
        """ rel解析後: 
        <Relationship Id="rId324"
        Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/image"
        Target="media/image309.png" />
        """
        rid = rel.get('Id')  #"rId324"
        target = rel.get('Target') #"media/image309.png"
        
        # 正面表列：如果是 media 或 embeddings，且 rId 沒被使用，就標記刪除
        if target.startswith(('media/', 'embeddings/')):
            if rid not in used_rids:
                rels_to_remove.append(rel)
                # continue # 跳過，不加入 keep_files
            else:
                # 要保留的圖片，補上 word/ 前綴以匹配 Zip 路徑
                keep_files.add('word/' + target)
        
        # 非媒體類 (如 styles.xml, theme.xml) 全部保留
        else:
            pass 

    # C. 從 XML 樹中移除未使用的 Relationship 節點
    if rels_to_remove:
        for r in rels_to_remove:
            r.getparent().remove(r)
        # 更新 rels 內容
        new_rels_xml = etree.tostring(rels_tree, encoding='utf-8', xml_declaration=True, standalone=True) #byes
    else:
        new_rels_xml = rels_xml

    # D. 重寫 Zip (過濾孤兒檔案)
    for item in zfin.infolist():
        # 如果是我們要修改的 rels 檔，寫入新內容
        if item.filename == 'word/_rels/document.xml.rels':
            zfout.writestr(item, new_rels_xml)
            continue
        
        # 如果是 media 資料夾下的檔案，且不在白名單中 -> 刪除 (跳過不寫入)
        if item.filename.startswith(('word/media/', 'word/embeddings/')):
            if item.filename not in keep_files:
                print(f"移除孤兒檔案: {item.filename}")
                continue
        
        # 其他檔案原樣複製
        zfout.writestr(item, zfin.read(item.filename))

# 移動暫存檔到最終路徑
shutil.move(tmp_zip_path, final_path)

print(f"\n最終檔案: {final_path}")
print(f"最終大小: {os.path.getsize(final_path)} bytes")

import zipfile
import tempfile
import shutil
from lxml import etree

final_path = 'final_clean.docx'

# 定義 XML Namespace
ns_map = {
    'w': 'http://schemas.openxmlformats.org/wordprocessingml/2006/main',
    'r': 'http://schemas.openxmlformats.org/officeDocument/2006/relationships',
    'pkg': 'http://schemas.openxmlformats.org/package/2006/relationships'
}

# 準備暫存檔
with tempfile.NamedTemporaryFile(delete=False) as tmp:
    tmp_zip_path = tmp.name

with zipfile.ZipFile(sliced_path, 'r') as zfin, \
     zipfile.ZipFile(tmp_zip_path, 'w', zipfile.ZIP_DEFLATED) as zfout:
    
    # A. 讀取 document.xml 找出所有使用中的 rId
    doc_xml = zfin.read('word/document.xml')
    doc_tree = etree.fromstring(doc_xml)
    # 抓取所有 r:embed, r:link, r:id 屬性值
    used_rids = set(doc_tree.xpath("//@r:embed | //@r:link | //@r:id", namespaces=ns_map))
    
    # B. 讀取 document.xml.rels 建立白名單 keep_files 
    # 以及黑名單 rels_to_remove ,要移除的Relationship節點
    rels_xml = zfin.read('word/_rels/document.xml.rels')
    rels_tree = etree.fromstring(rels_xml)
    
    keep_files = set()
    rels_to_remove = []
    
    for rel in rels_tree.xpath("//pkg:Relationship", namespaces=ns_map):
        """ rel解析後: 
        <Relationship Id="rId324"
        Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/image"
        Target="media/image309.png" />
        """
        rid = rel.get('Id')  #"rId324"
        target = rel.get('Target') #"media/image309.png"
        
        # 正面表列：如果是 media 或 embeddings，且 rId 沒被使用，就標記刪除
        if target.startswith(('media/', 'embeddings/')):
            if rid not in used_rids:
                rels_to_remove.append(rel)
                # continue # 跳過，不加入 keep_files
            else:
                # 要保留的圖片，補上 word/ 前綴以匹配 Zip 路徑
                keep_files.add('word/' + target)
        
        # 非媒體類 (如 styles.xml, theme.xml) 全部保留
        else:
            pass 

    # C. 從 XML 樹中移除未使用的 Relationship 節點
    if rels_to_remove:
        for r in rels_to_remove:
            r.getparent().remove(r)
        # 更新 rels 內容
        new_rels_xml = etree.tostring(rels_tree, encoding='utf-8', xml_declaration=True, standalone=True) #byes
    else:
        new_rels_xml = rels_xml

    # D. 重寫 Zip (過濾孤兒檔案)
    for item in zfin.infolist():
        # 如果是我們要修改的 rels 檔，寫入新內容
        if item.filename == 'word/_rels/document.xml.rels':
            zfout.writestr(item, new_rels_xml)
            continue
        
        # 如果是 media 資料夾下的檔案，且不在白名單中 -> 刪除 (跳過不寫入)
        if item.filename.startswith(('word/media/', 'word/embeddings/')):
            if item.filename not in keep_files:
                print(f"移除孤兒檔案: {item.filename}")
                continue
        
        # 其他檔案原樣複製
        zfout.writestr(item, zfin.read(item.filename))

# 移動暫存檔到最終路徑
shutil.move(tmp_zip_path, final_path)

print(f"\n最終檔案: {final_path}")
print(f"最終大小: {os.path.getsize(final_path)} bytes")

# %%
import pandas as pd

data = {
    'File': ['Source', 'Sliced (Bloated)', 'Final (Clean)'],
    'Size (Bytes)': [
        os.path.getsize(source_path),
        os.path.getsize(sliced_path),
        os.path.getsize(final_path)
    ]
}
df = pd.DataFrame(data)
df['Reduction'] = df['Size (Bytes)'].pct_change()
df

# %%
import pandas as pd

data = {
    'File': ['Source', 'Sliced (Bloated)', 'Final (Clean)'],
    'Size (Bytes)': [
        os.path.getsize(source_path),
        os.path.getsize(sliced_path),
        os.path.getsize(final_path)
    ]
}
df = pd.DataFrame(data)
df['Reduction'] = df['Size (Bytes)'].pct_change()
df

# %%
import os
from docx import Document
from docx.shared import Inches
from PIL import Image
import io
import numpy as np

# 建立測試用的圖片 (使用隨機雜訊，讓檔案變大且難以壓縮)
def create_heavy_image(color, filename, size=(1000, 1000)):
    # 產生隨機雜訊 (這樣 PNG 壓縮不下來，檔案會很大)
    # 使用 numpy 產生隨機像素
    arr = np.random.randint(0, 255, (size[0], size[1], 3), dtype=np.uint8)
    img = Image.fromarray(arr, 'RGB')
    img.save(filename)
    return filename

print("正在產生大型測試圖片 (可能需要幾秒鐘)...")
create_heavy_image('red', 'red.png')   # 這張會很大 (約 2-3 MB)
create_heavy_image('blue', 'blue.png') # 這張也會很大

# 建立原始 Word 文件
doc = Document()
doc.add_heading('Chapter 1: Intro', level=1)
doc.add_paragraph('This is the introduction.')

doc.add_heading('Chapter 2: Target Content', level=1)
doc.add_paragraph('We want to KEEP this section and the red image.')
doc.add_picture('red.png', width=Inches(3.0))

doc.add_heading('Chapter 3: Unwanted Content', level=1)
doc.add_paragraph('We want to REMOVE this section and the blue image.')
doc.add_picture('blue.png', width=Inches(3.0))

source_path = 'source_doc.docx'
doc.save(source_path)

print(f"原始檔案已建立: {source_path}")
print(f"原始大小: {os.path.getsize(source_path) / 1024 / 1024:.2f} MB")

# %%
import os
from docx import Document
from docx.shared import Inches
from PIL import Image
import io
import numpy as np

# 建立測試用的圖片 (使用隨機雜訊，讓檔案變大且難以壓縮)
def create_heavy_image(color, filename, size=(1000, 1000)):
    # 產生隨機雜訊 (這樣 PNG 壓縮不下來，檔案會很大)
    # 使用 numpy 產生隨機像素
    arr = np.random.randint(0, 255, (size[0], size[1], 3), dtype=np.uint8)
    img = Image.fromarray(arr, 'RGB')
    img.save(filename)
    return filename

print("正在產生大型測試圖片 (可能需要幾秒鐘)...")
create_heavy_image('red', 'red.png')   # 這張會很大 (約 2-3 MB)
create_heavy_image('blue', 'blue.png') # 這張也會很大

# 建立原始 Word 文件
doc = Document()
doc.add_heading('Chapter 1: Intro', level=1)
doc.add_paragraph('This is the introduction.')

doc.add_heading('Chapter 2: Target Content', level=1)
doc.add_paragraph('We want to KEEP this section and the red image.')
doc.add_picture('red.png', width=Inches(3.0))

doc.add_heading('Chapter 3: Unwanted Content', level=1)
doc.add_paragraph('We want to REMOVE this section and the blue image.')
doc.add_picture('blue.png', width=Inches(3.0))

source_path = 'source_doc.docx'
doc.save(source_path)

print(f"原始檔案已建立: {source_path}")
print(f"原始大小: {os.path.getsize(source_path) / 1024 / 1024:.2f} MB")

            # --- Step C: 分析其他 .rels 檔 (建立全域白名單) ---
            # 核心任務：保留。單純掃描，把所有看到的 media 引用都加入 keep_files 白名單。
            # 副作用：無。只讀不改。
            # 即使您的檔案目前只有 document.xml.rels，但若文件包含頁首/頁尾圖片，
            # 關聯會存在 header1.xml.rels / footer1.xml.rels 中。
            # 若跳過此步，這些僅存在於頁首尾的圖片會被誤判為孤兒而刪除。
            other_rels = [f for f in all_files if f.endswith('.rels') and f != doc_rels_path]
            
            for rel_file in other_rels:
                try:
                    xml_bytes = zf.read(rel_file)
                    root = etree.fromstring(xml_bytes, parser=parser)
                    
                    is_in_word_rels = rel_file.startswith('word/_rels/')
                    
                    for rel in root.xpath("//pkg:Relationship", namespaces=ns_map):
                        target = rel.get('Target')
                        
                        # 正面表列: 只保留明確指向 media/ 或 embeddings/ 的參照
                        # 這是因為我們只打算刪除這類型的孤兒檔案，其他類型的參照 (如 styles.xml) 
                        # 即使沒加入 keep_files 也不會被刪除
                        # 且在 Word 結構中，參照內部圖片通常都是相對路徑且以 media/ 開頭
                        if target and target.startswith(('media/', 'embeddings/')):
                            resolved = target
                            if is_in_word_rels and not target.startswith('word/'):
                                resolved = 'word/' + target
                            keep_files.add(resolved)
                except Exception:
                    pass

            # --- Step C: 分析其他 .rels 檔 (建立全域白名單) ---
            # 核心任務：保留。單純掃描，把所有看到的 media 引用都加入 keep_files 白名單。
            # 副作用：無。只讀不改。
            # 即使您的檔案目前只有 document.xml.rels，但若文件包含頁首/頁尾圖片，
            # 關聯會存在 header1.xml.rels / footer1.xml.rels 中。
            # 若跳過此步，這些僅存在於頁首尾的圖片會被誤判為孤兒而刪除。
            other_rels = [f for f in all_files if f.endswith('.rels') and f != doc_rels_path]
            
            for rel_file in other_rels:
                try:
                    xml_bytes = zf.read(rel_file)
                    root = etree.fromstring(xml_bytes, parser=parser)
                    
                    is_in_word_rels = rel_file.startswith('word/_rels/')
                    
                    for rel in root.xpath("//pkg:Relationship", namespaces=ns_map):
                        target = rel.get('Target')
                        
                        # 正面表列: 只保留明確指向 media/ 或 embeddings/ 的參照
                        # 這是因為我們只打算刪除這類型的孤兒檔案，其他類型的參照 (如 styles.xml) 
                        # 即使沒加入 keep_files 也不會被刪除
                        # 且在 Word 結構中，參照內部圖片通常都是相對路徑且以 media/ 開頭
                        if target and target.startswith(('media/', 'embeddings/')):
                            resolved = target
                            if is_in_word_rels and not target.startswith('word/'):
                                resolved = 'word/' + target
                            keep_files.add(resolved)
                except Exception:
                    pass