脚本之家,脚本语言编程技术及教程分享平台!
分类导航

Python|VBS|Ruby|Lua|perl|VBA|Golang|PowerShell|Erlang|autoit|Dos|bat|

服务器之家 - 脚本之家 - Python - libreoffice python 操作word及excel文档的方法

libreoffice python 操作word及excel文档的方法

2021-08-01 00:44鸭子船长 Python

这篇文章主要介绍了libreoffice python 操作word及excel文档的方法,文中通过示例代码介绍的非常详细,对大家的学习或者工作具有一定的参考学习价值,需要的朋友们下面随着小编来一起学习学习吧

1、开始、关闭libreoffice服务;

开始之前同步字体文件时间,是因为创建soffice服务时,服务会检查所需加载的文件的时间,如果其认为时间不符,则其可能会重新加载,耗时较长,因此需事先统一时间。

使用时如果需要多次调用,最后每次调用均开启后关闭,否则libreoffice会创建一个缓存文档并越用越大,处理时间会增加。

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
class OfficeProcess(object):
  def __init__(self):
    self.p = 0
    subprocess.Popen('find /usr/share/fonts | xargs touch -m -t 201801010000.00', shell=True)
 
  def start_office(self):
    self.p = subprocess.Popen('soffice --pidfile=sof.pid --invisible --accept="socket,host=localhost,port=2002;urp;"', shell=True)
    while True:
      try:
        local_context = uno.getComponentContext()
        resolver = local_context.getServiceManager().createInstanceWithContext('com.sun.star.bridge.UnoUrlResolver', local_context)
        resolver.resolve('uno:socket,host=localhost,port=2002;urp;StarOffice.ComponentContext')
        return
      except:
        print(ts(), "wait for connecting soffice...")
        time.sleep(1)
        continue
 
  def stop_office(self):
    with open("sof.pid", "rb") as f:
      try:
        os.kill(int(f.read()), signal.SIGTERM)
        self.p.wait()
      except:
        pass

2、init service manager

?
1
2
3
4
5
6
local_context = uno.getComponentContext()
    service_manager = local_context.getServiceManager()
    resolver = service_manager.createInstanceWithContext('com.sun.star.bridge.UnoUrlResolver', local_context)
    self.ctx = resolver.resolve('uno:socket,host=localhost,port=2002;urp;StarOffice.ComponentContext')
    self.smgr = self.ctx.ServiceManager
    self.desktop = self.smgr.createInstanceWithContext('com.sun.star.frame.Desktop', self.ctx)

3、从二进制数据中读取doc文档

?
1
2
3
4
5
6
7
8
9
10
11
12
def ImportFromMemory(self, data):
    istream = self.smgr.createInstanceWithContext('com.sun.star.io.SequenceInputStream', self.ctx)
    istream.initialize((uno.ByteSequence(data), ))
    pv = PropertyValue()
    pv.Name = 'InputStream'
    pv.Value = istream
    self.doc = {'doc': []}
    try:
      self.document = self.desktop.loadComponentFromURL('private:stream/swriter', '_blank', 0, (pv, ))
      self.text = self.document.getText()
    except:
      self.text = None

4、读取doc文档中的数据

?
1
2
3
4
5
6
7
8
9
10
11
12
13
def ExportToJson(self):
    try:
      l = self.__ParseText(self.text, self.__Callback(self.doc['doc']))
      self.doc['length'] = l
    except:
      self.doc = {'doc': [], 'length': 0}
    return json.dumps(self.doc)
 
@staticmethod
  def __Callback(alist):
    def Append(sth):
      alist.append(sth)
    return Append
?
1
2
3
4
5
6
7
8
9
10
11
12
def __ParseText(self, text, func):
    l = 0
    text_it = text.createEnumeration()
    while text_it.hasMoreElements():
      element = text_it.nextElement()
      if element.supportsService('com.sun.star.text.Paragraph'):
        l += self.__ParseParagraph(element, func)
      elif element.supportsService('com.sun.star.text.TextTable'):
        l += self.__ParseTable(element, func)
      else:
        pass
    return l
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
def __ParseParagraph(self, paragraph, func):
    p = {'paragraph': []}
    l = 0
    paragraph_it = paragraph.createEnumeration()
    while paragraph_it.hasMoreElements():
      portion = paragraph_it.nextElement()
      if portion.TextPortionType == 'Text':
        l += self.__ParsePortionText(portion, self.__Callback(p['paragraph']))
      elif portion.TextPortionType == 'SoftPageBreak':
        pass
      elif portion.TextPortionType == 'TextField':
        l += self.__ParsePortionText(portion, self.__Callback(p['paragraph']))
      else:
        l += self.__ParseTextContent(portion, self.__Callback(p['paragraph']))
    if hasattr(paragraph, 'createContentEnumeration'):
      l += self.__ParseTextContent(paragraph, self.__Callback(p['paragraph']))
    p['length'] = l
    func(p)
    return l
 
  def __ParseTextContent(self, textcontent, func):
    l = 0
    content_it = textcontent.createContentEnumeration('com.sun.star.text.TextContent')
    while content_it.hasMoreElements():
      element = content_it.nextElement()
      if element.supportsService('com.sun.star.text.TextGraphicObject'):
        l += self.__ParsePortionGraphic(element, func)
      elif element.supportsService('com.sun.star.text.TextEmbeddedObject'):
        pass
      elif element.supportsService('com.sun.star.text.TextFrame'):
        l += self.__ParseFrame(element, func)
      elif element.supportsService('com.sun.star.drawing.GroupShape'):
        l += self.__ParseGroup(element, func)
      else:
        pass
    return l
 
  def __ParseFrame(self, frame, func):
    f = {'frame': []}
    l = self.__ParseText(frame.getText(), self.__Callback(f['frame']))
    f['length'] = l
    func(f)
    return l
 
  def __ParseGroup(self, group, func):
    l = 0
    for i in range(group.getCount()):
      it = group.getByIndex(i)
      if it.supportsService('com.sun.star.drawing.Text'):
        l += self.__ParseFrame(it, func)
      else:
        pass
    return l
 
  def __ParsePortionText(self, portion_text, func):
    func({'portion': portion_text.String, 'length': len(portion_text.String)})
    return len(portion_text.String)
 
  def __ParsePortionGraphic(self, portion_graphic, func):
    gp = self.smgr.createInstanceWithContext('com.sun.star.graphic.GraphicProvider', self.ctx)
    stream = self.smgr.createInstanceWithContext('com.sun.star.io.TempFile', self.ctx)
    pv1 = PropertyValue()
    pv1.Name = 'OutputStream'
    pv1.Value = stream
    pv2 = PropertyValue()
    pv2.Name = 'MimeType'
    pv2.Value = 'image/png'
    gp.storeGraphic(portion_graphic.Graphic, (pv1, pv2))
    stream.getOutputStream().flush()
    stream.seek(0)
    l = stream.getInputStream().available()
    b = uno.ByteSequence(b'')
    stream.seek(0)
    l, b = stream.getInputStream().readBytes(b, l)
    img = {'image': base64.b64encode(b.value).decode('ascii')}
    img['height'] = portion_graphic.Height
    img['width'] = portion_graphic.Width
    img['actualheight'] = portion_graphic.ActualSize.Height
    img['actualwidth'] = portion_graphic.ActualSize.Width
    img['croptop'] = portion_graphic.GraphicCrop.Top
    img['cropbottom'] = portion_graphic.GraphicCrop.Bottom
    img['cropleft'] = portion_graphic.GraphicCrop.Left
    img['cropright'] = portion_graphic.GraphicCrop.Right
    img['length'] = 0
    func(img)
    return 0
 
  def __ParseTable(self, table, func):
    l = 0
    try:
      matrix = self.__GetTableMatrix(table)
      seps = self.__GetTableSeparators(table)
      t = {}
      count = 0
      for ri in matrix.keys():
        t[ri] = {}
        for ci in matrix[ri].keys():
          t[ri][ci] = dict(matrix[ri][ci])
          del t[ri][ci]['cell']
          t[ri][ci]['content'] = []
          l += self.__ParseText(matrix[ri][ci]['cell'], self.__Callback(t[ri][ci]['content']))
          count += t[ri][ci]['rowspan'] * t[ri][ci]['colspan']
      if count != len(t) * len(seps):
        raise ValueError('count of cells error')
      func({'table': t, 'row': len(t), 'column': len(seps), 'length': l, 'tableid': self.table_id})
      self.table_id += 1
    except:
      l = 0
      print('discard wrong table')
    return l
 
  @staticmethod
  def __GetTableSeparators(table):
    result = [table.TableColumnRelativeSum]
    for ri in range(table.getRows().getCount()):
      result += [s.Position for s in table.getRows().getByIndex(ri).TableColumnSeparators]
    result = sorted(set(result))
    for i in range(len(result) - 1):
      result[i] += 1 if result[i] + 1 == result[i + 1] else 0
    return sorted(set(result))
 
  @staticmethod
  def __NameToRC(name):
    r = int(re.sub('[A-Za-z]', '', name)) - 1
    cstr = re.sub('[0-9]', '', name)
    c = 0
    for i in range(len(cstr)):
      if cstr[i] >= 'A' and cstr[i] <= 'Z':
        c = c * 52 + ord(cstr[i]) - ord('A')
      else:
        c = c * 52 + 26 + ord(cstr[i]) - ord('a')
    return r, c
 
  @staticmethod
  def __GetTableMatrix(table):
    result = {}
    for name in table.getCellNames():
      ri, ci = WordToJson.__NameToRC(name)
      cell = table.getCellByName(name)
      if ri not in result:
        result[ri] = {}
      result[ri][ci] = {'cell': cell, 'rowspan': cell.RowSpan, 'name': name}
 
    seps = WordToJson.__GetTableSeparators(table)
    for ri in result.keys():
      sep = [s.Position for s in table.getRows().getByIndex(ri).TableColumnSeparators] + [table.TableColumnRelativeSum]
      sep = sorted(set(sep))
      for ci in result[ri].keys():
        right = seps.index(sep[ci]) if sep[ci] in seps else seps.index(sep[ci] + 1)
        left = -1 if ci == 0 else seps.index(sep[ci - 1]) if sep[ci - 1] in seps else seps.index(sep[ci - 1] + 1)
        result[ri][ci]['colspan'] = right - left
    return result

5、写doc文档

?
1
2
3
4
self.doco = self.desktop.loadComponentFromURL('private:factory/swriter', '_blank', 0, ())
    self.texto = self.doco.getText()
    self.cursoro = self.texto.createTextCursor()
    self.cursoro.ParaBottomMargin = 500
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
def __WriteText(self, text, texto, cursoro):
    for it in text:
      if 'paragraph' in it:
        self.__WriteParagraph(it, texto, cursoro)
      elif 'image' in it:
        self.__WritePortionGraphic(it, texto, cursoro)
      elif 'table' in it:
        self.__WriteTable(it, texto, cursoro)
 
  def __WriteParagraph(self, paragraph, texto, cursoro):
    if paragraph['length'] > 0:
      if 'result' in paragraph:
        for it in paragraph['result']:
          texto.insertString(cursoro, it['trans_sen'], False)
      else:
        texto.insertString(cursoro, paragraph['paragraph'], False)
      texto.insertControlCharacter(cursoro, ControlCharacter.PARAGRAPH_BREAK, False)
 
  def __WritePortionGraphic(self, portion_graphic, texto, cursoro):
    png_base64 = portion_graphic['image']
    png = base64.b64decode(png_base64)
    gp = self.smgr.createInstanceWithContext('com.sun.star.graphic.GraphicProvider', self.ctx)
    istream = self.smgr.createInstanceWithContext('com.sun.star.io.SequenceInputStream', self.ctx)
    istream.initialize((uno.ByteSequence(png), ))
    pv = PropertyValue()
    pv.Name = 'InputStream'
    pv.Value = istream
 
    actualsize = uno.createUnoStruct('com.sun.star.awt.Size')
    actualsize.Height = portion_graphic['actualheight'] if 'actualheight' in portion_graphic else portion_graphic['height']
    actualsize.Width = portion_graphic['actualwidth'] if 'actualwidth' in portion_graphic else portion_graphic['width']
    graphiccrop = uno.createUnoStruct('com.sun.star.text.GraphicCrop')
    graphiccrop.Top = portion_graphic['croptop'] if 'croptop' in portion_graphic else 0
    graphiccrop.Bottom = portion_graphic['cropbottom'] if 'cropbottom' in portion_graphic else 0
    graphiccrop.Left = portion_graphic['cropleft'] if 'cropleft' in portion_graphic else 0
    graphiccrop.Right = portion_graphic['cropright'] if 'cropright' in portion_graphic else 0
 
    image = self.doco.createInstance('com.sun.star.text.TextGraphicObject')
    image.Surround = NONE
    image.Graphic = gp.queryGraphic((pv, ))
    image.Height = portion_graphic['height']
    image.Width = portion_graphic['width']
    image.setPropertyValue('ActualSize', actualsize)
    image.setPropertyValue('GraphicCrop', graphiccrop)
    texto.insertTextContent(cursoro, image, False)
    texto.insertControlCharacter(cursoro, ControlCharacter.PARAGRAPH_BREAK, False)
 
  def __WriteTable(self, table, texto, cursoro):
    tableo = self.doco.createInstance('com.sun.star.text.TextTable')
    tableo.initialize(table['row'], table['column'])
    texto.insertTextContent(cursoro, tableo, False)
#    texto.insertControlCharacter(cursoro, ControlCharacter.PARAGRAPH_BREAK, False)
    tcursoro = tableo.createCursorByCellName("A1")
 
    hitbug = False
    if table['row'] > 1:
      tcursoro.goDown(1, True)
      hitbug = tcursoro.getRangeName() == 'A1'
 
    for ri in sorted([int(r) for r in table['table'].keys()]):
      rs = table['table'][str(ri)]
      for ci in sorted([int(c) for c in rs.keys()]):
        cell = rs[str(ci)]
        if hitbug == False and (cell['rowspan'] > 1 or cell['colspan'] > 1):
          tcursoro.gotoCellByName(cell['name'], False)
          if cell['rowspan'] > 1:
            tcursoro.goDown(cell['rowspan'] - 1, True)
          if cell['colspan'] > 1:
            tcursoro.goRight(cell['colspan'] - 1, True)
          tcursoro.mergeRange()
        ctexto = tableo.getCellByName(cell['name'])
        if ctexto == None:
          continue
        ccursoro = ctexto.createTextCursor()
        ccursoro.CharWeight = FontWeight.NORMAL
        ccursoro.CharWeightAsian = FontWeight.NORMAL
        ccursoro.ParaAdjust = LEFT
        self.__WriteText(cell['content'], ctexto, ccursoro)

6、生成二进制的doc文档数据

?
1
2
3
4
streamo = self.smgr.createInstanceWithContext('com.sun.star.io.Pipe', self.ctx)
self.doco.storeToURL('private:stream', (PropertyValue('FilterName', 0, 'MS Word 2007 XML', 0), PropertyValue('OutputStream', 0, streamo, 0)))
streamo.flush()
_, datao = streamo.readBytes(None, streamo.available())

7、从doc文档数据生成pdf的二进制数据

?
1
2
3
4
streamo = self.smgr.createInstanceWithContext('com.sun.star.io.Pipe', self.ctx)
self.doco.storeToURL('private:stream', (PropertyValue('FilterName', 0, 'writer_pdf_Export', 0), PropertyValue('OutputStream', 0, streamo, 0)))
streamo.flush()
_, datap = streamo.readBytes(None, streamo.available())

8、读取excel二进制数据

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
def ImportFromMemory(self, data):
  istream = self.smgr.createInstanceWithContext('com.sun.star.io.SequenceInputStream', self.ctx)
  istream.initialize((uno.ByteSequence(data), ))
  pv = PropertyValue()
  pv.Name = 'InputStream'
  pv.Value = istream
  self.doc = {'doc': []}
  try:
    print("before loadComponentFromURL")
    self.document = self.desktop.loadComponentFromURL('private:stream/scalc', '_blank', 0, (pv, ))
    self.sheets = self.document.getSheets()
    print("ImportFromMemory done")
  except:
    print("ImportFromMemory failed")
    self.sheets = None

9、读取excel的文本数据

?
1
2
3
4
5
6
7
def ExportToJson(self):
  try:
    l = self.__ParseText(self.sheets, self.__Callback(self.doc['doc']))
    self.doc['length'] = l
  except:
    self.doc = {'doc': [], 'length': 0}
  return json.dumps(self.doc)
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
def __ParseText(self, sheets, func):
  l = 0
  sheets_it = sheets.createEnumeration()
  while sheets_it.hasMoreElements():
    element = sheets_it.nextElement()
    if element.supportsService('com.sun.star.sheet.Spreadsheet'):
      l += self.__ParseSpreadsheet(element, func)
  return l
 
def __ParseSpreadsheet(self, spreadsheet, func):
  l = 0
  p = {'spreadsheet': []}
  visible_cells_it = spreadsheet.queryVisibleCells().getCells().createEnumeration()
  while visible_cells_it.hasMoreElements():
    cell = visible_cells_it.nextElement()
    type = cell.getType()
    if type == self.EMPTY:
      print("cell.type==empty")
    elif type == self.VALUE:
      print("cell.type==VALUE", "value=", cell.getValue(), cell.getCellAddress ())
    elif type == self.TEXT:
      print("cell.type==TEXT","content=", cell.getString().encode("UTF-8"), cell.getCellAddress ())
      l += self.__ParseCellText(spreadsheet, cell, self.__Callback(p['spreadsheet']))
      print("__ParseCellText=", p)
    elif type == self.FORMULA:
      print("cell.type==FORMULA", "formula=", cell.getValue())
  p['length'] = l
  func(p)
  return l
 
def __ParseCellText(self, sheet, cell, func):
  try:
    x = cell.getCellAddress().Column
    y = cell.getCellAddress().Row
    sheetname = sheet.getName()
  except:
    x = -1
    y = -1
    sheetname = None
  func({'celltext': cell.getString(), 'x': x, 'y': y, 'sheetname': sheetname, 'length': len(cell.getString())})
  return len(cell.getString())
?
1
2
3
4
 self.EMPTY = uno.Enum("com.sun.star.table.CellContentType", "EMPTY")
self.TEXT = uno.Enum("com.sun.star.table.CellContentType", "TEXT")
self.FORMULA = uno.Enum("com.sun.star.table.CellContentType", "FORMULA")
self.VALUE = uno.Enum("com.sun.star.table.CellContentType", "VALUE")

10、替换excel的文本信息

?
1
2
3
4
5
6
def ImportFromJson(self, data):
  doc = json.loads(data)
  try:
    self.__WriteText(doc['doc'])
  except:
    pass
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
def __WriteText(self, text):
    print("__WriteText begin:", text)
    sheet = None
    for it in text:
      if 'paragraph' in it and 'sheetname' in it:
        if sheet == None or sheet.getName() != it['sheetname']:
          try:
            sheet = self.sheets.getByName(it['sheetname'])
            print("getsheet:", it['sheetname'], "=", sheet.getName())
          except:
            sheet = None
            continue
        self.__WriteParagraph(it, sheet)
 
  def __WriteParagraph(self, paragraph, sheet):
    print("__WriteParagraph")
    if paragraph['length'] > 0:
      try:
        x = paragraph['x']
        y = paragraph['y']
        print("getcell:", x, y)
        cell = sheet.getCellByPosition(x, y)
        print("getcell done")
      except:
        return
      if 'result' in paragraph:
        for it in paragraph['result']:
          print("cell=", cell.getString())
          cell.setString(it['trans_sen'])
          print("cell,", cell.getString(), ",done")

11、生成excel文档二进制数据

?
1
2
3
4
streamo = self.smgr.createInstanceWithContext('com.sun.star.io.Pipe', self.ctx)
self.document.storeToURL('private:stream', (PropertyValue('FilterName', 0, 'Calc MS Excel 2007 XML', 0), PropertyValue('OutputStream', 0, streamo, 0)))
streamo.flush()
_, datao = streamo.readBytes(None, streamo.available())

12、生成excel的pdf文档

?
1
2
3
4
streamo = self.smgr.createInstanceWithContext('com.sun.star.io.Pipe', self.ctx)
self.document.storeToURL('private:stream', (PropertyValue('FilterName', 0, 'calc_pdf_Export', 0), PropertyValue('OutputStream', 0, streamo, 0)))
streamo.flush()
_, datap = streamo.readBytes(None, streamo.available())

以上就是本文的全部内容,希望对大家的学习有所帮助,也希望大家多多支持服务器之家。

原文链接:https://www.cnblogs.com/zl1991/p/10615881.html

延伸 · 阅读

精彩推荐