亲宝软件园·资讯

展开

python修改properties 在python中修改.properties文件的操作

十年23 人气:2

在java 编程中,很多配置文件用键值对的方式存储在 properties 文件中,可以读取,修改。而且在java 中有 java.util.Properties 这个类,可以很方便的处理properties 文件, 在python 中虽然也有读取配置文件的类ConfigParser, 但如果习惯java 编程的人估计更喜欢下面这个用python 实现的读取 properties 文件的类:

"""
A Python replacement for java.util.Properties class
This is modelled as closely as possible to the Java original. 
"""

import sys,os
import re
import time

class IllegalArgumentException(Exception):

  def __init__(self, lineno, msg):
    self.lineno = lineno
    self.msg = msg

  def __str__(self):
    s='Exception at line number %d => %s' % (self.lineno, self.msg)
    return s

class Properties(object):
  """ A Python replacement for java.util.Properties """

  def __init__(self, props=None):

    # Note: We don't take a default properties object
    # as argument yet

    # Dictionary of properties.
    self._props = {}
    # Dictionary of properties with 'pristine' keys
    # This is used for dumping the properties to a file
    # using the 'store' method
    self._origprops = {}

    # Dictionary mapping keys from property
    # dictionary to pristine dictionary
    self._keymap = {}

    self.othercharre = re.compile(r'(?<!\\)(\s*\=)|(?<!\\)(\s*\:)')
    self.othercharre2 = re.compile(r'(\s*\=)|(\s*\:)')
    self.bspacere = re.compile(r'\\(?!\s$)')

  def __str__(self):
    s='{'
    for key,value in self._props.items():
      s = ''.join((s,key,'=',value,', '))

    s=''.join((s[:-2],'}'))
    return s

  def __parse(self, lines):
    """ Parse a list of lines and create
    an internal property dictionary """

    # Every line in the file must consist of either a comment
    # or a key-value pair. A key-value pair is a line consisting
    # of a key which is a combination of non-white space characters
    # The separator character between key-value pairs is a '=',
    # ':' or a whitespace character not including the newline.
    # If the '=' or ':' characters are found, in the line, even
    # keys containing whitespace chars are allowed.

    # A line with only a key according to the rules above is also
    # fine. In such case, the value is considered as the empty string.
    # In order to include characters '=' or ':' in a key or value,
    # they have to be properly escaped using the backslash character.

    # Some examples of valid key-value pairs:
    #
    # key   value
    # key=value
    # key:value
    # key   value1,value2,value3
    # key   value1,value2,value3 \
    #     value4, value5
    # key
    # This key= this value
    # key = value1 value2 value3

    # Any line that starts with a '#' is considerered a comment
    # and skipped. Also any trailing or preceding whitespaces
    # are removed from the key/value.

    # This is a line parser. It parses the
    # contents like by line.

    lineno=0
    i = iter(lines)

    for line in i:
      lineno += 1
      line = line.strip()
      # Skip null lines
      if not line: continue
      # Skip lines which are comments
      if line[0] == '#': continue
      # Some flags
      escaped=False
      # Position of first separation char
      sepidx = -1
      # A flag for performing wspace re check
      flag = 0
      # Check for valid space separation
      # First obtain the max index to which we
      # can search.
      m = self.othercharre.search(line)
      if m:
        first, last = m.span()
        start, end = 0, first
        flag = 1
        wspacere = re.compile(r'(?<![\\\=\:])(\s)')    
      else:
        if self.othercharre2.search(line):
          # Check if either '=' or ':' is present
          # in the line. If they are then it means
          # they are preceded by a backslash.

          # This means, we need to modify the
          # wspacere a bit, not to look for
          # : or = characters.
          wspacere = re.compile(r'(?<![\\])(\s)')    
        start, end = 0, len(line)

      m2 = wspacere.search(line, start, end)
      if m2:
        # print 'Space match=>',line
        # Means we need to split by space.
        first, last = m2.span()
        sepidx = first
      elif m:
        # print 'Other match=>',line
        # No matching wspace char found, need
        # to split by either '=' or ':'
        first, last = m.span()
        sepidx = last - 1
        # print line[sepidx]


      # If the last character is a backslash
      # it has to be preceded by a space in which
      # case the next line is read as part of the
      # same property
      while line[-1] == '\\':
        # Read next line
        nextline = i.next()
        nextline = nextline.strip()
        lineno += 1
        # This line will become part of the value
        line = line[:-1] + nextline

      # Now split to key,value according to separation char
      if sepidx != -1:
        key, value = line[:sepidx], line[sepidx+1:]
      else:
        key,value = line,''

      self.processPair(key, value)

  def processPair(self, key, value):
    """ Process a (key, value) pair """

    oldkey = key
    oldvalue = value

    # Create key intelligently
    keyparts = self.bspacere.split(key)
    # print keyparts

    strippable = False
    lastpart = keyparts[-1]

    if lastpart.find('\\ ') != -1:
      keyparts[-1] = lastpart.replace('\\','')

    # If no backspace is found at the end, but empty
    # space is found, strip it
    elif lastpart and lastpart[-1] == ' ':
      strippable = True

    key = ''.join(keyparts)
    if strippable:
      key = key.strip()
      oldkey = oldkey.strip()

    oldvalue = self.unescape(oldvalue)
    value = self.unescape(value)

    self._props[key] = value.strip()

    # Check if an entry exists in pristine keys
    if self._keymap.has_key(key):
      oldkey = self._keymap.get(key)
      self._origprops[oldkey] = oldvalue.strip()
    else:
      self._origprops[oldkey] = oldvalue.strip()
      # Store entry in keymap
      self._keymap[key] = oldkey

  def escape(self, value):

    # Java escapes the '=' and ':' in the value
    # string with backslashes in the store method.
    # So let us do the same.
    newvalue = value.replace(':','\:')
    newvalue = newvalue.replace('=','\=')

    return newvalue

  def unescape(self, value):

    # Reverse of escape
    newvalue = value.replace('\:',':')
    newvalue = newvalue.replace('\=','=')

    return newvalue  

  def load(self, stream):
    """ Load properties from an open file stream """

    # For the time being only accept file input streams
    if type(stream) is not file:
      raise TypeError,'Argument should be a file object!'
    # Check for the opened mode
    if stream.mode != 'r':
      raise ValueError,'Stream should be opened in read-only mode!'

    try:
      lines = stream.readlines()
      self.__parse(lines)
    except IOError, e:
      raise

  def getProperty(self, key):
    """ Return a property for the given key """

    return self._props.get(key,'')

  def setProperty(self, key, value):
    """ Set the property for the given key """

    if type(key) is str and type(value) is str:
      self.processPair(key, value)
    else:
      raise TypeError,'both key and value should be strings!'

  def propertyNames(self):
    """ Return an iterator over all the keys of the property
    dictionary, i.e the names of the properties """

    return self._props.keys()

  def list(self, out=sys.stdout):
    """ Prints a listing of the properties to the
    stream 'out' which defaults to the standard output """

    out.write('-- listing properties --\n')
    for key,value in self._props.items():
      out.write(''.join((key,'=',value,'\n')))

  def store(self, out, header=""):
    """ Write the properties list to the stream 'out' along
    with the optional 'header' """

    if out.mode[0] != 'w':
      raise ValueError,'Steam should be opened in write mode!'

    try:
      out.write(''.join(('#',header,'\n')))
      # Write timestamp
      tstamp = time.strftime('%a %b %d %H:%M:%S %Z %Y', time.localtime())
      out.write(''.join(('#',tstamp,'\n')))
      # Write properties from the pristine dictionary
      for prop, val in self._origprops.items():
        out.write(''.join((prop,'=',self.escape(val),'\n')))

      out.close()
    except IOError, e:
      raise

  def getPropertyDict(self):
    return self._props

  def __getitem__(self, name):
    """ To support direct dictionary like access """

    return self.getProperty(name)

  def __setitem__(self, name, value):
    """ To support direct dictionary like access """

    self.setProperty(name, value)

  def __getattr__(self, name):
    """ For attributes not found in self, redirect
    to the properties dictionary """

    try:
      return self.__dict__[name]
    except KeyError:
      if hasattr(self._props,name):
        return getattr(self._props, name)

if __name__=="__main__":
  p = Properties()
  p.load(open('test2.properties'))
  p.list()
  print p
  print p.items()
  print p['name3']
  p['name3'] = 'changed = value'
  print p['name3']  
  p['new key'] = 'new value'
  p.store(open('test2.properties','w'))

当然,测试这个类你需要在程序目录下简历test2.properties 文件。才可以看到效果,基本可以达到用python 读写 properties 文件的效果.

补充知识:python修改配置文件某个字段

思路:要修改的文件filepath

将修改后的文件写入f2,删除filepath,将f2名字改为filepath,从而达到修改

修改的字段可以参数化,即下面出现的 lilei 可以参数化

imort os
tag=“jdbc.cubedata.username=”
midifyInfo=“jdbc.cubedata.username=lilei”
f1=filepath
f2=application.application
fileInfo=open(filepath)

以上这篇在python中修改.properties文件的操作就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持。

加载全部内容

相关教程
猜你喜欢
用户评论