I want to loop over the contents of a text file and do a search and replace on some lines and write the result back to the file. I could first load the whole file in memory
Based on the answer by Thomas Watnedal. However, this does not answer the line-to-line part of the original question exactly. The function can still replace on a line-to-line basis
This implementation replaces the file contents without using temporary files, as a consequence file permissions remain unchanged.
Also re.sub instead of replace, allows regex replacement instead of plain text replacement only.
Reading the file as a single string instead of line by line allows for multiline match and replacement.
import re
def replace(file, pattern, subst):
    # Read contents from file as a single string
    file_handle = open(file, 'r')
    file_string = file_handle.read()
    file_handle.close()
    # Use RE package to allow for replacement (also allowing for (multiline) REGEX)
    file_string = (re.sub(pattern, subst, file_string))
    # Write contents to file.
    # Using mode 'w' truncates the file.
    file_handle = open(file, 'w')
    file_handle.write(file_string)
    file_handle.close()
                                                                        As lassevk suggests, write out the new file as you go, here is some example code:
fin = open("a.txt")
fout = open("b.txt", "wt")
for line in fin:
    fout.write( line.replace('foo', 'bar') )
fin.close()
fout.close()
                                                                        The shortest way would probably be to use the fileinput module. For example, the following adds line numbers to a file, in-place:
import fileinput
for line in fileinput.input("test.txt", inplace=True):
    print('{} {}'.format(fileinput.filelineno(), line), end='') # for Python 3
    # print "%d: %s" % (fileinput.filelineno(), line), # for Python 2
What happens here is:
print statements write back into the original filefileinput has more bells and whistles. For example, it can be used to automatically operate on all files in sys.args[1:], without your having to iterate over them explicitly. Starting with Python 3.2 it also provides a convenient context manager for use in a with statement.
While fileinput is great for throwaway scripts, I would be wary of using it in real code because admittedly it's not very readable or familiar. In real (production) code it's worthwhile to spend just a few more lines of code to make the process explicit and thus make the code readable.
There are two options:
Here's another example that was tested, and will match search & replace patterns:
import fileinput
import sys
def replaceAll(file,searchExp,replaceExp):
    for line in fileinput.input(file, inplace=1):
        if searchExp in line:
            line = line.replace(searchExp,replaceExp)
        sys.stdout.write(line)
Example use:
replaceAll("/fooBar.txt","Hello\sWorld!$","Goodbye\sWorld.")
                                                                        A more pythonic way would be to use context managers like the code below:
from tempfile import mkstemp
from shutil import move
from os import remove
def replace(source_file_path, pattern, substring):
    fh, target_file_path = mkstemp()
    with open(target_file_path, 'w') as target_file:
        with open(source_file_path, 'r') as source_file:
            for line in source_file:
                target_file.write(line.replace(pattern, substring))
    remove(source_file_path)
    move(target_file_path, source_file_path)
You can find the full snippet here.
Create a new file, copy lines from the old to the new, and do the replacing before you write the lines to the new file.