Python - dealing with mixed-encoding files

If you try to decode this string as utf-8, as you already know, you will get an "UnicodeDecode" error, as these spurious cp1252 characters are invalid utf-8 -

However, Python codecs allow you to register a callback to handle encoding/decoding errors, with the codecs.register_error function - it gets the UnicodeDecodeerror a a parameter - you can write such a handler that atempts to decode the data as "cp1252", and continues the decoding in utf-8 for the rest of the string.

In my utf-8 terminal, I can build a mixed incorrect string like this:

>>> a = u"maçã ".encode("utf-8") + u"maçã ".encode("cp1252")
>>> print a
maçã ma�� 
>>> a.decode("utf-8")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python2.6/encodings/utf_8.py", line 16, in decode
    return codecs.utf_8_decode(input, errors, True)
UnicodeDecodeError: 'utf8' codec can't decode bytes in position 9-11: invalid data

I wrote the said callback function here, and found a catch: even if you increment the position from which to decode the string by 1, so that it would start on the next chratcer, if the next character is also not utf-8 and out of range(128), the error is raised at the first out of range(128) character - that means, the decoding "walks back" if consecutive non-ascii, non-utf-8 chars are found.

The worka round this is to have a state variable in the error_handler which detects this "walking back" and resume decoding from the last call to it - on this short example, I implemented it as a global variable - (it will have to be manually reset to "-1" before each call to the decoder):

import codecs

last_position = -1

def mixed_decoder(unicode_error):
    global last_position
    string = unicode_error[1]
    position = unicode_error.start
    if position <= last_position:
        position = last_position + 1
    last_position = position
    new_char = string[position].decode("cp1252")
    #new_char = u"_"
    return new_char, position + 1

codecs.register_error("mixed", mixed_decoder)

And on the console:

>>> a = u"maçã ".encode("utf-8") + u"maçã ".encode("cp1252")
>>> last_position = -1
>>> print a.decode("utf-8", "mixed")
maçã maçã 

With thanks to jsbueno and a whack of other Google searches and other pounding I solved it this way.

#The following works very well but it does not allow for any attempts to FIX the data.
xmlText = unicode(xmlText, errors='replace').replace(u"\uFFFD", "?")

This version allows for a limited opportunity to repair invalid characters. Unknown characters are replaced with a safe value.

import codecs    
replacement = {
   '85' : '...',           # u'\u2026' ... character.
   '96' : '-',             # u'\u2013' en-dash
   '97' : '-',             # u'\u2014' em-dash
   '91' : "'",             # u'\u2018' left single quote
   '92' : "'",             # u'\u2019' right single quote
   '93' : '"',             # u'\u201C' left double quote
   '94' : '"',             # u'\u201D' right double quote
   '95' : "*"              # u'\u2022' bullet
}

#This is is more complex but allows for the data to be fixed.
def mixed_decoder(unicodeError):
    errStr = unicodeError[1]
    errLen = unicodeError.end - unicodeError.start
    nextPosition = unicodeError.start + errLen
    errHex = errStr[unicodeError.start:unicodeError.end].encode('hex')
    if errHex in replacement:
        return u'%s' % replacement[errHex], nextPosition
    return u'%s' % errHex, nextPosition   # Comment this line out to get a question mark
    return u'?', nextPosition

codecs.register_error("mixed", mixed_decoder)

xmlText = xmlText.decode("utf-8", "mixed")

Basically I attempt to turn it into utf8. For any characters that fail I just convert it to HEX so I can display or look it up in a table of my own.

This is not pretty but it does allow me to make sense of messed up data