There is an encoding issue between codecs.open and sax (see attached file). The issue is reproducible on Python 3.3.1, it is working fine on Python 3.3.0

Since this is a regression, setting (temporarily perhaps) as release blocker.

It looks like a regression of introduced by the fix of the issue #1470548, changeset 66f92f76b2ce.

Extracted test from report.txt. Test with Python 3.4:

$ ./python test_codecs.py 
Traceback (most recent call last):
  File "test_codecs.py", line 7, in <module>
    xml.startDocument()
  File "/home/haypo/prog/python/default/Lib/xml/sax/saxutils.py", line 148, in startDocument
    self._encoding)
  File "/home/haypo/prog/python/default/Lib/codecs.py", line 699, in write
    return self.writer.write(data)
  File "/home/haypo/prog/python/default/Lib/codecs.py", line 355, in write
    data, consumed = self.encode(object, self.errors)
TypeError: Can't convert 'bytes' object to str implicitly

_gettextwriter() of xml.sax.saxutils does not recognize codecs classes. (See also the PEP 400 :-)).

It is not working fine on Python 3.3.0.

>>> with codecs.open('/tmp/test.txt', 'w', encoding='iso-8859-1') as f:
...     xml = XMLGenerator(f, encoding='iso-8859-1')
...     xml.startDocument()
...     xml.startElement('root', {'attr': u'\u20ac'})
...     xml.endElement('root')
...     xml.endDocument()
... 
Traceback (most recent call last):
  File "<stdin>", line 4, in <module>
  File "/home/serhiy/py/cpython-3.Lib/xml/sax/saxutils.py", line 141, in startElement
    self._write(' %s=%s' % (name, quoteattr(value)))
  File "/home/serhiy/py/cpython-3.Lib/xml/sax/saxutils.py", line 96, in _write
    self._out.write(text)
  File "/home/serhiy/py/cpython-3.Lib/codecs.py", line 699, in write
    return self.writer.write(data)
  File "/home/serhiy/py/cpython-3.Lib/codecs.py", line 355, in write
    data, consumed = self.encode(object, self.errors)
UnicodeEncodeError: 'latin-1' codec can't encode character '\u20ac' in position 7: ordinal not in range(256)

And shouldn't. On Python 2 XMLGenerator works only with binary files and "works" with text files only due implicit str->unicode converting. On Python 3 working with binary files was broken. Issue1470548 restores working with binary file (for which only XMLGenerator can work correctly), but for backward compatibility accepting of text files was left. The problem is that there no trustworthy method to determine whenever a file-like object is binary or text.

Accepting of text streams in XMLGenerator should be deprecated in future versions.

> Accepting of text streams in XMLGenerator should be deprecated in future versions.

I agree that the following pattern is strange:

with codecs.open('/tmp/test.txt', 'w', encoding='iso-8859-1') as f:
   xml = XMLGenerator(f, encoding='iso-8859-1')

Why would I specify a codec twice? What happens if I specify two
different codecs?

with codecs.open('/tmp/test.txt', 'w', encoding='utf-8') as f:
   xml = XMLGenerator(f, encoding='iso-8859-1')

It may be simpler (and safer?) to reject text files. If you cannot
detect that f is a text file, just make it explicit in the
documentation that f must be a binary file.

2013/5/7 Serhiy Storchaka <report@bugs.python.org>:
>
> Serhiy Storchaka added the comment:
>
> It is not working fine on Python 3.3.0.
>
>>>> with codecs.open('/tmp/test.txt', 'w', encoding='iso-8859-1') as f:
> ...     xml = XMLGenerator(f, encoding='iso-8859-1')
> ...     xml.startDocument()
> ...     xml.startElement('root', {'attr': u'\u20ac'})
> ...     xml.endElement('root')
> ...     xml.endDocument()
> ...
> Traceback (most recent call last):
>   File "<stdin>", line 4, in <module>
>   File "/home/serhiy/py/cpython-3.Lib/xml/sax/saxutils.py", line 141, in startElement
>     self._write(' %s=%s' % (name, quoteattr(value)))
>   File "/home/serhiy/py/cpython-3.Lib/xml/sax/saxutils.py", line 96, in _write
>     self._out.write(text)
>   File "/home/serhiy/py/cpython-3.Lib/codecs.py", line 699, in write
>     return self.writer.write(data)
>   File "/home/serhiy/py/cpython-3.Lib/codecs.py", line 355, in write
>     data, consumed = self.encode(object, self.errors)
> UnicodeEncodeError: 'latin-1' codec can't encode character '\u20ac' in position 7: ordinal not in range(256)
>
> And shouldn't. On Python 2 XMLGenerator works only with binary files and "works" with text files only due implicit str->unicode converting. On Python 3 working with binary files was broken. Issue1470548 restores working with binary file (for which only XMLGenerator can work correctly), but for backward compatibility accepting of text files was left. The problem is that there no trustworthy method to determine whenever a file-like object is binary or text.
>
> Accepting of text streams in XMLGenerator should be deprecated in future versions.
>
> ----------
>
> _______________________________________
> Python tracker <report@bugs.python.org>
> <http://bugs.python.org/issue17915>
> _______________________________________

Here is a patch which adds explicit checks for codecs stream writers and adds tests for these cases. The tests are not entirely honest, they test only that XMLGenerator works with some specially prepared streams. XMLGenerator doesn't work with a stream with arbitrary encoding and errors handler.

Of course, if this patch will be committed, perhaps it will be worth to apply it also for 3.2 which has the same regression.

Perhaps we should add a deprecation warning for codecs streams right in this patch?

New changeset 1c01571ce0f4 by Georg Brandl in branch '3.2':
Issue #17915: Fix interoperability of xml.sax with file objects returned by
http://hg.python.org/cpython/rev/1c01571ce0f4

Fixed in 3.2, 3.3 and default.

thanks everybody !