In Python 2.4.2 and trunk, when StreamReader.readline
is used with a size argument, it will eventually
truncate some lines, even if every line is less than
the size.  It works correctly in Python 2.3.5, I
haven't tried other versions.

>>>
f=codecs.getreader('ascii')(StringIO.StringIO("hello\nworld\n"))
>>> f.readline(8)
u'hello\n'
>>> f.readline(8)
u'wo'
>>> f.readline(8)
u'rld\n'

I've attached a patch which fixes this and modifies
test_codecs.py to test this.  I don't know if this is
the best way to fix it, but I didn't want to make
deeper changes to the caching mechanism, being
unfamiliar with this code.