iterating over the lines of a file - difference between Python 2.7 and 3?

[Peter Otten]

Wolfgang Maier wrote:

> I just came across an unexpected behavior in Python 3.3, which has to do
> with file iterators and their interplay with other methods of file/IO
> class methods, like readline() and tell(): Basically, I got used to the
> fact that it is a bad idea to mix them because the iterator would use that
> hidden read-ahead buffer, so what you got with subsequent calls to
> readline() or tell() was what was beyond that buffer, but not the next
> thing after what the iterator just returned.
> 
> Example:
> 
> in_file_object=open(‘some_file’,’rb’)
> 
> for line in in_file_object:
> 
>                 print (line)
> 
>                 if in_file_object.tell() > 300:
> 
>                                # assuming that individual lines are
>                                # shorter
> 
>                                break
> 
>  
> 
> This wouldn´t print anything in Python 2.7 since next(in_file_object)
> would read ahead beyond the 300 position immediately, as evidenced by a
> subsequent call to in_file_object.tell() (returning 8192 on my system).
> 
> However, I find that under Python 3.3 this same code works: it prints some
> lines from my file and after completing in_file_object.tell() returns a
> quite reasonable 314 as the current position in the file.
> 
> I couldn´t find this difference anywhere in the documentation. Is the 3.3
> behavior official, and if so, when was it introduced and how is it
> implemented? I assume the read-ahead buffer still exists?

You can get the Python 3 behaviour with io.open() in Python 2.7. There is an 
implementation in Python in _pyio.py:

    def tell(self):
        return _BufferedIOMixin.tell(self) - len(self._read_buf) + 
self._read_pos


> By the way, the 3.3 behavior only works in binary mode. In text mode, the
> code will raise an OSError:  telling position disabled by next() call. In
> Python 2.7 there was no difference between the binary and text mode
> behavior. Could not find this documented either.