Base class for file-like objects? (a.k.a "Stream" in Java)

tleeuwenburg at gmail.com tleeuwenburg at gmail.com
Wed Jul 25 09:00:01 CEST 2007


On Jul 25, 12:28 pm, Asun Friere <afri... at yahoo.co.uk> wrote:
> On Jul 25, 8:51 am, Boris Dušek <boris.du... at gmail.com> wrote:
>
> > In Python, I found that "file" objects exist. While specifying
> > argument types in Python is not possible as in Java, it is possible to
> > check whether an object is an instance of some class and that's what I
> > need - I need to check if an argument is a "file"-like object, and if
> > yes, behave accordingly, if not, treat the argument as string with
> > URL.
>
> > But I am afraid there is no such a base class - I tried the following:
>
> Yup you are right.  Look.>>> type(g).mro()
>
> [<type 'instance'>, <type 'object'>]>>> type(f).mro()
>
> [<type 'file'>, <type 'object'>]
>
> "File-like object" refers the the behaviour of the object, not it's
> inheritance.  Python is an equal opportunity employer, we don't ask
> and object what race it is, we simply ask "can you do the job?"  This
> is known colloquially as "duck-typing."  In other words Python has
> 'polymorphism by interface,' in contradistinction to Java's
> 'polymorphism by implementation.'  Until you understand this you will
> be trying to write Java in Python, and you will not have much joy.
>
> > P.S.: The code should finally look in esence something like this:
>
> > if isinstance(f, file):
> >    pass
> > elif isinstance(f, string):
> >    f = urllib.urlopen(f)
> > else:
> >    raise "..."
> > process_stream(f)
>
> Because of duck-typing, you should almost never use the isintance()
> (or type(), hasattr(), or any others that don't immediately come to
> mind ...) in actual code. It (usually) breaks Python's polymorphism!
> If you find these methods popping up in your code it strongly
> indicates that you should be using a try/except statement, if not a
> complete change of your code's logic.  Search for LBYL (look before
> you leap) vs EAFP (easier to ask forgiveness than permission) for a
> full explanation.
>
> I'm not sure why you would ever be sending a file object to urlopen
> (or is the test isinstance(f, file) supposed to test for an already
> opened url?), but you final code should actually look more like
> something along these lines:
>
> try :
>   f = urlopen(f)
> except AttributeError :
>   pass
>
> This is not as elegant as it could be, as it will pass not only on
> open files (or urls), but on any type that lacks the .strip method
> (ie. it doesn't account for your else condition).  You'd probably have
> to catch an exception later when you try to use what should be an open
> url, or rewrite your code not to get in this position in the first
> place.  But however you refactor your code, it is advisable to
> concentrate on what an object can do (and catch an exception where it
> fails to perform), rather on the type of the object.

I agree that using try statements is often, perhaps usually, a good
idea. That said...

One good reason to avoid using a try: except: methodology is if the
actions taken in the try block could potentially modify data
structures or the underlying data during processing, or if the action
to be tried is expensive.

Being able to check before taking an action that it will be safe can
be important.

In these cases, you can use isinstance. Python 3k, I believe, is
thinking of being able to implement Abstract Base Classes for this
very reason.

As things stand, however, you will not be able to catch all cases with
a simple isinstance call. Try except should work in almost all cases.
If you want to do your own duck-typing, you can use dir(object) to get
its methods and check whether what you need is in there.

Anyway, good luck with it.




More information about the Python-list mailing list