Message 119888 - Python tracker (original) (raw)

Antoine Pitrou wrote:

Antoine Pitrou <pitrou@free.fr> added the comment:

That's what I'm referring to: most Python applications are written with the fact in mind, that garbage collection will close the files or socket.

That's a perfectly fine way of writing Python applications,

Some people would disagree, especially Windows users who cannot timely delete files when some file descriptors still point to them.

There is no difference here:

Whether you write an application with automatic closing of the file/socket at garbage collection time in mind, or you explicitly close the file/socket, the timing is the same.

The only difference is with Python implementations that don't use synchronous garbage collection, e.g. Jython, but not with CPython.

so why should the programmer get warned about this regular approach to Python programming ?

Again: it is an optional warning. It is disabled by default, except when compiled --with-pydebug.

Yes, I know. I still find the warning rather useless, since it warns about code that's perfectly ok.

The same applies for sockets.

It is definitely a mistake if the socket has been bound to a local address and/or connected to a remote endpoint.

I don't follow you. Where's the difference between writing:

s.close() or s = None

for an open socket s ?

Think of the simple idiom:

data = open(filename).read()

This would always create a warning under the proposal.

We have had many Windows buildbot failures because of such coding style.

Again: where's the difference between writing:

data = open(filename).read() and f = open(filename) data = f.read() f.close()

If the above coding style causes problems, the reasons must be elsewhere, since there is no difference between those two styles (other than cluttering up your locals).

The for-loop file iterator support was explicitly added to make writing:

for line in open(filename): print line

possible.

If you want to monitor resource usage in your application it would be a lot more useful to provide access to the number of currently open FDs

Agreed it would be useful as well, but please tell that to operating system vendors. Python has no way to calculate such a statistic.

At least for Linux, that's not hard and I doubt it is for other OSes.

On other Unixes, you can simply use fcntl() to scan all possible FDs for open FDs.

On Windows you can use one of these functions for the same effect: http://msdn.microsoft.com/en-us/library/kdfaxaay(v=VS.90).aspx