On Thu, Dec 20, 2012 at 3:47 AM, Glyph <glyph@twistedmatrix.com> wrote:

On Dec 19, 2012, at 2:14 PM, anatoly techtonik <techtonik@gmail.com> wrote:

On Sun, Dec 9, 2012 at 7:14 AM, Glyph <glyph@twistedmatrix.com> wrote:
On Dec 7, 2012, at 5:10 PM, anatoly techtonik <techtonik@gmail.com> wrote:

What about reading from other file descriptors?  subprocess.Popen allows arbitrary file descriptors to be used.  Is there any provision here for reading and writing non-blocking from or to those?

On Windows it is WriteFile/ReadFile and PeekNamedPipe. On Linux it is select. Of course a test is needed, but why it should not just work?

This is exactly why the provision needs to be made explicitly.

On Windows it is WriteFile and ReadFile and PeekNamedPipe - unless the handle is a socket in which case it needs to be WSARecv.  Or maybe it's some other weird thing - like, maybe a mailslot - and you need to call a different API.

IIRC on Windows there is no socket descriptor that can be used as a file descriptor. Seems reasonable to limit the implementation to standard file descriptors in this platform.


hFile [in]
A handle to the device (for example, a file, file stream, physical disk, volume, console buffer, tape drive, socket, communications resource, mailslot, or pipe). (...) For asynchronous read operations, hFile can be any handle that is opened with the FILE_FLAG_OVERLAPPED flag by the CreateFilefunction, or a socket handle returned by the socket or accept function.

(emphasis mine).

So, you can treat sockets as regular files in some contexts, and not in others.  Of course there are other reasons to use WSARecv instead of ReadFile sometimes, which is why there are multiple functions.

handle != descriptor, and Python documentation explicitly says that socket descriptor is limited, so it's ok to continue not supporting socket descriptors for pipes.
http://docs.python.org/2/library/socket.html#socket.socket.fileno
 
On *nix it really shouldn't be select.  select cannot wait upon a file descriptor whose value is greater than FD_SETSIZE, which means it sets a hard (and small) limit on the number of things that a process which wants to use this facility can be doing.

I didn't know that. Should a note be added to http://docs.python.org/2/library/select ?

The note that should be added there is simply "you should know how the select system call works in C if you want to use this module".

Why spreading FUD if it is possible to define a good entrypoint for those who want to learn, but don't have enough time? Why not to say directly that select interface is outdated?
On the other hand, if you hard-code another arbitrary limit like this into the stdlib subprocess module, it will just be another great reason why Twisted's spawnProcess is the best and everyone should use it instead, so be my guest ;-).

spawnProcess requires a reactor. This PEP is an alternative for the proponents of green energy. =)

Do you know what happens when you take something that is supposed to be happening inside a reactor, and then move it outside a reactor?  It's not called "green energy", it's called "a bomb" ;-).

The biggest complain about nuclear physics is that to understand what's going on it should have been gone 3D long ago. =) I think Twisted needs to organize competition on the best visualization of underlying concepts. It will help people grasp the concepts behind and different problems much faster (as well as gain an ability to compare different reactors).