Issue1350

In python 2.5.4: 

Python 2.5.4 (r254:67916, Apr 24 2009, 09:21:35) 
[GCC 4.3.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys
>>> type(sys.stdin.fileno())
<type 'int'>
>>> type(sys.stdout.fileno())
<type 'int'>
>>> type(sys.stderr.fileno())
<type 'int'>
>>> print sys.stdin.fileno()
0
>>> print sys.stdout.fileno()
1
>>> print sys.stderr.fileno()
2

In Jython trunk:
Jython 2.5rc2+ (trunk:6353M, maj 17 2009, 10:37:03) 
[Java HotSpot(TM) 64-Bit Server VM (Sun Microsystems Inc.)] on 
java1.6.0_13
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys
>>> type(sys.stdin.fileno())
<type 'org.python.core.io.StreamIO'>
>>> type(sys.stdout.fileno())
<type 'org.python.core.io.StreamIO'>
>>> type(sys.stderr.fileno())
<type 'org.python.core.io.StreamIO'>
>>> print sys.stdin.fileno()
org.python.core.io.StreamIO@68fc8e75
>>> print sys.stdout.fileno()
org.python.core.io.StreamIO@7afccada
>>> print sys.stderr.fileno()
org.python.core.io.StreamIO@f8600d6

This is intentional as Java doesn't provide us direct access to real int 
fds

Emulating them as ints is possible but doesn't seem worth the effort, at 
least for the moment. The only real apparent issue is the std 
descriptors 0/1/2 are sometimes passed around. We've fixed os.isatty to 
recognize them and handle them appropriately, but some other functions 
may need similar workarounds

Does our current implementation actually break anything? All the related 
stdlib tests should be passing. If so, please reopen this and provide a 
specific use case (preferably a test)

So far it doesn't break anything :-) But I am trying to implement
termios module using jna under the hood, and because linux/unix termios
functions need an int file descriptor in order to work, I'm stuck. Since
fds are not going to be changed, I could use an advice on how can I
recognize if given jython fd is generated by sys.stdin.fileno(), etc, so
I could call termios methods with appropriate int fd.

We can provide access to the real FDs in some environments. They are 
available via a private field in java.io.FileDescriptor (at least on Sun 
and IIRC other common JVMs). You can circumvent the private field's 
protection via reflection unless the java security policy disallows you 
from doing so. I believe JRuby is doing this

If you just need to work on the std streams, couldn't you check if the 
passed in fd in (fp.fileno() for fp in (sys.stdin, sys.stdout, 
sys.stderr))? Then map them to 0/1/2 accordingly? At least for the 
meantime, anyway. As we're not going to change fileno() until after the 
upcoming 2.5 release