[Patches] [ python-Patches-487458 ] bugfix and extension to Lib/asyncore.py

noreply@sourceforge.net noreply@sourceforge.net
Fri, 22 Nov 2002 00:51:07 -0800


Patches item #487458, was opened at 2001-11-30 20:14
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=305470&aid=487458&group_id=5470

Category: Library (Lib)
Group: None
Status: Open
Resolution: Later
Priority: 5
Submitted By: Hye-Shik Chang (perky)
Assigned to: Jeremy Hylton (jhylton)
Summary: bugfix and extension to Lib/asyncore.py

Initial Comment:
fix: asyncore of 2.2b2+ fails to initialize when passed
'sock' is listening.
extension(or bugfix?): more flexible handling own map

----------------------------------------------------------------------

>Comment By: Hye-Shik Chang (perky)
Date: 2002-11-22 17:51

Message:
Logged In: YES 
user_id=55188

No. All problems are resolved right after 2.2a3. Thank you!


----------------------------------------------------------------------

Comment By: Martin v. L÷wis (loewis)
Date: 2002-11-22 17:26

Message:
Logged In: YES 
user_id=21627

Is this patch still needed? If yes, can someone provide a
test case of an example that breaks without this patch but
works with the patch?



----------------------------------------------------------------------

Comment By: Hye-Shik Chang (perky)
Date: 2001-12-15 01:25

Message:
Logged In: YES 
user_id=55188

Hmm. thanks for your appreciation. ;)
I'm using this way for non-connected sockets.

class asyncore_accepting_mixin:
    def __init__(self, sock=None, map=None):
        if map is not None:
            self.map = map
        asyncore.dispatcher.__init__(self, sock, map)
        self.connected = 0
        self.accepting = 1

I call asyncore.dispatcher.__init__ for adapting any
version of asyncore.
I thought this way:

def __init__(.... of subclass or mixin..

try:
    asyncore.dispatcher.__init__(self, sock, map)
except socket.error, why:
    if why[0] != ENOTCONN:
        raise socket.error, why

but, this can be reasonable only if sock.getpeername() is on
last line of __init__. So, subclass can't use 
dispatch.__init__ clearly on this condition.

Anyway, thanks for your help and commit. ;)

----------------------------------------------------------------------

Comment By: Jeremy Hylton (jhylton)
Date: 2001-12-15 00:21

Message:
Logged In: YES 
user_id=31392

The dispatcher constructor sets connected to 1 when you pass
it a socket.  This suggests to me that the contract is that
you only pass a connected socket to dispatcher.  Is this
what you're doing?

If not, perhaps you need to create a subclass of dispatcher
that behaves appropriately for non-connected sockets.


----------------------------------------------------------------------

Comment By: Jeremy Hylton (jhylton)
Date: 2001-12-15 00:14

Message:
Logged In: YES 
user_id=31392

I see.  It was hard to tell what was bug and what was
feature.    It does sound like you've identified a bug
introduced during 2.2 development.  I'll see if it's not too
late.


----------------------------------------------------------------------

Comment By: Hye-Shik Chang (perky)
Date: 2001-12-14 13:39

Message:
Logged In: YES 
user_id=55188

But, failure of sock.getpeername() is fatal for me.
It interrupts initializing dispatcher class on programs 
which runs perfectly on all of past versions of Python.

Without this patch, difference between 1.19 and 1.20 of 
asyncore.py must be rollbacked. (It is not important feature
but fancy for __repr__)

----------------------------------------------------------------------

Comment By: Jeremy Hylton (jhylton)
Date: 2001-12-14 04:09

Message:
Logged In: YES 
user_id=31392

I'm not comfortable making changes like this with only the
release candidate for testing.  I think we should postpone
until 2.3.


----------------------------------------------------------------------

Comment By: Hye-Shik Chang (perky)
Date: 2001-12-05 14:09

Message:
Logged In: YES 
user_id=55188

Oops! sorry. It was a mistake.
changed to corrected patch.

----------------------------------------------------------------------

Comment By: Hye-Shik Chang (perky)
Date: 2001-12-04 13:14

Message:
Logged In: YES 
user_id=55188

On the ENOTCONN error:
 the error will be raised only if 'sock' was bound outside.
(even if self.accepting get set)
 but using out-bound socket is unavoidable for some server 
models. and I and some co-workers are using it.
 How about this way instead of previous patch?

 if not self.accepting:
     self.addr = sock.getpeername()

 
On passing maps:
 some part of asyncore supports using own socket map.
 (__init__, add_channel, del_channel), but when
 __init__ registers socket into map which passed by
 argument, asyncore can't del_channel from own map and
 it will raises KeyError because asyncore.socket_map
 doesn't have such fileno.
 And, current version of asynchat doesn't support
 to use another maps.


----------------------------------------------------------------------

Comment By: Martin v. L÷wis (loewis)
Date: 2001-12-02 20:03

Message:
Logged In: YES 
user_id=21627

On the ENOTCONN error: Shouldn't self.accepting get set if
the socket is listening? But then, the socket may not be
even listening?

On passing maps: Can you please explain why you use self.map
if it is None? I had expected that you use socket_map in
this case.



----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=305470&aid=487458&group_id=5470