[Python-Dev] asyncore fixes in Python 2.6 broke Zope's version of medusa

[glyph at divmod.com]

On 10:01 am, greg.ewing at canterbury.ac.nz wrote:
>Hrvoje Niksic wrote:
>>        Under Linux, select() may report a socket file descriptor
>>        as "ready for reading",  while  nevertheless
>>        a subsequent read blocks.
>
>Blarg. Linux is broken, then. This should not happen.

You know what else is broken?  MacOS, FreeBSD, Solaris, and every 
version of Windows.  I haven't tried running Twisted on AmigaOS but I 
bet it has some problems too.

On occasion Linux has been so badly broken that Twisted has motivated a 
fix.  For example, 
http://lkml.indiana.edu/hypermail/linux/kernel/0502.3/1160.html

But even if we ignore difficulties at the OS level (which should, after 
all, be worked around rather than catered to in API design) there are 
other good reasons why the general async API should be fairly distant 
from both the select/poll wrapper and the questions of blocking vs. non- 
blocking sockets.  For another example, consider the issue of non- 
blocking SSL sockets.  Sometimes, in order to do a "read", you actually 
need to do a write and then another read.  Which means that application 
code, if it wants to be compatible with SSL, needs to deal with any 
failure that may come from a write as coming from a read, unless you 
abstract all this away somehow.