There are three (minor?) problems with raising
ValueError. 

1) Compatibility.  I had some code for 1.5.2 that
was trying to load web pages checking for various
errors, and it was expecting this error to cause
a socket error, not a value error. 

2) Accuracy.  ValueError can be caused by
anything.  The 'non-numeric port' error is much 
more informative.  I don't want to catch
ValueError, because it can be caused in too 
many situations.  I also cannot check 
myself that the port is fine, because the 
port and the URL are often given by a  
redirect (errors 301 and 302, if I remember
correctly).  This in fact was the situation 
that caused the problem. 
Hence, my only real solution was to patch my version of httplib. 

3) Style.  I am somewhat new to Python, but I was
under the impression that, stilistically, 
a ValueError was used to convey a situation that
was the fault of the programmer, while other 
more specific errors were used for unexpected 
situations (communication, etc).  Since the 
socket is the result of a URL redirection 
(errors 301 or 302), the programmer is not in 
a position to prevent this error by "better 
checking".  Hence, I would consider a
network-relted exception to be more appropriate 
here. 

But who am I to argue with the creator of Python? 
;-)

Luca