Adding a header also for each open call is a frequent
need for me (much more than e.g. proxies per call in
urllib1/urlopen). 

For example:

urlopen(url,data,addheaders=[('Referer':lasturl)])

So far one is forced to set the _complete_ re-rendered
bunch of opener.addheaders each time (and to maintain a
separate opener for each thread )

open/OpenerDirector.open maybe should distribute
*args,**kwargs

---

Note on disciplined use of protocol modules in urllib2
( see also #1046077 ):

urllib2 still draws in all kind of (future?) protocol
modules all in advance =>  slows down app startup / cgi
script's ...; ugly non-pythonic/non-modular; unused
prots cannot be excluded for cx_Freeze,setup ... 
Now the fat cookielib was also added in that manner.
Thus, when you use urllib2 for "ftp://xy", you load
also all kind of fun (down to things like
_MozillaCookieJar) into Python. 

The need for those imports is very local (search e.g.
"ftplib."). ==> local import in those few places.

I saw, this jam style already encouraged more
undisciplined dependencies like :
"origin_req_host = cookielib.request_host(self)"
(~urllib2-py2.4/line 191) in the general Request.__init__ 


Robert

Logged In: YES 
user_id=341410

Whether or not urllib2 imports generally unneeded modules is
a different matter than whether inserting headers should be
easier.  In the future, please separate feature requests
from random chatter about a module.

I'm personally for making it easier to include alternate
headers, though am ambiguous as to your suggested change.