Why are "broken iterators" broken?

Cameron Simpson cs at zip.com.au
Mon Sep 22 06:32:17 CEST 2008


On 21Sep2008 18:36, Fredrik Lundh <fredrik at pythonware.com> wrote:
> Roy Smith wrote:
>> There are plausible examples of collections which grow while you're  
>> iterating over them.  I'm thinking specifically of a queue in a  
>> multi-threaded application.  One thread pushes work onto the back of 
>> the queue while another pops from the front.  The queue could certainly 
>> go empty at times.  But, maybe a Python iterator is just the wrong way 
>> to model such behavior.
>
> you probably want the consumer thread to block when it catches up with  
> the producer, rather than exit.
> (that's the default behaviour of Python's Queue object, btw)

It sounds like he wants non-blocking behaviour in his consumer.
A common example is "try to gather a lot of stuff into a single packet,
but send a smaller packet promptly if there isn't much stuff".

You could make the next() method return a sentinal value like None
when the queue is empty. That would mean your consumer must recognise
the special value and also precludes delivering that value through the
queue. I'm not convinced my suggestion here is any better than just
doubling up every call to next() with an empty() check immediately
beforehand.

You could write a trivial wrapping generator to take the original
blocking queue and return a sentinel value on empty, too.

My suggestion is also an excellent way of getting programs that
fail-busy (i.e. they spin out) if you make a logic error in your
consumer. Ouch.

Cheers,
-- 
Cameron Simpson <cs at zip.com.au> DoD#743
http://www.cskk.ezoshosting.com/cs/

Kill, v.t.  To create a vacancy without nominating a successor.
Ambrose Bierce (1842-1914), U.S. author. The Devil's Dictionary (1881-1906).



More information about the Python-list mailing list