Re: [Webware-devel] Re: [Cheetahtemplate-discuss] 2.0 bugs or warts?

[Ian B. <ia...@co...>]

Chuck Esterbrook wrote:
> I'm working strictly off memory here, but I thought the point of the
> implementation was that it used a special library (when available)
> that eliminated the need for polling. Instead, the op sys would notify
> the appserver process when a file changed.
> 
> Does that sound familiar?

I'm not sure that was actually the original intent, but yes, that is an 
advantage.  OTOH, you could look at sys.modules.keys() once a second to 
test for new files without any substantial overhead.

> Polling has the well known disadvantages of (1) the gap between polls
> (too big? too small? never just right), (2) wasted effort checking for
> something repeatedly and in this case, (3) the increased amount of
> time required per poll as your app grows larger and uses more modules.
> 
> Although, I don't see why we couldn't have an option to use one
> technique vs. another. We could even default it to polling sys.modules
> since that seems less likely to disturb other Python packages like
> Cheetah.

Especially since most people don't have the non-polling libraries 
installed (e.g., famd), this would be safer all around.

-- 
Ian Bicking  |  ia...@co...  |  http://blog.ianbicking.org