[Patrick]
> Ouch! Any idea how much overhead the ptl stuff adds?

On the surface, I wouldn't expect PTL's runtime to cost very much.  It's
just Python with a different way of accumulating and returning strings
from functions.  But I could be wrong...

[Neil]
> It costs some.  Here are the numbers for a version of the demo that
> doesn't use ptl:
>
>     Requests per second:    10.54 [#/sec] (mean)
>
> With ptl the demo only did about 6 hits per second.

Hmmm!  That surprises me.

> The ptl import hooks make '__import__' much slower.  My cimport
> extension module fixes that problem but it hasn't been tested too much
> yet.

My intuition tells me that importing is worse for performance than PTL's
runtime stuff.  Neil, did you run that benchmark with cimport or with
ihooks?

I've been running with cimport for the past couple of days, and I've
been doing everything through plain CGI (otherwise I'd be restarting the
SCGI daemon on every hit -- development is like that).  Anyways, that
means I've been doing boatloads of imports through cimport, with zero
problems.  It's nice that import-time tracebacks are a lot shorter too.
;-) Why don't we drop cimport.so into our /www/plat and let everyone,
including our live site, bash on it?  That should build confidence.

        Greg