[Zope-dev] dtml-in performance

[Joachim Werner]

> > What can I do to maximize the peformance of a dtml-in statement?
> > When iterating over a 2000 row database query, it takes about 20
> > seconds - time mostly spent doing security checks, and calling
> > __getitem__ in DT_InSV.py.  Is the performance of dtml-in just slow?
> > -Brett
> >
>
> Another performance consideration with looping in DTML is that whatever
> is in your DTML block is reinterpreted each time through. Although it is
> parsed (hopefully) only the first time through, this overhead is still
> considerable when multiplied over 2000 rows. You are essentially using
> an interpreter (Python) as an interpreter for DTML. I would seriously
> consider moving this entire operation (query, iteration and html
> generation) to native Python if performance is a big consideration.

We have had similar performance problems with the first version of our
Content Management demo. It used both dtml-in loops and ZClasses (with other
ZCLasses as base classes) heavily. The combination of the both can be very
efficient in slowing down Zope. I remember the first versions of Martijn
Faassen's XML-Widgets that had a similar performance problem. The reason is
that in addition to the dtml-in overhead ZClasses take a lot of time for
acquiring from their containers and inheriting from their "base classes" as
all the inheritance stuff is only simulated - in fact almost everything is
just stored in property sheets and has to be processed by the security
engine if I got it right.

In Python these problems go away. Maybe it is time for a good
reimplementation of ZClasses, inlcuding the creation of REAL Python classes
in the filesystem on  the fly.

BTW: Does anybody know if the Zope Page Templates (ZPT) parser is faster
with looping than the DTML one?

Cheers,

Joachim.