Venus thoughts

[Sam Ruby]

Jeff Waugh wrote:
> <quote who="Sam Ruby">
> 
>> For memory constrained environments, XSLT may turn out to be be preferred.
> 
> Wow, really? Why is that? (Seems counter-intuitive.)

Does seem counter intuitive, but libxml2 and related components are 
blazingly fast.  Also, I'm starting from a normalized XML representation 
that I simply concatenate and pass to XSLT.  For HTMLTMPL, I need to 
build a second dictionary based representation of the input data 
structures.  Finally, that data structure must contain everything that 
might be referenced, including dates in several formats.

Disclaimer: I have not measured memory consumption.  All I know is that 
on my machine, there is a small but human noticeable pause between 
templates invocations when HTMLTMPL is used; by contrast, there is no 
perceptible delay between template invocations when XSLT is used.

As a final footnote: libxml2 is noticeably faster than the default 
minidom implementation, but isn't installed by default in Python, so 
I've stuck with mindom.  I did prototype building an encapsulation layer 
so that Venus would use libxml2 when it was available, but fall back to 
minidom.  On the test I tried it with, there was about a 33% reduction 
in elapsed time.  This may merit additional exploration, but I didn't 
commit that change at the time.

It may very well turn out that no xml parser is required at all in the 
splice phase -- everything is normalized, so I should be able to simply 
strcat the entries and then wrap them in a feed.

- Sam Ruby