RE: Parser torture documents

["Arnold, Curt" <Curt.Arnold@hyprotech.com>]

I'd also add SVG with the svg.dtd as a decent torture test due
the svg.dtd's use of parameter entities.

Last time I checked it broke Xerces-C (but I have been a dog and haven't submitted a bug or fix.

Definitely XMI is a edge case and one that goes against most
reasonable DOM optimizations.  If you have never looked at XMI,
it loves to use ludicrously long tag names, such as:

<Foundation.Core.Classifier.abstract xmi.value="true"/>

Since tag names in most documents are short, trying to 
share references to tag names may not be worth the hit
to find the match.  However, for XMI, it would definitely
be beneficial.

I'd definitely be careful about using non-representative
documents as performance benchmarks.  You don't want to 
encourage the parser writers to optimize for oddball cases
at the expense of more normal cases.