I originally downloaded this code from http://www.xml-benchmark.org a few years
ago. That site is no longer active, so I've posted the code here, as is.
It's copyrighted (C) by Florian Waas. See the original contents of the README
below the build instructions.

The tool is described in a paper from 2002, a copy of which is also in this
repo.

Update from March 2022: a kind reader informed me that the source for xmlgen
can also be obtained from https://projects.cwi.nl/xmark/generator.html

----

./build.sh

After that run './xmlgen'

./xmlgen -f 1     produces ~116 MiB
./xmlgen -f 0.5   produces 58 MiB

----

xmlgen, version 0.92 by Florian Waas (flw@mx4.org)
Copyright (C) Florian Waas


1. What is xmlgen?

xmlgen is an XML data generator which produces scaled documents according
to the DTD specified in The XML Benchmark Project. xmlgen is part of the
benchmark suite and can be found at http://www.xml-benchmark.org. It has
been one of the major design goals to achieve a maximum degree of portability
and to date, xmlgen has been used on a number of platforms including Windows,
Solaris, various Linux distributions, and IRIX.
xmlgen was designed to produce large and very large XML documents in an
efficient manner with low constant main memory requirements.


2. How to use xmlgen?

xmlgen comes with a number of options to influence the output behavior:

 -f <factor>    scaling factor of the document, float value;
		0 produces the "minimal document"

 -o <file>	direct output to file

 -h 		show usage info

 -d 		use doctype preamble

 -i 		renders the document somewhat more readable

 -v 		shows current version, intended for bug reporting

 -t 		display elapsed time, meant for profiling

 -s <cnt>	split the doc in smaller chunks of only <cnt> elements each;
		useful for systems which cannot cope with large input 
		documents

 -e		dumps the DTD the doc is complying with (version 0.92 and later)


3. Why is there no noise in the text?

Well, it's Shakespeare. In fact, the text has only little noise and many
text indexing programs seem to be a little baffled by that. We plan to
change this in a future release together with a more contemporary vocabulary. 
Also, Shakespeare is not quite politically correct.


4. Can I control the level of recursion?

No, we purposely reduced the number of tuning parameters to only one single
one: the scaling factor. Otherwise, the space of possible combinations is 
growing too quickly.