- Only minimize empty elements when the content model is empty,
  keeping an explicit list of which elements have an empty content
  model (e.g. <img>, etc.).  This checks whether the tagname is in an
  explicit list of such elements; the check is case insensitive.

- Minimize certain attributes that are considered Boolean by HTML,
  e.g. ismap.  Again, this does a case-insensitive check whether the
  attribute name is in a list of known Boolean attributes.

Note: this should really be refactored, making HTML support a subclass
rather than a keyword argument.  Later...

- metal to do macro expansion (default on)

- tal to do TAL expansion (default on)

- html for XHTML heuristics (default off).  See
  http://www.w3.org/TR/xhtml1/#guidelines for what this should do;
  currently all it does is to suppress the XML declaration and insert
  a space before the /> in minimized elements.

Changed the TALCompiler class to generate enough code to be able to
implement the tal=0 and metal=0 options to TALInterpreter; this meant
e.g. compiling the original contents of nodes subject to replacement.

TALVisitor.  Also added wrap=0 to the TALInterpreter call, so the
regression test isn't affected by the new wrapping code.

aligns them with the end of the tag.  This required introducing
self.write_stream() which keeps track of the current output column.

The default wrap column is 60.  Specifying 0 disables wrapping (good
for the regression test).

preserved.

them froM TALVisitor into the other modules.

class constructor, so a caller can pass in an expression compiler.
Every expression occurring in the syntax is compiled first (even the
macro name in use-macro; but not slot names, nor the macro name in
define-macro).  If the expressionCompiler argument is left
unspecified, a default compiler is chosen whose compiled output is
identical to its input.

  value.  This affects TALCompiler.py and TALInterpreter.py, and
  affects the output of the non-local-macro expansion tests.  The
  TALVisitor.py module is *not* yet fixed to do this.

- In order to do this, I had to change the "program code" format for
  the attribute list: the 3rd item of an attribute description tuple,
  if present, now names an action, and the remaining tuple items are
  arguments for that action.  Two actions are recognized: "replace"
  replaces the attribute with the outcome of evaluating the expression
  given by the next tuple  item; "macroHack" replaces the attribute
  name with "use-macro" and its value with the current macro, but
  only if we're inside macro expansion and the attribute name ends in
  ":define-macro".

- The test suite is now error-free for "./runtest.py -c" but shows some
  errors for "./runtest.py".

- Added a primitive debugging flag to the TALInterpreter class.

- Update the README.

- Should I check for spurious TAL attributes even when doing just
  METAL (macro) expansion?

- Should I turn all the other errors into exceptions too?  (Currently
  they are all print statements.)

lock-refresh request.

to minidom.  (Note that at the moment, only the latest Python CVS
version of minidom works; 2.1b1 will work too when it is released, but
2.1a2 or anything before won't.)

- Lose redundant semicolon.

- Rearrange pickle imports.

built and a dictionary was being found where a list was expected (the
dictionary was in the form of {'rolename':1}, used to set the default
roles for a permission.  Changed the line to get the keys of the
dictionary).

Document running the regression test.

Add TO DO list.