other versions of expat that might pop up elsewhere.

Eventually, this version needs to go away.

end.  This would have caught the problem exposed in test10.html
earlier.

To prevent this in the future, read the path from a file .path.

Properly pop the interpreter level even when an exception happens.

Add a few asserts to ensure that the scope level and the interpreter
level are correctly maintained.

Properly implement tal:replace="structure ..." combined with
tal:attributes="...", both in HTML and in XML mode.  This is pretty
tricky, and requires instantiating a parser with a modified code
generator which suppresses any TAL/METAL interpretation but does the
attribute replacement on the first encountered start tag as required
by the TAL 1.0 specs.

that are now generated.

DummyCompiler class), and make sure to call it everywhere it is
needed (in particular for attribute replacements).

Emit a new opcode, "mode", with an argument "html" or "xml", depending
on the value of the xml argument to the constructor.  (This requires a
matching checkin of TALInterpreter.)  (This comment was mistakenly
added to the previous checkin of DummyCompiler.py.)

Don't append to self.program directly; always call self.emit().

on the value of the xml argument to the constructor.  (This requires a
matching checkin of TALInterpreter.)

Move the compile() method into this class.

Make compile() enclose the expression in $dollars$, and add an
uncompile() method to strip them again.  This keeps me honest, making
sure that compile() is always called.

code.

Add the exception raised by Expat to the module, under the name
XMLParseError (which was the name used by the test suite).

started, calls to it would fail with a NameError (map).  The fix is to
using the global _looping to hold the socket map used by the current
loop() call.

Also reformat the code and add some doc strings.

Very preliminary test for the XMLParser class.

Remove an unused import.

Don't use the file's revision number in __version__, although keep it
in a comment.  After a recent exchange on python-dev, it wasn't the
right way to extract the revision number anyway.