the tracker.

mode opcodes.

Fix check_code_short_endtags(), which should have used <img /> since
the previous batch of changes.

- that the TALcode version of the macro is the same as that of the
  main program;

- that the mode (html or xml) of the macro is the same as that of the
  main program.

If either of these tests fails, raise METALError().

Added three new inquiries for TALcode programs: isCurrentVersion(),
getProgramVersion(), and getProgramMode().

generating HTML, and "/>" for XML, instead of always using "/>".

This is necessary for proper XHTML generation.

self.endsep to the proper value there.

this fixes the Windows crashes?

checkSimpleTransactionalUndo(): Modify tests to check zombification
and subsequent re-birth of objects via transactionalUndo().

works by writing a metadata record which contains a special lrevid
(pickle pointer).  Upon loading a zombified object, naive applications
will get a KeyError as would be expected for loading a non-existant
object.

The actual exception though is ObjectDoesNotExist -- derived from
KeyError -- which contains a `revid' attribute pointing to the
revision of the object which zombified it.  Thus, by undoing this
revid, the object can be reborn to its initial state.

Specifically,

ObjectDoesNotExist: New exception, derived from KeyError.

load(): If an object's metadata has an lrevid with the special `dne'
value (64-bits full of 1's), the object is a zombie and
ObjectDoesNotExist is raised.

transactionalUndo(): If prevrevid for the revision we're undoing is
zero, it means that we're undoing the object's creation.  Instead of
raising an UndoError, write a new metadata record containing the
original record, but with lrevid == dne, and prevrevid == tid.

This change also contains initial (untested) support for pack(),
specifically,

_zaprevision(): pack doesn't happen in a transaction, so remove the
txn argument.  Add the referencesf argument, passed from the pack()
call.

pack(): Initial implementation.  Seems fairly simple really (too
simple? ;)

XMLParserTestCase.check_bad_nesting():
    Use self.fail() instead of naming AssertionError directly; how PyUnit
    deals with the error is it's problem, not ours.  ;-)

Fix two buffer boundary issues; this restores this to passing the test
suite without any large restructuring.

Guido & I will be looking at how this is structured later; buffer
boundary checks will make this nearly unmaintainable if we can't
bring about a better structure to the code.  (Better tests would also
be nice!)

character at a time.  This uncovers two new bugs in HTMLParser that
I'm going to attack next.

default true.  When this is false, structure insertion without
attribute replacement does not parse the inserted text.  This is
(expected to be) much faster, but allows invalid HTML/XML to be
inserted, so it is off by default.

Evan can choose to turn this on in ZPT if he's more concerned about
efficiency than about correctness. :-)

<script> and <style> elements aren't mangled.

By nature of the parser, bare < and & cannot occur in the input data
(the parser raises an exception when it encounters these); bare > is
harmless.

buffer boundaries in the middle of the cdata stretch.

Fixed this, and added a clear_cdata_mode() callback after a successful
</endtag>.  (I hope this doesn't break other things.  This parser is
getting horribly ad-hoc. :-( )

code.