cStringIO.

items and start and end tags without attributes into a single rawtext
item.

We could do more, by also changing start tags with only plain
(non-substituted) attributes into rawtext and collapsing.  After
breakfast.

- add getCode() interface to TALGenerator

- remove premature optimization from emitRawText()

- rework unEmitNewlineWhitespace() so it works with unoptimized code

- use getCode() in HTMLTALParser and TALParser

Move quote() function here from TALInterpreter.

references, introduced by the line number / offset code.

Changed the interface for the NestingError exception; it now has tag,
lineno, and offset attributes, and reports these in its __str__; the
tagstack is no longer needed.

Moved all the "internal thingies" to the end; renamed some for more
consistent naming.

retrieve.

Added a loop over all filename arguments.  It now prints the filename,
the compilation time, and the interpretation time for each argument.

Add new test files for HTML, plus their output (not yet hand-checked).

Fix the input *and* output for test9 -- it was referencing an
undefined macro before.

When checking for auto-closure of list- and table-related tags, be sensitive
to the specific tag in hand so that we do not close too many tags.

Allow colons and underscores in element type names.

Make sure we always flush stdout before switch to stderr.

Do not start the traceback on the same line as the name of the file we are
testing.

Define a NestingError exception that gives a reasonable error message when
the open element stack does not contain the tag passed to finish_endtag().
The new message gives us the tag that was seen and the contents of the
tag stack at the time.

stopping on the first exception.

Add -q option that supprresses most output (diffs and tracebacks).

Remove code to support the null-end-tag support (<span/.../); this is
*never* used with HTML.  Removing this simply reduces the number of
special cases that clutter the start-tag parsing.

Add support for XHTML-style end-tags (<img src="foo.gif" />).

to TALGenerator, so it can be reused by HTMLTALParser.

Also got rid of the CharacterData and Comment handlers in TALParser
-- the Default handler does the right thing already.

HTML TAL parser.  This moves the "fixing" of the attributes (really
changing them from "<namespace> name" to "<prefix>:name") to the same
loop over the attributes where the presence of unrecognized attributes
is reported, simplifying life later.

Update the HTML test to ensure that automagical placement of end tags
only happens for implied end tags and not all end tags in the affected
group.

This also corrects for the bug where paragraph-level tags would close
themselves (before they opened!) if they implicitly closed anything at
all.

Fix a bug that caused paragraph-level markup to close itself when implicitly
closing anything.

For some elements, when the end tag is implied, insert the tag before any
whitespace at the end of the text being enclosed.  This makes the generated
end tags "feel" more like human-generated markup.

If there are errors, generate a non-zero return code.