Thank you for the updated code.  I had an implmentation question about
one part:

if s and s.isspace() and s[-1] != '\n':

May I suggest three aspects of what I would consider an ideal
compromise:

1) Improve block detection so that blocks could be arbitrarily
indented.  So this is okay:
<?py if ?>
    <?py if ?>
    <?py else ?>
    <?py #end ?>
  <?py if ?>
  <?py else ?>
  <?py #end ?>
<?py end?>

2) Disallow inconsistent indenting of a single block level control
structure.  For instance, this wound NOT be okay:
<?py if ?>
  <?py else ?>
    <?py #end ?>

3) Generate python code with normalized indentation instead of using
"if True:" dummy blocks.

I think these three principals could be accomplished relatively easily
with the following sort of pseudo-code algorithm:

py_parse = re.compile(r'(\s*)<\?py (\s*)(.*?)\?>')
output = []
new_blocks = set(('if', 'while', 'for', 'try', 'with', 'def',
'class'))
cont_blocks = set(('elif', 'else', 'except', 'finally'))
indent_spaces = 4
indent = ' ' * indent_spaces
depth = -1
for line in py_lines:
  pre_space, in_space, statement = py_parse.match(line).groups()
  if statement in new_blocks:
    depth += 1
    output.append(''.join((indent * depth, statement)))
  elif statement in cont_blocks:
    output.append(''.join((indent * depth, statement)))
  elif statement.startswith('#end'):
    depth -= 1
  else:
    output.append(''.join((indent * (depth+1), statement)))
output = '\n'.join(output)

I admit that this only handles single-line <?py> statements with no
trailing text.  But, I think trailing text is easy and mutli-line <?py
\n?> statements could be handled with an additional regex.

I would personally like to see this functionality built-in to tenjin
because I think it makes sense to better be able to combined html
indentation levels with python statement indentation.  However, if
this is not feasible it's not a very big problem for me.  I
implemented a slightly simpler algorithm as a pre-process python
script that strips out almost all leading whitespace.  It uses a
similar technique to preserve <?py > indentation levels but only
outputs a single space per indentation depth.  With this done, it is
not hard to see and hand-edit inconsistent <?py> indentations.

In a previous email I had mentioned that I used an HTML "compressor"
to reduce file size by ~30%.  With the leading-space elimination pre-
process script I managed to reduce the file size by ~25%.  That's
enough that I don't feel bad about missing the last 5%.