|
|
@ -82,6 +82,7 @@ def minimize_source(source):
|
|
|
|
tokens = tokenize.generate_tokens(cStringIO.StringIO(source).readline)
|
|
|
|
tokens = tokenize.generate_tokens(cStringIO.StringIO(source).readline)
|
|
|
|
tokens = strip_comments(tokens)
|
|
|
|
tokens = strip_comments(tokens)
|
|
|
|
tokens = strip_docstrings(tokens)
|
|
|
|
tokens = strip_docstrings(tokens)
|
|
|
|
|
|
|
|
tokens = reindent(tokens)
|
|
|
|
return tokenize.untokenize(tokens)
|
|
|
|
return tokenize.untokenize(tokens)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@ -142,6 +143,25 @@ def strip_docstrings(tokens):
|
|
|
|
yield t
|
|
|
|
yield t
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def reindent(tokens, indent=' '):
|
|
|
|
|
|
|
|
old_levels = []
|
|
|
|
|
|
|
|
old_level = 0
|
|
|
|
|
|
|
|
new_level = 0
|
|
|
|
|
|
|
|
for typ, tok, (start_row, start_col), (end_row, end_col), line in tokens:
|
|
|
|
|
|
|
|
if typ == tokenize.INDENT:
|
|
|
|
|
|
|
|
old_levels.append(old_level)
|
|
|
|
|
|
|
|
old_level = len(tok)
|
|
|
|
|
|
|
|
new_level += 1
|
|
|
|
|
|
|
|
tok = indent * new_level
|
|
|
|
|
|
|
|
elif typ == tokenize.DEDENT:
|
|
|
|
|
|
|
|
old_level = old_levels.pop()
|
|
|
|
|
|
|
|
new_level -= 1
|
|
|
|
|
|
|
|
start_col = max(0, start_col - old_level + new_level)
|
|
|
|
|
|
|
|
if start_row == end_row:
|
|
|
|
|
|
|
|
end_col = start_col + len(tok)
|
|
|
|
|
|
|
|
yield typ, tok, (start_row, start_col), (end_row, end_col), line
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def flags(names):
|
|
|
|
def flags(names):
|
|
|
|
"""Return the result of ORing a set of (space separated) :py:mod:`termios`
|
|
|
|
"""Return the result of ORing a set of (space separated) :py:mod:`termios`
|
|
|
|
module constants together."""
|
|
|
|
module constants together."""
|
|
|
|