Tokenizing `' 1 2 3` versus `''' 1 2 3` yields different results.

Tokenizing `' 1 2 3` gives:

1,0-1,1:	ERRORTOKEN	"'"
1,2-1,3:	NUMBER	'1'
1,4-1,5:	NUMBER	'2'
1,6-1,7:	NUMBER	'3'
2,0-2,0:	ENDMARKER	''

while tokenizing `''' 1 2 3` yields:

Traceback (most recent call last):
  File "prog.py", line 4, in <module>
    tokenize.tokenize(iter(["''' 1 2 3"]).next)
  File "/usr/lib/python2.6/tokenize.py", line 169, in tokenize
    tokenize_loop(readline, tokeneater)
  File "/usr/lib/python2.6/tokenize.py", line 175, in tokenize_loop
    for token_info in generate_tokens(readline):
  File "/usr/lib/python2.6/tokenize.py", line 296, in generate_tokens
    raise TokenError, ("EOF in multi-line string", strstart)
tokenize.TokenError: ('EOF in multi-line string', (1, 0))


Apparently tokenize decides to re-tokenize after the erroneous quote in the case of a single-quote, but not a triple-quote. I guess that this is because retokenizing the rest of the file after an unclosed triple-quote would be expensive; however, I've also been told it's very strange and possibly wrong for tokenize to be inconsistent this way.

If this is the right behavior, I guess I'd like it if it were documented. This sort of thing is confusing / potentially misleading for users of the tokenize module. Or at least, when I saw how single quotes were handled, I assumed incorrectly that all quotes were handled that way.