In section 17.5 of the Python 2.0 Library Reference, 
the constant "NL" is not documented.

Logged In: YES 
user_id=3066

Ping, should NL be documented?

Logged In: YES 
user_id=31435

Yes, it should be, but Ping doesn't know about NL:  Guido 
added that one.  Python's real tokenizer has no such token 
type.  It was added because-- unlike Python's real 
tokenizer --tokenizer.py passes *everything* on to the 
client, including blank lines and comments, so passes on 
newlines that aren't NEWLINEs too.  Reassigned to Guido in 
case he remembers more about the details ...

Logged In: YES 
user_id=39128

tokenize.NL seems to be an empty line, as opposed to 
token.NEWLINE which is a logical line end. Insofar NL is 
comparable to COMMENT, since it has no semantic content and 
is ignorable by a parser.

1 a = \
2 2
3
4 b=4

emits these tokens:

1,0-1,1:        NAME    'a'
1,2-1,3:        OP      '='
2,0-2,1:        NUMBER  '2'
2,1-2,2:        NEWLINE '\012'
3,0-3,1:        NL      '\012'
4,0-4,1:        NAME    'b'
4,1-4,2:        OP      '='
4,2-4,3:        NUMBER  '4'
4,3-4,4:        NEWLINE '\012'
5,0-5,0:        ENDMARKER       ''

Logged In: YES 
user_id=6380

The time machne strikes again.  Skip Montanaro added docs
for NL in February.  For proof, see:

http://python.sourceforge.net/devel-docs/lib/module-tokenize.html