When tokenizing with tokenize.generate_tokens, if the code ends with whitespace (no newline), the tokenizer produces an ERRORTOKEN for each space.  Additionally, the regex that fails to find tokens in those spaces is linear in the number of spaces, so the overall performance is O(n**2).

I found this while tokenizing code samples uploaded to a public web site.  One sample for some reason ended with 40,000 spaces, which was taking two hours to tokenize.

Demonstration:

{{{
import token
import tokenize

try:
    from cStringIO import StringIO
except:
    from io import StringIO

code = "@"+" "*10000
code_reader = StringIO(code).readline

for num, (ttyp, ttok, _, _, _) in enumerate(tokenize.generate_tokens(code_reader)):
    print("%5d %15s %r" % (num, token.tok_name[ttyp], ttok))
}}}

Here's a patch for 3.3.

I would like to also fix 2.7...

Updated with new (better) patch, for v2.7 and v3.3.  They are the same except for the test.

Ned, could you possibly send a Contributor Form Agreement? http://www.python.org/psf/contrib/

Jesús, done!

New changeset eb7ea51e658e by Ezio Melotti in branch '2.7':
#16152: fix tokenize to ignore whitespace at the end of the code when no newline is found.  Patch by Ned Batchelder.
http://hg.python.org/cpython/rev/eb7ea51e658e

New changeset 3ffff1798ed5 by Ezio Melotti in branch '3.2':
#16152: fix tokenize to ignore whitespace at the end of the code when no newline is found.  Patch by Ned Batchelder.
http://hg.python.org/cpython/rev/3ffff1798ed5

New changeset 1fdeddabddda by Ezio Melotti in branch '3.3':
#16152: merge with 3.2.
http://hg.python.org/cpython/rev/1fdeddabddda

New changeset ed091424f230 by Ezio Melotti in branch 'default':
#16152: merge with 3.3.
http://hg.python.org/cpython/rev/ed091424f230

Fixed, thanks for the report and the patch!

PR 6273 is mentioned, but I think 6573 is the correct number.