The tokenize module's tokenizer functions output incorrect (or at least misleading) information when the content being tokenized does not end in a line ending character.  This is related to the fix for issue<33899>  which added the NEWLINE tokens for this case but did not fill out the whole token tuple correctly.

The bug can be seen by running a version of the test in Lib/test/test_tokenize.py:

import io, tokenize

newline_token_1 = list(tokenize.tokenize(io.BytesIO("x\n".encode('utf-8')).readline))[-2]
newline_token_2 = list(tokenize.tokenize(io.BytesIO("x".encode('utf-8')).readline))[-2]

print(newline_token_1)
print(newline_token_2)

# Prints:
# TokenInfo(type=4 (NEWLINE), string='\n', start=(1, 1), end=(1, 2), line='x\n')
# TokenInfo(type=4 (NEWLINE), string='', start=(1, 1), end=(1, 2), line='')  # bad "end" and "line"!

Notice that "len(newline_token_2.string) == 0" but "newline_token_2.end[1] - newline_token_2.start[1] == 1".  Seems more consistent if the newline_token_2.end == (1, 1).

Also, newline_token_2.line should hold the physical line rather than the empty string.  This would make it consistent with newline_token_1.line.

I'll add a PR shortly with a change so the output from the two cases is:

TokenInfo(type=4 (NEWLINE), string='\n', start=(1, 1), end=(1, 2), line='x\n')
TokenInfo(type=4 (NEWLINE), string='', start=(1, 1), end=(1, 1), line='x')

If this looks reasonable, I can backport it for the other branches.  Thanks!