Python's documentation includes 2 source codes for alternate xrange implementations, which, at least in my tests, give unexpected results.

# from file:///usr/share/doc/python2.6-doc/html/library/functions.html#xrange
takewhile(lambda x:x<stop, (start+i*step for i in count()))

and:


# from: http://docs.python.org/library/functions.html?highlight=xrange#xrange
islice(count(start, step), (stop-start+step-1)//step)


I'll attach a file with source code showing that, and propose 3 different versions which seems to work fine. (I've prefer the first one, but python lacks of sign() function).

Hi.  Python 2.6 is in security mode, its documentation is not updated nor released anymore.  Is this bug present in the documentation of 2.7 or 3.x versions?

Yes it is. I copied both versions but forgot to specify the second is in 2.7.

This is read in the current (2.7) documentation:
# from: http://docs.python.org/library/functions.html?highlight=xrange#xrange
islice(count(start, step), (stop-start+step-1)//step)

and test_xrange.py shows the output of this code against range(). The code includes 2 samples of parameters which gives the unexpected results.

I verified bug on winxp with 2.7
xrangef [5, 4, 3, 2, 1, 0]
py26 []
py27 [5, 4, 3, 2, 1, 0, -1, -2]
v1 [5, 4, 3, 2, 1, 0]
v2 [5, 4, 3, 2, 1, 0]
v3 [5, 4, 3, 2, 1, 0]
-----------------------
xrangef [5, 3, 1, -1, -3]
py26 []
py27 [5, 3, 1, -1, -3, -5]
v1 [5, 3, 1, -1, -3]
v2 [5, 3, 1, -1, -3]
v3 [5, 3, 1, -1, -3]

The problem is that -1 should instead be +1 for negative steps. The 2.6 version completely failed for negative steps because the comparison needs to be reversed. The doc example could add "for positive steps. For negative steps, use '+1' instead of '-1'". Or change expression to
islice(count(start, step), (stop-start+step-1+2*(step<0))//step)
(the addition is  +2*(step<0)).
Or change -1 to +(-1 if step>0 else 1). I tested both.

I think that if this note should stay in the docs at all, it should be as concise as possible, so I like Terry's -1+2*(step<0) option. I also tested it on a few more inputs and it works fine.

If there are no objections, I can commit it to python 2.7 docs in a couple of days.

The docs should value readability over conciseness IMHO; the examples here with seven operations in a row are a bit scary.

Éric, I'm not sure that the note is necessary at all, but once it's there, it should value *correctness* over conciseness and readability.

Given that the note is already gone* as obsolete in 3.x, I think a minimal maintenance fix for correctness should be fine for 2.7.

* It is replaced, in essence, by "Ranges containing absolute values larger than sys.maxsize are permitted but some features (such as len()) will raise OverflowError."

New changeset 76e5fe8e21fd by Eli Bendersky in branch '2.7':
Issue 12003: fixing error in xrange alternative sample
http://hg.python.org/cpython/rev/76e5fe8e21fd

Agreed. 
Fix committed & issue closed.