itertools.count() falls back to fast (integer) mode when step rounds to 1

[StyXman]

BPO	28019
Nosy	@rhettinger, @facundobatista, @serhiy-storchaka
Files	fix_28019.diff fix_28019.diff itertools_count_truncated_step.patch

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/serhiy-storchaka'
closed_at = <Date 2016-09-10.07:11:34.570>
created_at = <Date 2016-09-08.12:48:32.348>
labels = ['type-bug', 'library']
title = 'itertools.count() falls back to fast (integer) mode when step rounds to 1'
updated_at = <Date 2016-09-10.07:11:34.569>
user = 'https://bugs.python.org/StyXman'

bugs.python.org fields:

activity = <Date 2016-09-10.07:11:34.569>
actor = 'serhiy.storchaka'
assignee = 'serhiy.storchaka'
closed = True
closed_date = <Date 2016-09-10.07:11:34.570>
closer = 'serhiy.storchaka'
components = ['Library (Lib)']
creation = <Date 2016-09-08.12:48:32.348>
creator = 'StyXman'
dependencies = []
files = ['44466', '44467', '44471']
hgrepos = []
issue_num = 28019
keywords = ['patch']
message_count = 6.0
messages = ['275007', '275016', '275025', '275038', '275082', '275583']
nosy_count = 5.0
nosy_names = ['rhettinger', 'facundobatista', 'StyXman', 'python-dev', 'serhiy.storchaka']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue28019'
versions = ['Python 3.5', 'Python 3.6']

[StyXman]

If the step parameter for itertools.count() rounds to 1 (f.i., 1.1, 1.437643, 1.99999), then it fallsback to fast (integer) mode and increases the counter by 1. Here's an example:

Python 3.6.0a4+ (default:ddc95a9bc2e0+, Sep  8 2016, 14:46:19)
>>> import itertools
>>> for i in itertools.count(1, step=1.5):
...     print(i)
...     if i > 10:
...         break
1
2
3
4
5
6
7
8
9
10
11

[StyXman]

Here's a first approach on fixing this bug. I'm not sure how to handle the case where step=1.0.

[facundobatista]

I think the fix nails it; all the problem was that the "fast" mode was wrongly detected, and all the problems (counting badly, or a bad repr, etc) is a problem after setting cnt into PY_SSIZE_T_MAX.

IIUC there is nothing special to do when step=1.0, as later on the original objects are used (they don't suffer a wrong cast to integer anymore).

[StyXman]

New patch following comments from SilentGhost.

[serhiy-storchaka]

PyLong_AsLong() not just converts Python integer to C long, but it also converts Python float to Python integer (with the lost of the fractional part). We need to check that the value is Python integer.

Here is a patch that fixes this issue and few other bugs.

[python-dev]

New changeset b23f963a7b45 by Serhiy Storchaka in branch '3.5':
Issue bpo-28019: itertools.count() no longer rounds non-integer step in range
https://hg.python.org/cpython/rev/b23f963a7b45

New changeset 74667320778e by Serhiy Storchaka in branch '2.7':
Issue bpo-28019: Backported additional tests for itertools.count().
https://hg.python.org/cpython/rev/74667320778e

New changeset 51dfab4f28e7 by Serhiy Storchaka in branch 'default':
Issue bpo-28019: itertools.count() no longer rounds non-integer step in range
https://hg.python.org/cpython/rev/51dfab4f28e7