time.localtime(float("NaN")) does not raise a ValueError on all platforms

[gpshead]

BPO	26669
Nosy	@gpshead, @mdickinson, @abalkin, @vstinner, @Mariatta, @isidentical
PRs	bpo-26669: Fix nan arg value error in pytime.c #3085 [3.6] bpo-26669: Fix nan arg value error in pytime.c (GH-3085) #3467 bpo-26669: Reject float infinity in time functions #11507 bpo-26669: Reject float infinity in time functions #11507 bpo-26669: Reject float infinity in time functions #11507 bpo-26669: Reject float infinity in time functions #11507

^{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/abalkin'
closed_at = <Date 2019-10-29.11:12:36.394>
created_at = <Date 2016-03-30.07:17:23.433>
labels = ['extension-modules', 'interpreter-core', 'type-bug']
title = 'time.localtime(float("NaN")) does not raise a ValueError on all platforms'
updated_at = <Date 2019-10-29.11:12:36.392>
user = 'https://github.com/gpshead'

bugs.python.org fields:

activity = <Date 2019-10-29.11:12:36.392>
actor = 'vstinner'
assignee = 'belopolsky'
closed = True
closed_date = <Date 2019-10-29.11:12:36.394>
closer = 'vstinner'
components = ['Extension Modules', 'Interpreter Core']
creation = <Date 2016-03-30.07:17:23.433>
creator = 'gregory.p.smith'
dependencies = []
files = []
hgrepos = []
issue_num = 26669
keywords = ['patch']
message_count = 10.0
messages = ['262653', '300237', '301041', '301754', '301755', '301760', '301761', '333403', '355640', '355644']
nosy_count = 6.0
nosy_names = ['gregory.p.smith', 'mark.dickinson', 'belopolsky', 'vstinner', 'Mariatta', 'BTaskaya']
pr_nums = ['3085', '3467', '11507', '11507', '11507', '11507']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue26669'
versions = ['Python 2.7', 'Python 3.4', 'Python 3.5', 'Python 3.6']

[gpshead]

time.localtime(float("NaN")) raises a ValueError on x86_64 using the few compilers I have tested it with. (this makes sense)

>>> time.localtime(float("NaN"))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: (75, 'Value too large for defined data type')

On an arm and arm64 system, it does not and treats NaN as 0. (nonsense!)

>>> time.localtime(float("NaN"))
time.struct_time(tm_year=1970, tm_mon=1, tm_mday=1, tm_hour=0, tm_min=0, tm_sec=0, tm_wday=3, tm_yday=1, tm_isdst=0)

The root of this problem appears to be the (potentially questionable? I'll ask a C compiler person...) code in Python/pytime.c's (3.x) Modules/timemodule.c's (2.7) double to time_t conversion function.

I'm not sure what it does is supposed to be well defined behavior with NaN...

The easy fix is to add:

#include <math.h>

and

add || isnan(x) || isinf(x) to the check that raises a ValueError in

https://hg.python.org/cpython/file/4c903ceeb4d1/Python/pytime.c#l149 (3.x)
https://hg.python.org/cpython/file/2.7/Modules/timemodule.c#l102 (2.7)

Along with a relevant assertRaises(ValueError) unittest for NaN, inf and -inf in test_time.py.

[mdickinson]

potentially questionable? I'll ask a C compiler person...

Questionable indeed. Attempting to cast a NaN to an integer type results in undefined behaviour. Unfortunately, so does attempting to cast any double value that's outside the range represented by the time_t type, so the questionability extends beyond just the NaN handling.

[vstinner]

I like PR 3085 to raise explicitly a ValueError with an helpful error message.

[vstinner]

New changeset 829dacc by Victor Stinner (Han Lee) in branch 'master':
bpo-26669: Fix nan arg value error in pytime.c (bpo-3085)
829dacc

[vstinner]

For Python 2.7, you have at least to fix these 2 functions: parse_time_double_args(), _PyTime_DoubleToTimet().

[Mariatta]

New changeset a4baf1c by Mariatta (Miss Islington (bot)) in branch '3.6':
[3.6] bpo-26669: Fix nan arg value error in pytime.c (GH-3085) (GH-3467)
a4baf1c

[Mariatta]

This has been backported to 3.6.
Is backport to 2.7 needed? If not we'll close the issue.

[vstinner]

I wrote PR 11507 to raise ValueError rather than OverflowError for -inf and +inf.

[isidentical]

Victor's PR 11507 is closed, what actions are going to be taken next? Close the issue as Mariatta said?

[vstinner]

Python 2 users have this bug since 2010, I think that it's fine to not fix it. Python 2 support ends at the end of the year. I close the issue.

The issue has been fixed in Python 3.6 and newer.