time_hash() reads the wrong bytes to get microseconds

[vstinner]

BPO	29140
Nosy	@abalkin, @vstinner
Files	time_hash.patch check.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 = None
closed_at = <Date 2017-02-01.17:15:10.446>
created_at = <Date 2017-01-03.13:57:09.547>
labels = ['3.7', 'library']
title = 'time_hash() reads the wrong bytes to get microseconds'
updated_at = <Date 2017-02-01.17:15:10.445>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2017-02-01.17:15:10.445>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2017-02-01.17:15:10.446>
closer = 'vstinner'
components = ['Library (Lib)']
creation = <Date 2017-01-03.13:57:09.547>
creator = 'vstinner'
dependencies = []
files = ['46123', '46130']
hgrepos = []
issue_num = 29140
keywords = ['patch', '3.6regression']
message_count = 4.0
messages = ['284561', '284596', '284597', '284598']
nosy_count = 3.0
nosy_names = ['belopolsky', 'vstinner', 'python-dev']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = None
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue29140'
versions = ['Python 3.6', 'Python 3.7']

[vstinner]

When the time is folded, time_hash() uses DATE_xxx() macros, instead of TIME_xxx() macros, and so reads microseconds from the wrong bytes.

Bug introduced by the implementation of the PEP-495 (Local Time Disambiguation).

#define PyDateTime_DATE_GET_MICROSECOND(o)              \
    ((((PyDateTime_DateTime*)o)->data[7] << 16) |       \
     (((PyDateTime_DateTime*)o)->data[8] << 8)  |       \
      ((PyDateTime_DateTime*)o)->data[9])

#define PyDateTime_TIME_GET_MICROSECOND(o)              \
    ((((PyDateTime_Time*)o)->data[3] << 16) |           \
     (((PyDateTime_Time*)o)->data[4] << 8)  |           \
      ((PyDateTime_Time*)o)->data[5])

Attached patch fixes time_hash().

I guess that it's a dummy copy-paste issue.

[vstinner]

I used attached check.patch to check if macros are misused in other functions. Hopefully, only time_hash() has bugs.

I don't think that it's worth it to apply check.patch, I dislike such complex macro.

[python-dev]

New changeset 72e48ff7b169 by Victor Stinner in branch '3.6':
Issue bpo-29140: Fix hash(datetime.time)
https://hg.python.org/cpython/rev/72e48ff7b169

[vstinner]

Note: I found this bug while working on the issue bpo-29100.