Fixing issue #798

[jalberdi004]

Fixing issue #798, thanks to @reaperhulk; removing undocumented '%s' option to convert the date to an int and getting the date in a more robust way.

Closes #798

[alex]

imports should go at teh top of the file

[jalberdi004]

Moved

[alex]

Why was this change made?

[jalberdi004]

Not intended for the PR (I'll remove it as soon as possible).

The error occurs while installing PyOpenSSL in Windows machines (I tried in 3 different machines). The command that fails is:

python setup.py --install

With the output:

Traceback (most recent call last):
File "<string>", line 1, in <module>
File "C:\Users\USER\pyopenssl\setup.py", line 54, in <module>
"\n\n`Full changelog " +
AttributeError: 'NoneType' object has no attribute 'group'

[jalberdi004]

There are python2.7 failures. But those are happening before tests are started, in setup.py
https://travis-ci.org/github/pyca/pyopenssl/jobs/671712647
https://travis-ci.org/github/pyca/pyopenssl/jobs/671712626

ERROR:   flake8: commands failed

The command "if [[ "$(uname -s)" == 'Darwin' ]]; then

  # set our flags to use homebrew openssl

  if [[ "${OPENSSL}"  == "1.1.1" ]]; then

    export LDFLAGS="-L/usr/local/opt/openssl@1.1/lib"

    export CFLAGS="-I/usr/local/opt/openssl@1.1/include"

    export PATH="/usr/local/opt/openssl@1.1/bin:$PATH"

  else

    export LDFLAGS="-L/usr/local/opt/openssl/lib"

    export CFLAGS="-I/usr/local/opt/openssl/include"

    export PATH="/usr/local/opt/openssl/bin:$PATH"

  fi

[alex]

the flake8 issue is real, the failing test is related to cryptography dropping openssl 1.0.1. we'll have to solve that outside of this PR.

[jalberdi004]

Good, thank you for the info

[reaperhulk]

Since this was originally done the CI has been changed significantly and we black'd the codebase. If you could rebase this we can get it merged though!

[jalberdi004]

It fails with the following error (https://travis-ci.com/github/pyca/pyopenssl/jobs/401351492):

flake8 run-test: commands[0] | black --check .

[5934] /home/travis/build/pyca/pyopenssl$ /home/travis/build/pyca/pyopenssl/.tox/flake8/bin/black --check .

would reformat /home/travis/build/pyca/pyopenssl/src/OpenSSL/crypto.py

Oh no! 💥 💔 💥

1 file would be reformatted, 23 files would be left unchanged.

ERROR: InvocationError for command /home/travis/build/pyca/pyopenssl/.tox/flake8/bin/black --check . (exited with code 1)

___________________________________ summary ____________________________________

ERROR:   flake8: commands failed

The command "~/.venv/bin/tox -v" exited with 1.

I don't understand what does it mean and how to fix it.

[alex]

All the python files int his repo are formatted with black. That error is saying that one of these files is not correctly formatted.

[jalberdi004]

Now is solved the problem with formatting

[orosam]

Unfortunately, timegm(timetuple()) also disregards time zones, just the other way around.
Let me demonstrate with a small script (best results with python3, plus python-dateutil is also needed):

from __future__ import print_function

from calendar import timegm
from datetime import datetime
from os import putenv

from dateutil import tz

TZ = "CET"  # it's UTC + 2 at the tested time below
putenv("TZ", TZ)

tzs = [
    ("UTC", tz.gettz("UTC")),
    ("Naive (local zone is {})".format(TZ), None),
    ("EEST", tz.gettz("EET")),  # It will be UTC + 3 hours at the tested time 
]

dates = [(name, datetime(2020, 8, 1, 11, 0, 0, tzinfo=tzinfo)) for name, tzinfo in tzs]

for name, date in dates:
    print("{}: {!r}".format(name, date))

    try:
        print("  python3 timestamp()        : {}".format(int(date.timestamp())))
    except AttributeError:
        pass

    print("  timestamp with timetuple   : {} ({!r})".format(timegm(date.timetuple()), date.timetuple()))
    print("  timestamp with utctimetuple: {} ({!r})".format(timegm(date.utctimetuple()), date.utctimetuple()))
    print("  timestamp with old strftime: {}".format(date.strftime("%s")))
    print()

Result:

UTC: datetime.datetime(2020, 8, 1, 11, 0, tzinfo=tzfile('/usr/share/zoneinfo/UTC'))
  python3 timestamp()        : 1596279600
  timestamp with timetuple   : 1596279600 (time.struct_time(tm_year=2020, tm_mon=8, tm_mday=1, tm_hour=11, tm_min=0, tm_sec=0, tm_wday=5, tm_yday=214, tm_isdst=0))
  timestamp with utctimetuple: 1596279600 (time.struct_time(tm_year=2020, tm_mon=8, tm_mday=1, tm_hour=11, tm_min=0, tm_sec=0, tm_wday=5, tm_yday=214, tm_isdst=0))
  timestamp with old strftime: 1596276000

Naive (local zone is CET): datetime.datetime(2020, 8, 1, 11, 0)
  python3 timestamp()        : 1596272400
  timestamp with timetuple   : 1596279600 (time.struct_time(tm_year=2020, tm_mon=8, tm_mday=1, tm_hour=11, tm_min=0, tm_sec=0, tm_wday=5, tm_yday=214, tm_isdst=-1))
  timestamp with utctimetuple: 1596279600 (time.struct_time(tm_year=2020, tm_mon=8, tm_mday=1, tm_hour=11, tm_min=0, tm_sec=0, tm_wday=5, tm_yday=214, tm_isdst=0))
  timestamp with old strftime: 1596272400

EEST: datetime.datetime(2020, 8, 1, 11, 0, tzinfo=tzfile('/usr/share/zoneinfo/EET'))
  python3 timestamp()        : 1596268800
  timestamp with timetuple   : 1596279600 (time.struct_time(tm_year=2020, tm_mon=8, tm_mday=1, tm_hour=11, tm_min=0, tm_sec=0, tm_wday=5, tm_yday=214, tm_isdst=1))
  timestamp with utctimetuple: 1596268800 (time.struct_time(tm_year=2020, tm_mon=8, tm_mday=1, tm_hour=8, tm_min=0, tm_sec=0, tm_wday=5, tm_yday=214, tm_isdst=0))
  timestamp with old strftime: 1596272400

So while the old strftime(%s) always treated vfy_time as a datetime in local time, the current implementation always treats it as a datetime in UTC. Both disregards time zones.

Using utctimetuple() could be better, although it would also treat timezone-naive datetime objects as being in UTC, which might be surprising. But that could be documented.