Add Windows CI

[nicoddemus]

Fix #59

Also:

• Renamed README to rst so GH renders it properly;
• Updated classifiers;
• Added badges.

[nicoddemus]

Some tests are failing because I'm guessing they were not being executed on Windows yet.

I will take a look later to make them pass or skip them if not possible.

[nicoddemus]

It seems this was a bug on Windows:

>>> import shlex
>>> shlex.split(r'X:\execnet\.env36\Scripts\python.exe')
['X:execnet.env36Scriptspython.exe']

Which of course is not a valid Python file. My quick and dirty solution at least gives a valid file:

import shlex
>>> shlex.split(r'X:\execnet\.env36\Scripts\python.exe'.replace('\\', '/'))
['X:/execnet/.env36/Scripts/python.exe']

[nicoddemus]

flakes is not passing; given the errors and the fact that Travis does not run them as well I think it is not passing on Linux as well;

pypy is a little more tricky, I tried to debug this a little but no luck.

I can reproduce the failure with this simple script:

import execnet
from execnet.gateway import rinfo_source

def rinfo_source(channel):
    import sys
    import os
    d = dict(
        executable=sys.executable,
        version_info=sys.version_info[:5],
        platform=sys.platform,
        cwd=os.getcwd(),
        pid=os.getpid(),
    )
    channel.send(d)

gw = execnet.makegateway()
channel = gw.remote_exec(rinfo_source)
print(channel.receive())

This fails with:

Traceback (most recent call last):
  File "X:/execnet/.tmp/foo.py", line 31, in <module>
    print(channel.receive())
  File "X:\execnet\execnet\gateway_base.py", line 737, in receive
    raise self._getremoteerror() or EOFError()
EOFError

If I send a simple string (channel.send("hello")) it works, but the dictionary fails for some reason.

I used PyCharm's remote debugging and managed to get to this line on the remote:

# gateway_base.py
    def _send(self, msgcode, channelid=0, data=bytes()):
        message = Message(msgcode, channelid, data)
        try:
            message.to_io(self._io)   # << terminates here
            self._trace('sent', message)  
        except (IOError, ValueError):
            e = sys.exc_info()[1]
            self._trace('failed to send', message, e)
            # ValueError might be because the IO is already closed
            raise IOError("cannot send (already closed?)")

In the line highlighted above, the process simply terminates without any discernible message. Any pointers?

[nicoddemus]

Played around a little bit more:

---

import sys
import execnet

def rinfo_source(channel):
    import sys
    import os
    d = dict(
        executable=sys.executable,
        version_info=sys.version_info[:5],
        platform=sys.platform,
        cwd=os.getcwd(),
        pid=os.getpid(),
    )
    channel.send(d)

gw = execnet.makegateway()
channel = gw.remote_exec(rinfo_source)
print 'local executable:', repr(sys.executable)
print 'local version_info:', repr(sys.version_info)
print 'remote receive:', channel.receive()

local executable: 'X:\\execnet\\.envpypy\\bin\\pypy.exe'
local version_info: (major=2, minor=7, micro=10, releaselevel='final', serial=42)
remote receive:Traceback (most recent call last):
  File "X:/execnet/.tmp/foo.py", line 20, in <module>
    print 'remote receive:', channel.receive()
  File "X:\execnet\execnet\gateway_base.py", line 737, in receive
    raise self._getremoteerror() or EOFError()
EOFError

---

Commenting out both executable and version_info from the remote dict makes the example work:

import sys
import execnet

def rinfo_source(channel):
    import sys
    import os
    d = dict(
        #executable=sys.executable,
        #version_info=sys.version_info[:5],
        platform=sys.platform,
        cwd=os.getcwd(),
        pid=os.getpid(),
    )
    channel.send(d)

gw = execnet.makegateway()
channel = gw.remote_exec(rinfo_source)
print 'local executable:', repr(sys.executable)
print 'local version_info:', repr(sys.version_info)
print 'remote receive:', channel.receive()

local executable: 'X:\\execnet\\.envpypy\\bin\\pypy.exe'
local version_info: (major=2, minor=7, micro=10, releaselevel='final', serial=42)
remote receive: {'platform': 'win32', 'cwd': 'X:\\execnet\\.tmp', 'pid': 16360}

Leaving either of them on the dict gives the same error.

---

Strangely, renaming the executable entry to something else:

import sys
import execnet

def rinfo_source(channel):
    import sys
    import os
    d = dict(
        _executable=sys.executable,
        #version_info=sys.version_info[:5],
        platform=sys.platform,
        cwd=os.getcwd(),
        pid=os.getpid(),
    )
    channel.send(d)

gw = execnet.makegateway()
channel = gw.remote_exec(rinfo_source)
print 'local executable:', repr(sys.executable)
print 'local version_info:', repr(sys.version_info)
print 'remote receive:', channel.receive()

Works:

remote receive: {'_executable': 'X:\\execnet\\.envpypy\\bin\\pypy.exe', 'platform': 'win32', 'cwd': 'X:\\execnet\\.tmp', 'pid': 16076}

Perhaps there's some bug on the marshaler which fails on PyPy with the name executable?

---

I tried the same trick of changing the name for version_info (and even making it a plain tuple for good measure):

import sys
import execnet

def rinfo_source(channel):
    import sys
    import os
    d = dict(
        #executable=sys.executable,
        _version_info=tuple(sys.version_info[:5]),
        platform=sys.platform,
        cwd=os.getcwd(),
        pid=os.getpid(),
    )
    channel.send(d)

gw = execnet.makegateway()
channel = gw.remote_exec(rinfo_source)
print 'local executable:', repr(sys.executable)
print 'local version_info:', repr(sys.version_info)
print 'remote receive:', channel.receive()

But in this case I still get the same error.

[RonnyPfannschmidt]

for linting i wan to use a different service, imho its just bullshit to make it part of testing and run it on all platforms

[RonnyPfannschmidt]

also execnet is in a shit place wrt linting, dont even try making it a ci req right now is just wasting time

[nicoddemus]

also execnet is in a shit place wrt linting, dont even try making it a ci req right now is just wasting time

No problem, I tried to reproduce the environments I saw defined in tox.ini.

btw, Travis and tox are inconsistent to each other. Should we getting them in sync?

I would like to support the same versions we guarantee pytest-xdist to run, what do you think?

[nicoddemus]

What about pypy? I won't have too much time to look deep into it, so I suppose we can create an issue for it and mark the affected tests as xfail (it is a couple of tests IIRC)

[RonnyPfannschmidt]

@nicoddemus yes, we can do that bit and add pypy

[nicoddemus]

Yay finally 😅

[RonnyPfannschmidt]

Lol nice one art complex