Cannot use threading with Mac OSX Backend

The following does not work on Mac OSX:

from __future__ import division

import threading
import time
import sys
import numpy as np
import matplotlib
matplotlib.use('MacOSX')

import matplotlib.pyplot as plt


class async_worker(threading.Thread):
    def __init__(self, fig, axis):
        threading.Thread.__init__(self)
        self.fig = fig
        self.axis = axis

    def run(self):
        i = 0
        while True:
            print "Run thread %d" % i
            sys.stdout.flush()
            i += 1
            time.sleep(0.1)


def main():
    fig = plt.figure()
    ax = fig.add_subplot(1, 1, 1)
    ax.plot(np.random.rand(251, 1))

    async_worker(fig, ax).start()
    plt.show()


if __name__ == "__main__":
    main()

I expect to see "Run thread" repeated every 0.1 seconds, no matter if I interact with the plot or not. Instead I only see it repeat when I click and drag around in the MPL figure.

When switching to matplotlib.use('TkAgg'), the second thread runs as expected.

@mdehoon

Cc: @mdehoon

I cannot replicate this with the current matplotlib in master; I get the same behavior with the MacOSX backend and the tkagg backend.
Which version of matplotlib are you using?
Can you try with the matplotlib release candidate:
https://github.com/matplotlib/matplotlib/releases/tag/v1.5.0rc2

I have the same problem with 1.4.3 and after cloning and pip install -eing e7e8bcb: The second thread only runs when I am moving my mouse.

@nils-werner

I can confirm the bug if I have interactive : False in matplotlibrc ;
the bug does not appear if I set interactive : True.
@nils-werner Can you confirm this?

When I add plt.interactive(True) to the top of the script

The second thread runs
Matplotlib does not plot anything, the window seems to freeze (spinning ball of death)

@nils-werner

@nils-werner Can you try with the latest master (or release candidate 1.5rc2)?
If I add plt.interactive(True) to the top of the script, the second thread runs and the window behaves normally.

I tried it with both stable and master, the behaviour is always the same: The figure window stays blank.

how was python installed/built for you? Maybe it is a subtle difference
like the whole frameworks/nonframeworks issue?

On Fri, Oct 16, 2015 at 11:29 AM, Nils Werner [email protected]
wrote:

I tried it with both stable and master, the behaviour is always the same:
The figure window stays blank.

—
Reply to this email directly or view it on GitHub
#5251 (comment)
.

brew install python --with-brewed-openssl --framework

and fixed the virtualenv framework issue using https://github.com/gldnspud/virtualenv-pythonw-osx

@nils-werner

@nils-werner
Can you try the following:
In the function static PyObject* show(PyObject* self) in src/_macosx.m, replace

    [NSApp run];

by

    Py_BEGIN_ALLOW_THREADS
    [NSApp run];
    Py_END_ALLOW_THREADS

That should get you the correct behavior in the non-interactive case.

Yes, that indeed fixes the issue! Shall I prepare a PR with that fix?

@nils-werner

@nils-werner We're not quite there yet; I have not yet been able to reproduce the spinning ball of death in the interactive case. Also there may be a more appropriate place to put the Py_BEGIN_ALLOW_THREADS / Py_END_ALLOW_THREADS block.

@nils-werner

@nils-werner On second thought, since we are in the process of release candidates for the 1.5 release, perhaps it's better to wait with more extensive changes to _macosx.m. So if you could prepare a PR with the simply fix of putting Py_BEGIN_ALLOW_THREADS / Py_END_ALLOW_THREADS around [NSApp run], then I can accept it.

Fix Mac OSX backend threading issues, issue matplotlib#5251

@mdehoon

@mdehoon Should this be closed or left open for a more through fix?

@tacaswell

@tacaswell Let's leave it open. I will have another look after 1.5 final is out.

I have discovered another issue while trying to draw() from another thread:

from __future__ import division

import threading
import time
import sys
import numpy as np
import matplotlib
matplotlib.use('MacOSX')

import matplotlib.pyplot as plt


class async_worker(threading.Thread):
    def __init__(self, fig, axis):
        threading.Thread.__init__(self)
        self.fig = fig
        self.axis = axis

    def run(self):
        i = 0
        while True:
            self.fig.canvas.draw()
            print "Run thread %d" % i
            sys.stdout.flush()
            i += 1
            time.sleep(0.1)


def main():
    fig = plt.figure()
    ax = fig.add_subplot(1, 1, 1)
    ax.plot(np.random.rand(251, 1))

    async_worker(fig, ax).start()
    plt.show()


if __name__ == "__main__":
    main()

Whenever I click any of the navigation bar buttons, the entire process deadlocks. Again, when using TkAgg, this works just fine.

You are trying to draw from the secondary thread, while typically you should draw from the main thread only. If you replace self.fig.canvas.draw by self.fig.canvas.invalidate, your script should work. Then the secondary thread merely informs the main thread that it should redraw itself, while the main thread does the actual drawing. I haven't done much testing with threads, so you may still run into some bugs; please let us know if you do.

When I try to use invalidate I get

AttributeError: 'FigureCanvasTkAgg' object has no attribute 'invalidate'

I tried in both mpl 1.4.3 and 1.5.0

What does matplotlib.get_backend() return? invalidate is an OSX backend specific thing, I would not expect it to work with TkAgg.

You could use self.fig.canvas.draw_idle (which calls self.fig.canvas.invalidate on MacOSX backend). The draw_idle function is defined for all backends. In practice, this seems to work for the MacOSX backend only; tkagg gives "RuntimeError: main thread is not in main loop" and gtkagg gives a segmentation fault.

Ah ok. Well obviously I want to write platform and backend independent code.

Wasn't draw_idle supposed to be deprecated?

@nils-werner

@nils-werner Actually can you provide an example script that shows how you want to use threads in practice? From your current example, it's not clear to me what you are trying to achieve.

Pull request #5562 provides a comprehensive fix for the first issue (October 15).
For the second issue (November 2), the following script should work:

import threading
import time
import sys
import numpy as np
import matplotlib
matplotlib.use('MacOSX')
matplotlib.interactive(False)

import matplotlib.pyplot as plt


class async_worker(threading.Thread):
    def __init__(self, fig, axis):
        threading.Thread.__init__(self)
        self.fig = fig
        self.axis = axis

    def run(self):
        i = 0
        while True:
            self.axis.clear()
            self.axis.plot(np.random.rand(251, 1))
            self.fig.canvas.draw_idle()
            print "Run thread %d" % i
            sys.stdout.flush()
            i += 1
            time.sleep(0.1)


def main():
    fig = plt.figure()
    ax = fig.add_subplot(1, 1, 1)
    async_worker(fig, ax).start()
    plt.show()

if __name__ == "__main__":
    main()

So far as I can tell the original issue now works in master on my mac, so I'm going to close this one.

Navigation Menu

Repository navigation

Cannot use threading with Mac OSX Backend #5251

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Cannot use threading with Mac OSX Backend #5251

Description

Activity

mdboom commented on Oct 15, 2015

mdehoon commented on Oct 16, 2015

nils-werner commented on Oct 16, 2015

mdehoon commented on Oct 16, 2015

nils-werner commented on Oct 16, 2015

mdehoon commented on Oct 16, 2015

nils-werner commented on Oct 16, 2015

WeatherGod commented on Oct 16, 2015

nils-werner commented on Oct 16, 2015

mdehoon commented on Oct 16, 2015

nils-werner commented on Oct 19, 2015

mdehoon commented on Oct 19, 2015

mdehoon commented on Oct 23, 2015

tacaswell commented on Oct 24, 2015

mdehoon commented on Oct 25, 2015

nils-werner commented on Nov 2, 2015

mdehoon commented on Nov 7, 2015

nils-werner commented on Nov 7, 2015

tacaswell commented on Nov 7, 2015

mdehoon commented on Nov 8, 2015

nils-werner commented on Nov 12, 2015

mdehoon commented on Nov 13, 2015

mdehoon commented on Nov 25, 2015

jklymak commented on Aug 16, 2018

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions