cx_freeze converted GUI-app (tkinter) crashes afte

Question

I've been dealing with this for days now and hope to finde some help. I developed a GUI-application with imported modules tkinter, numpy, scipy, matplotlib, which runs fine in python itself. After having converted to an *.exe everything works as expected, but NOT the matplotlib section. When I press my defined plot-Button, the *.exe simply closes and doesn't show any plots. So I thought to make a minimalistic example, where I simply plot a sin-function and I'm facing the same issue: Works perfect in python, when converting it to an *.exe it crashes when pressing the plot Button. The minimalistic example is here:

import tkinter as tk
import matplotlib.pyplot as plt
import numpy as np

class MainWindow(tk.Frame):
    def __init__(self):
        tk.Frame.__init__(self,bg='#9C9C9C',relief="flat", bd=10)
        self.place(width=x,height=y)
        self.create_widgets()

    def function(self):
        datax = np.arange(-50,50,0.1)
        datay = np.sin(datax)
        plt.plot(datax,datay)
        plt.show()

    def create_widgets(self):
        plot = tk.Button(self, text='PLOT', command=self.function)
        plot.pack()


x,y=120,300
root=tk.Tk()
root.geometry(str(x)+"x"+str(y))
app = MainWindow()
app.mainloop()

And see my corresponding "setup.py" for converting with cx_freeze.

import cx_Freeze
import matplotlib
import sys
import numpy
import tkinter

base = None

if sys.platform == "win32":
    base = "Win32GUI"

executables = [cx_Freeze.Executable("test.py", base=base)]


build_exe_options = {"includes":   ["matplotlib.backends.backend_tkagg","matplotlib.pyplot",
                             "tkinter.filedialog","numpy"],
                     "include_files":[(matplotlib.get_data_path(), "mpl-data")],
                     "excludes":[],
                    }

cx_Freeze.setup(
    name = "test it",
    options = {"build_exe": build_exe_options},
    version = "1.0",
    description = "I test it",
    executables = executables)

Any ideas that might solve the issue are highly appreciated. I'm working on a 64-bit Windows10 machine and I'm using the WinPython Distribution with Python 3.4.3.

Answer 1

Check if you have numpy+mkl package installed. Uninstalling numpy and installing numpy+mkl package solved my issue of getting error related to mkl_intel_thread.dll

Answer 2

I have followed @J.J. Hakala's answer, but I found that it's not necessary copy all mkl_*.dll and libiomp5md.dll files. For me it worked with libiomp5md.dll mkl_core.dll mkl_def.dll mkl_intel_thread.dll. This helps to reduce the final bundle size in ~500MB.

Also, you can include the files you want to copy in the include_files option. You also could only want to include them if sys.platform is win32.

I'm using Anaconda as well as @Matt Williams, so, changing a bit the OP's code:

import cx_Freeze
import matplotlib
import sys
import numpy
import tkinter
import os

PYTHON_INSTALL_DIR = os.path.dirname(os.path.dirname(os.__file__))

build_exe_options = {"includes":   ["matplotlib.backends.backend_tkagg","matplotlib.pyplot",
                             "tkinter.filedialog","numpy"],
                     "include_files":[(matplotlib.get_data_path(), "mpl-data")],
                     "excludes":[],
                    }

base = None

if sys.platform == "win32":
    base = "Win32GUI"
    DLLS_FOLDER = os.path.join(PYTHON_INSTALL_DIR, 'Library', 'bin')

    dependencies = ['libiomp5md.dll', 'mkl_core.dll', 'mkl_def.dll', 'mkl_intel_thread.dll']

    for dependency in dependencies:
        build_exe_options['include_files'].append(os.path.join(DLLS_FOLDER, dependency))

executables = [cx_Freeze.Executable("test.py", base=base)]

cx_Freeze.setup(
    name = "test it",
    options = {"build_exe": build_exe_options},
    version = "1.0",
    description = "I test it",
    executables = executables)

Answer 3

I found a potential solution (or at least an explanation) for this problem while testing PyInstaller with the same test.py. I received error message about a dll file being missing, that file being mkl_intel_thread.dll.

I searched for that file and it was found inside numpy folder. I copied files matching mkl_*.dll and also libiomp5md.dll to the same directory where the test.exe created by python setup.py build was. After this the minimal test.exe showed the matplotlib window when pressing the plot button.

The files were located in folder lib\site-packages\numpy\core.

Answer 4

I really wanted to post this as a comment, but I don't have the reputation. This is mostly a followup to J.J. Hakala's answer about how to find the cause.

If one changes the base to "Console", i.e. using

base = "Console"

rather than

base = "Win32GUI"

a console will also pop up when the program starts and this error is printed

Intel MKL FATAL ERROR: Cannot load mkl_intel_thread.dll.

Which can help finding the cause of the problem pretty faster.

I thought this would be worth mentioning, since this trick can also be helpful to diagnose other problems. In the final release, one can revert back to Win32GUI to avoid the extra console. I should give the credits to this other stackoverflow post