PyPlot crashes Julia 1.0.1 on macOS Mojave

[portig]

Using a Python 3.7 installation with matplotlib via MacPorts. Below is the output demonstrating the crash. The same type of installation on a different mac running macOS High Sierra does not have this issue. When using a nightly build of Julia 1.1.0 the "using PyPlot" command finishes with the message "Warning: No working GUI backend found for matplotlib".

I did try to use the Conda that is installed by Julia when setting ENV["PYTHON"] = "", and it caused a similar crash to what's shown here.

Starting Julia 1.0.1 . . .
julia> using PyPlot

julia> plot(rand(2,2))
2018-11-03 22:04:13.610 julia[25104:1185443] -[NSApplication _setup:]: unrecognized selector sent to instance 0x7fea7671cce0
2018-11-03 22:04:13.611 julia[25104:1185443] *** Terminating app due to uncaught exception 'NSInvalidArgumentException', reason: '-[NSApplication _setup:]: unrecognized selector sent to instance 0x7fea7671cce0'
*** First throw call stack:
(
0 CoreFoundation 0x00007fff37eb2e65 __exceptionPreprocess + 256
1 libobjc.A.dylib 0x00007fff63f0d720 objc_exception_throw + 48
2 CoreFoundation 0x00007fff37f3022d -[NSObject(NSObject) __retain_OA] + 0
3 CoreFoundation 0x00007fff37e54820 forwarding + 1486
4 CoreFoundation 0x00007fff37e541c8 _CF_forwarding_prep_0 + 120
5 libtk8.6.dylib 0x000000013452231d TkpInit + 413
6 libtk8.6.dylib 0x000000013447a17e Initialize + 2622
7 _tkinter.cpython-37m-darwin.so 0x00000001342a2a0f _tkinter_create + 1183
8 libpython3.7m.dylib 0x00000001115922b9 _PyMethodDef_RawFastCallKeywords + 681
9 libpython3.7m.dylib 0x00000001115923bd _PyCFunction_FastCallKeywords + 45
10 libpython3.7m.dylib 0x000000011169bc4b call_function + 603
11 libpython3.7m.dylib 0x0000000111692e45 _PyEval_EvalFrameDefault + 7893
12 libpython3.7m.dylib 0x0000000111690ad9 _PyEval_EvalCodeWithName + 3305
13 libpython3.7m.dylib 0x00000001115911b1 _PyFunction_FastCallDict + 481
14 libpython3.7m.dylib 0x0000000111592ba4 _PyObject_Call_Prepend + 164
15 libpython3.7m.dylib 0x00000001115fa5aa slot_tp_init + 298
16 libpython3.7m.dylib 0x0000000111603ca7 type_call + 295
17 libpython3.7m.dylib 0x0000000111591e73 _PyObject_FastCallKeywords + 691
18 libpython3.7m.dylib 0x000000011169bcb4 call_function + 708
19 libpython3.7m.dylib 0x0000000111692f6c _PyEval_EvalFrameDefault + 8188
20 libpython3.7m.dylib 0x0000000111591a6e function_code_fastcall + 254
21 libpython3.7m.dylib 0x000000011169bcbe call_function + 718
22 libpython3.7m.dylib 0x0000000111692e45 _PyEval_EvalFrameDefault + 7893
23 libpython3.7m.dylib 0x0000000111690ad9 _PyEval_EvalCodeWithName + 3305
24 libpython3.7m.dylib 0x00000001115911b1 _PyFunction_FastCallDict + 481
25 libpython3.7m.dylib 0x0000000111592ba4 _PyObject_Call_Prepend + 164
26 libpython3.7m.dylib 0x00000001115945b4 method_call + 36
27 libpython3.7m.dylib 0x00000001115925b6 PyObject_Call + 246
28 libpython3.7m.dylib 0x00000001116931b5 _PyEval_EvalFrameDefault + 8773
29 libpython3.7m.dylib 0x0000000111690ad9 _PyEval_EvalCodeWithName + 3305
30 libpython3.7m.dylib 0x00000001115911b1 _PyFunction_FastCallDict + 481
31 ??? 0x000000011117a3da 0x0 + 4581729242
32 ??? 0x000000011117fb17 0x0 + 4581751575
33 ??? 0x000000011119fca8 0x0 + 4581883048
34 libjulia.1.0.dylib 0x0000000103981ac3 jl_fptr_trampoline + 51
35 ??? 0x000000011119fbb6 0x0 + 4581882806
36 libjulia.1.0.dylib 0x00000001039924b2 jl_f__apply + 1218
37 libjulia.1.0.dylib 0x000000010399294f jl_f__apply_latest + 63
38 ??? 0x000000011119f785 0x0 + 4581881733
39 ??? 0x000000011119d5fe 0x0 + 4581873150
40 ??? 0x000000011119d724 0x0 + 4581873444
41 ??? 0x0000000111150bbc 0x0 + 4581559228
)
libc++abi.dylib: terminating with uncaught exception of type NSException

signal (6): Abort trap: 6
in expression starting at no file:0
__pthread_kill at /usr/lib/system/libsystem_kernel.dylib (unknown line)
Allocations: 24070144 (Pool: 24065397; Big: 4747); GC: 53
Abort trap: 6

[tkf]

Reading https://stackoverflow.com/a/34109240 it looks like it may be solved by installing a compatible matplotlib backend. See also #399 (comment)

Also, check it in Python without using Julia and PyPlot to see where the problem is.

[portig]

Thanks for your comments, @tkf.

When running the same Python 3.7 installation from the command line I have no problem using matplotlib. Pasting some of the examples from the matplotlib 3.0.0 tutorial into a Python session give correct results and the GUI buttons in the plot window work great.

[tkf]

Did you try activating Tk backend before importing PyPlot?

julia> using PyCall

julia> pygui(:tk)
:tk

julia> using PyPlot

If it does not work, you can try other backends https://matplotlib.org/tutorials/introductory/usage.html#what-is-a-backend

[portig]

When I try your commands, the response is that tk is not installed. With the command pygui() the response is ERROR: No supported Python GUI toolkit is installed..

When using pip to list the Python packages installed, both on the non-working Mojave system and the working High Sierra system, the list of packages and versions is identical.

Running Python from the command line, there is no problem using Matplotlib to produce graphics implying that some kind of backend is actually installed, however Julia is not able to make use of it.

[tkf]

You can use matplotlib.get_backend() to get the backend used by matplotlib.

As mentioned in the comment I linked above #399 (comment), it is known that macosx backend does not work with PyPlot. It looks like qt5 works.

[stevengj]

Running from Python, it uses the macosx backend by default, but it is really hard to get this working in Julia without losing the Julia event loop (#11).

You can edit your .matplotlibrc file or set the MPLBACKEND environment variable to change the default backend in both Julia and Python.

[stevengj]

Does the Conda installation work? i.e.

ENV["PYTHON"]=""  # use the Conda Python
Pkg.build("PyCall")
using PyPlot

?

[portig]

It turns out that command-line Python is using the MacOSX backend. Adding the environment variable MPLBACKEND=TkAgg caused Matplotlib in Python to no longer work. Eliminating that environment variable or changing it to MacOSX allowed Matplotlib in a Python session to work again.

After experimenting with MPLBACKEND, PyPlot in Julia still does not work, but differently than before: now, instead of crashing during a plot command, the using PyPlot command produces No working GUI backend found for matplotlib.

I then removed PyPlot and PyCall from the installation and replaced them while ENV["PYTHON"]="". After using PyCall the command pygui() now produced the response :tk, but after issuing the plot(rand(2,2) command, Julia crashed again with the following traceback:

julia> plot(rand(2,2))
2018-11-05 22:10:30.910 julia[29167:1988944] -[NSApplication _setup:]: unrecognized selector sent to instance 0x7fa432cf0770
2018-11-05 22:10:30.918 julia[29167:1988944] *** Terminating app due to uncaught exception 'NSInvalidArgumentException', reason: '-[NSApplication _setup:]: unrecognized selector sent to instance 0x7fa432cf0770'
*** First throw call stack:
(
	0   CoreFoundation                      0x00007fff37eb2e65 __exceptionPreprocess + 256
	1   libobjc.A.dylib                     0x00007fff63f0d720 objc_exception_throw + 48
	2   CoreFoundation                      0x00007fff37f3022d -[NSObject(NSObject) __retain_OA] + 0
	3   CoreFoundation                      0x00007fff37e54820 ___forwarding___ + 1486
	4   CoreFoundation                      0x00007fff37e541c8 _CF_forwarding_prep_0 + 120
	5   libtk8.6.dylib                      0x00000001295e331d TkpInit + 413
	6   libtk8.6.dylib                      0x000000012953b17e Initialize + 2622
	7   _tkinter.cpython-37m-darwin.so      0x0000000129363a0f _tkinter_create + 1183
	8   libpython3.7m.dylib                 0x0000000128d162b9 _PyMethodDef_RawFastCallKeywords + 681
	9   libpython3.7m.dylib                 0x0000000128d163bd _PyCFunction_FastCallKeywords + 45
	10  libpython3.7m.dylib                 0x0000000128e1fc4b call_function + 603
	11  libpython3.7m.dylib                 0x0000000128e16e45 _PyEval_EvalFrameDefault + 7893
	12  libpython3.7m.dylib                 0x0000000128e14ad9 _PyEval_EvalCodeWithName + 3305
	13  libpython3.7m.dylib                 0x0000000128d151b1 _PyFunction_FastCallDict + 481
	14  libpython3.7m.dylib                 0x0000000128d16ba4 _PyObject_Call_Prepend + 164
	15  libpython3.7m.dylib                 0x0000000128d7e5aa slot_tp_init + 298
	16  libpython3.7m.dylib                 0x0000000128d87ca7 type_call + 295
	17  libpython3.7m.dylib                 0x0000000128d15e73 _PyObject_FastCallKeywords + 691
	18  libpython3.7m.dylib                 0x0000000128e1fcb4 call_function + 708
	19  libpython3.7m.dylib                 0x0000000128e16f6c _PyEval_EvalFrameDefault + 8188
	20  libpython3.7m.dylib                 0x0000000128d15a6e function_code_fastcall + 254
	21  libpython3.7m.dylib                 0x0000000128e1fcbe call_function + 718
	22  libpython3.7m.dylib                 0x0000000128e16e45 _PyEval_EvalFrameDefault + 7893
	23  libpython3.7m.dylib                 0x0000000128e14ad9 _PyEval_EvalCodeWithName + 3305
	24  libpython3.7m.dylib                 0x0000000128d151b1 _PyFunction_FastCallDict + 481
	25  libpython3.7m.dylib                 0x0000000128d16ba4 _PyObject_Call_Prepend + 164
	26  libpython3.7m.dylib                 0x0000000128d185b4 method_call + 36
	27  libpython3.7m.dylib                 0x0000000128d165b6 PyObject_Call + 246
	28  libpython3.7m.dylib                 0x0000000128e171b5 _PyEval_EvalFrameDefault + 8773
	29  libpython3.7m.dylib                 0x0000000128e14ad9 _PyEval_EvalCodeWithName + 3305
	30  libpython3.7m.dylib                 0x0000000128d151b1 _PyFunction_FastCallDict + 481
	31  ???                                 0x000000012ef01eba 0x0 + 5082455738
	32  ???                                 0x000000012ef020e7 0x0 + 5082456295
	33  ???                                 0x000000012ef5a0d8 0x0 + 5082816728
	34  libjulia.1.0.dylib                  0x000000010a3d2ac3 jl_fptr_trampoline + 51
	35  ???                                 0x000000012ef59fe6 0x0 + 5082816486
	36  libjulia.1.0.dylib                  0x000000010a3e34b2 jl_f__apply + 1218
	37  libjulia.1.0.dylib                  0x000000010a3e394f jl_f__apply_latest + 63
	38  ???                                 0x000000012ef59bb5 0x0 + 5082815413
	39  ???                                 0x000000012ef579be 0x0 + 5082806718
	40  ???                                 0x000000012ef57ae4 0x0 + 5082807012
	41  ???                                 0x000000012ef573d7 0x0 + 5082805207
	42  ???                                 0x000000010f7de2dc 0x0 + 4554875612
)
libc++abi.dylib: terminating with uncaught exception of type NSException

signal (6): Abort trap: 6
in expression starting at no file:0
__pthread_kill at /usr/lib/system/libsystem_kernel.dylib (unknown line)
Allocations: 62842323 (Pool: 62826961; Big: 15362); GC: 139
Abort trap: 6

This is basically the same response as when using the Python installation outside of Conda.

The Conda installed by Julia is version 3.7. Running pip list in its bin subdirectory gives the following packages as being installed:

./pip list
Package         Version   
--------------- ----------
asn1crypto      0.24.0    
certifi         2018.10.15
cffi            1.11.5    
chardet         3.0.4     
conda           4.5.11    
cryptography    2.3.1     
cycler          0.10.0    
idna            2.7       
kiwisolver      1.0.1     
matplotlib      3.0.1     
mkl-fft         1.0.6     
mkl-random      1.0.1     
numpy           1.15.3    
pip             10.0.1    
pycosat         0.6.3     
pycparser       2.18      
pyOpenSSL       18.0.0    
pyparsing       2.3.0     
PySocks         1.6.8     
python-dateutil 2.7.5     
pytz            2018.7    
requests        2.19.1    
ruamel-yaml     0.15.46   
setuptools      40.2.0    
six             1.11.0    
tornado         5.1.1     
urllib3         1.23      
wheel           0.31.1    
You are using pip version 10.0.1, however version 18.1 is available.
You should consider upgrading via the 'pip install --upgrade pip' command.

It turns out that the Python installed in Conda is not able to plot either. Executing the following commands:

./python
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
matplotlib.get_backend()

This leads to the output TkAgg
The command that crashes this Python session is fig, ax = plt.subplots() and leads to the following traceback:

2018-11-05 22:27:48.455 python[29521:2028138] -[NSApplication _setup:]: unrecognized selector sent to instance 0x7ffc9b49e890
2018-11-05 22:27:48.457 python[29521:2028138] *** Terminating app due to uncaught exception 'NSInvalidArgumentException', reason: '-[NSApplication _setup:]: unrecognized selector sent to instance 0x7ffc9b49e890'
*** First throw call stack:
(
	0   CoreFoundation                      0x00007fff37eb2e65 __exceptionPreprocess + 256
	1   libobjc.A.dylib                     0x00007fff63f0d720 objc_exception_throw + 48
	2   CoreFoundation                      0x00007fff37f3022d -[NSObject(NSObject) __retain_OA] + 0
	3   CoreFoundation                      0x00007fff37e54820 ___forwarding___ + 1486
	4   CoreFoundation                      0x00007fff37e541c8 _CF_forwarding_prep_0 + 120
	5   libtk8.6.dylib                      0x000000011c63731d TkpInit + 413
	6   libtk8.6.dylib                      0x000000011c58f17e Initialize + 2622
	7   _tkinter.cpython-37m-darwin.so      0x000000011c3b7a0f _tkinter_create + 1183
	8   python                              0x00000001087ce8d6 _PyMethodDef_RawFastCallKeywords + 230
	9   python                              0x000000010890eb87 call_function + 247
	10  python                              0x000000010890c6ba _PyEval_EvalFrameDefault + 45274
	11  python                              0x0000000108900332 _PyEval_EvalCodeWithName + 418
	12  python                              0x00000001087cd5a7 _PyFunction_FastCallDict + 231
	13  python                              0x0000000108851091 slot_tp_init + 193
	14  python                              0x000000010885b051 type_call + 241
	15  python                              0x00000001087ce2a3 _PyObject_FastCallKeywords + 179
	16  python                              0x000000010890ec24 call_function + 404
	17  python                              0x000000010890c7ae _PyEval_EvalFrameDefault + 45518
	18  python                              0x00000001087ce098 function_code_fastcall + 120
	19  python                              0x000000010890eb3e call_function + 174
	20  python                              0x000000010890c6ba _PyEval_EvalFrameDefault + 45274
	21  python                              0x0000000108900332 _PyEval_EvalCodeWithName + 418
	22  python                              0x00000001087cd5a7 _PyFunction_FastCallDict + 231
	23  python                              0x00000001087d14c2 method_call + 130
	24  python                              0x00000001087ceef4 PyObject_Call + 100
	25  python                              0x000000010890c905 _PyEval_EvalFrameDefault + 45861
	26  python                              0x0000000108900332 _PyEval_EvalCodeWithName + 418
	27  python                              0x00000001087cd5a7 _PyFunction_FastCallDict + 231
	28  python                              0x000000010890c905 _PyEval_EvalFrameDefault + 45861
	29  python                              0x0000000108900332 _PyEval_EvalCodeWithName + 418
	30  python                              0x00000001087ce7a3 _PyFunction_FastCallKeywords + 195
	31  python                              0x000000010890eb3e call_function + 174
	32  python                              0x000000010890c6ba _PyEval_EvalFrameDefault + 45274
	33  python                              0x0000000108900332 _PyEval_EvalCodeWithName + 418
	34  python                              0x000000010896722b PyRun_InteractiveOneObjectEx + 1147
	35  python                              0x0000000108963e11 PyRun_InteractiveLoopFlags + 321
	36  python                              0x0000000108963c60 PyRun_AnyFileExFlags + 64
	37  python                              0x0000000108990d3a pymain_main + 10938
	38  python                              0x00000001087a0a6d main + 125
	39  libdyld.dylib                       0x00007fff64fdc08d start + 1
)
libc++abi.dylib: terminating with uncaught exception of type NSException
Abort trap: 6

[tkf]

Please quote code blocks with a pair of three backticks (not one). It's very hard to read. 😉

It turns out that the Python installed in Conda is not able to plot either.

If it doesn't work in Python then it won't work in Julia. Did you try other backends? How about Qt5Agg? You may need to run using Conda; Conda.add("pyqt") in Julia (or conda install pyqt in system shell).

[portig]

Thank you, @tkf, for your suggestion to install pyqt into Conda. This solved the problem. Surprisingly your procedure appears to have added something to Conda that now allows both :tk and :qt5 to be usable, both in Julia and in Python.

[portig]

After the Conda installed by Julia allowed PyPlot to work I added the same packages that the Conda.add("pyqt") had added to my system-wide Python 3.7 installation. I then removed PyPlot from Julia and reinstalled it with ENV["PYTHON"] set to the executable of the system-wide Python. This worked also. Now I have removed Conda from the Julia installation going as far as deleting the ~/.julia/conda subdirectory, over 2 GB in size.

[divan]

Any workaround for this? I have it on the latest MacOS X 10.14.3 and latest Julia 1.1.0.

Julia Output

$ julia riemann_sum.jl
2019-03-03 16:45:27.929 julia[28988:737043] -[NSApplication _setup:]: unrecognized selector sent to instance 0x7fdfcd7dfcc0
2019-03-03 16:45:27.936 julia[28988:737043] *** Terminating app due to uncaught exception 'NSInvalidArgumentException', reason: '-[NSApplication _setup:]: unrecognized selector sent to instance 0x7fdfcd7dfcc0'
*** First throw call stack:
(
	0   CoreFoundation                      0x00007fff438edded __exceptionPreprocess + 256
	1   libobjc.A.dylib                     0x00007fff6f9b9720 objc_exception_throw + 48
	2   CoreFoundation                      0x00007fff4396b195 -[NSObject(NSObject) __retain_OA] + 0
	3   CoreFoundation                      0x00007fff4388fa60 ___forwarding___ + 1486
	4   CoreFoundation                      0x00007fff4388f408 _CF_forwarding_prep_0 + 120
	5   libtk8.6.dylib                      0x000000013635531d TkpInit + 413
	6   libtk8.6.dylib                      0x00000001362ad17e Initialize + 2622
	7   _tkinter.cpython-37m-darwin.so      0x00000001335e7a0f _tkinter_create + 1183
	8   libpython3.7m.dylib                 0x000000010f97c2b9 _PyMethodDef_RawFastCallKeywords + 681
	9   libpython3.7m.dylib                 0x000000010f97c3bd _PyCFunction_FastCallKeywords + 45
	10  libpython3.7m.dylib                 0x000000010fa85c4b call_function + 603
	11  libpython3.7m.dylib                 0x000000010fa7ce45 _PyEval_EvalFrameDefault + 7893
	12  libpython3.7m.dylib                 0x000000010fa7aad9 _PyEval_EvalCodeWithName + 3305
	13  libpython3.7m.dylib                 0x000000010f97b1b1 _PyFunction_FastCallDict + 481
	14  libpython3.7m.dylib                 0x000000010f97cba4 _PyObject_Call_Prepend + 164
	15  libpython3.7m.dylib                 0x000000010f9e45aa slot_tp_init + 298
	16  libpython3.7m.dylib                 0x000000010f9edca7 type_call + 295
	17  libpython3.7m.dylib                 0x000000010f97be73 _PyObject_FastCallKeywords + 691
	18  libpython3.7m.dylib                 0x000000010fa85cb4 call_function + 708
	19  libpython3.7m.dylib                 0x000000010fa7cf6c _PyEval_EvalFrameDefault + 8188
	20  libpython3.7m.dylib                 0x000000010f97ba6e function_code_fastcall + 254
	21  libpython3.7m.dylib                 0x000000010fa85cbe call_function + 718
	22  libpython3.7m.dylib                 0x000000010fa7ce45 _PyEval_EvalFrameDefault + 7893
	23  libpython3.7m.dylib                 0x000000010fa7aad9 _PyEval_EvalCodeWithName + 3305
	24  libpython3.7m.dylib                 0x000000010f97b1b1 _PyFunction_FastCallDict + 481
	25  libpython3.7m.dylib                 0x000000010f97cba4 _PyObject_Call_Prepend + 164
	26  libpython3.7m.dylib                 0x000000010f97e5b4 method_call + 36
	27  libpython3.7m.dylib                 0x000000010f97c5b6 PyObject_Call + 246
	28  libpython3.7m.dylib                 0x000000010fa7d1b5 _PyEval_EvalFrameDefault + 8773
	29  libpython3.7m.dylib                 0x000000010fa7aad9 _PyEval_EvalCodeWithName + 3305
	30  libpython3.7m.dylib                 0x000000010f97b1b1 _PyFunction_FastCallDict + 481
	31  ???                                 0x000000010f556b75 0x0 + 4552223605
	32  ???                                 0x000000010f5652cc 0x0 + 4552282828
	33  ???                                 0x000000010f56a4f8 0x0 + 4552303864
	34  libjulia.1.1.dylib                  0x0000000109d39b03 jl_fptr_trampoline + 51
	35  ???                                 0x000000010f56a406 0x0 + 4552303622
	36  libjulia.1.1.dylib                  0x0000000109d4a312 jl_f__apply + 1218
	37  libjulia.1.1.dylib                  0x0000000109d4a68f jl_f__apply_latest + 63
	38  ???                                 0x000000010f56a145 0x0 + 4552302917
	39  ???                                 0x000000010f56846c 0x0 + 4552295532
	40  ???                                 0x000000010f568584 0x0 + 4552295812
	41  ???                                 0x000000010f539d2c 0x0 + 4552105260
)
libc++abi.dylib: terminating with uncaught exception of type NSException
signal (6): Abort trap: 6

in expression starting at /Users/divan/src/test/julia/riemann_sum.jl:11

__pthread_kill at /usr/lib/system/libsystem_kernel.dylib (unknown line)

Allocations: 10715487 (Pool: 10714032; Big: 1455); GC: 23

Abort trap: 6

[mac@julia ]$ julia --version

julia version 1.1.0

[stevengj]

The workaround is to use a different matplotlib backend as mentioned above.