Julia code returns lists with Blender's numpy

[paulmelis]

Comparing the returned values of the subdivide() in catmull-clark.jl call under Blender and from regular Python I noticed the following. This is based on the code in test/subdiv.py, which is also stored in test.blend under (text) in the Blender Text Editor.

Python 3.8.5 (command-line)

melis@juggle 20:46:~/concepts/blender-julia-test/test$ python subdiv.py
/usr/lib/python3.8/site-packages/numpy/__init__.py
1.19.1
/home/melis/.local/lib/python3.8/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.533ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 21.688ms
<class 'numpy.ndarray'>
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ]
<class 'numpy.ndarray'>
[ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93]
<class 'numpy.ndarray'>
[4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4]
<class 'numpy.ndarray'>
[15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]

Blender 2.90 (Run script from UI)

/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/numpy/__init__.py
1.17.5
/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 0.018ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 0.127ms
<class 'list'>
[-0.5555555820465088, -0.5555555820465088, -0.5555555820465088, -0.5061728358268738, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.6049383282661438, -0.5061728358268738, -0.5061728358268738, 0.5665295124053955, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.5665295124053955, -0.44581618905067444, 0.5208048224449158, 0.520804762840271, 0.520804762840271, -1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 1.0, -0.75, -0.75, 0.0, -0.75, 0.0, 0.75, -0.75, 0.75, 0.0, -0.75, 0.0, -0.75, 0.0, 0.75, 0.75, 0.75, 0.75, 0.0, 0.0, 0.75, -0.75, 0.75, 0.0, 0.75, 0.75, -0.75, 0.0, 0.75, 0.0, -0.75, 0.0, -0.75, 0.75, 0.0, -0.75, -0.75]
<class 'list'>
[1, 5, 9, 13, 17, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 61, 65, 69, 73, 77, 81, 85, 89, 93]
<class 'list'>
[4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4]
<class 'list'>
[15, 2, 16, 9, 16, 4, 17, 9, 17, 3, 18, 9, 18, 1, 15, 9, 17, 4, 19, 10, 19, 8, 20, 10, 20, 7, 21, 10, 21, 3, 17, 10, 20, 8, 22, 11, 22, 6, 23, 11, 23, 5, 24, 11, 24, 7, 20, 11, 23, 6, 25, 12, 25, 2, 15, 12, 15, 1, 26, 12, 26, 5, 23, 12, 21, 7, 24, 13, 24, 5, 26, 13, 26, 1, 18, 13, 18, 3, 21, 13, 19, 4, 16, 14, 16, 2, 25, 14, 25, 6, 22, 14, 22, 8, 19, 14]

Summary

Python 3.8.5 (command-line):

• subdivide() returns a tuple of 4 numpy.ndarray values, as expected
• NumPy version 1.19.1
• PyJulia version 0.5.4

Blender 2.90 (UI):

• subdivide() returns a tuple of 4 lists
• Python 3.7.7 (as included in Blender)
• NumPy 1.17.5
• PyJulia 0.5.4

So the differences are probably caused by a different version for Python and/or NumPy.

Rebuilding PyCall from within Blender ((rebuild pycall) text tab) makes no difference.

[paulmelis]

Checking differences in julia.install() output.

Command-line Python:

melis@juggle 20:47:~/concepts/blender-julia-test/test$ python
Python 3.8.5 (default, Jul 27 2020, 08:42:51) 
[GCC 10.1.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import julia
>>> julia.install()
[ Info: Julia version info
Julia Version 1.5.0
Commit 96786e22cc* (2020-08-01 23:44 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
      "Arch Linux"
  uname: Linux 5.7.12-arch1-1 #1 SMP PREEMPT Fri, 31 Jul 2020 17:38:22 +0000 x86_64 unknown
  CPU: Intel(R) Core(TM) i5-4460  CPU @ 3.20GHz: 
              speed         user         nice          sys         idle          irq
       #1  3337 MHz     194500 s         83 s      85782 s    6606330 s      14217 s
       #2  3287 MHz     188803 s         67 s      67846 s    6627848 s      23495 s
       #3  3391 MHz     178286 s         71 s      54574 s    6655102 s       4923 s
       #4  3351 MHz     159383 s         72 s      49577 s    6696040 s       2958 s
       
  Memory: 7.712959289550781 GB (3372.6640625 MB free)
  Uptime: 69968.0 sec
  Load Avg:  0.521484375  0.4716796875  0.3701171875
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-10.0.1 (ORCJIT, haswell)
Environment:
  XDG_SESSION_PATH = /org/freedesktop/DisplayManager/Session0
  HOME = /home/melis
  XDG_SEAT_PATH = /org/freedesktop/DisplayManager/Seat0
  PYTHONPATH = /home/melis/projects/mvq-py-git:/home/melis/software/pywebkitgtk/lib64/python2.7/site-packages/:
  TERM = xterm-color
  CUDA_PATH = /opt/cuda
  MOZ_PLUGIN_PATH = /usr/lib/mozilla/plugins
  PATH = /home/melis/software/miniconda3/condabin:/home/melis/projects/packages:/home/melis/.local/bin:/home/melis/software/ParaView-5.5.2-Qt5-MPI-Linux-64bit/bin:/home/melis/scripts:/usr/local/bin:/usr/bin:/bin:/usr/local/sbin:/opt/cuda/bin:/usr/lib/emscripten:/usr/lib/jvm/default/bin:/usr/bin/site_perl:/usr/bin/vendor_perl:/usr/bin/core_perl
[ Info: Julia executable: /usr/bin/julia
[ Info: Trying to import PyCall...
┌ Info: PyCall is already installed and compatible with Python executable.
│ 
│ PyCall:
│     python: /usr/bin/python3
│     libpython: /usr/lib/libpython3.8.so.1.0
│ Python:
│     python: /usr/bin/python
└     libpython: /usr/lib/libpython3.8.so.1.0

Blender:

[ Info: Julia version info
Julia Version 1.5.0
Commit 96786e22cc* (2020-08-01 23:44 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
      "Arch Linux"
  uname: Linux 5.7.12-arch1-1 #1 SMP PREEMPT Fri, 31 Jul 2020 17:38:22 +0000 x86_64 unknown
  CPU: Intel(R) Core(TM) i5-4460  CPU @ 3.20GHz: 
              speed         user         nice          sys         idle          irq
       #1  3327 MHz     194747 s         83 s      85909 s    6610245 s      14246 s
       #2  3390 MHz     189044 s         67 s      67909 s    6631992 s      23502 s
       #3  3306 MHz     178639 s         71 s      54635 s    6659083 s       4931 s
       #4  3298 MHz     159463 s         72 s      49620 s    6700351 s       2962 s
       
  Memory: 7.712959289550781 GB (3357.9375 MB free)
  Uptime: 70013.0 sec
  Load Avg:  0.44189453125  0.4521484375  0.36669921875
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-10.0.1 (ORCJIT, haswell)
Environment:
  XDG_SESSION_PATH = /org/freedesktop/DisplayManager/Session0
  HOME = /home/melis
  XDG_SEAT_PATH = /org/freedesktop/DisplayManager/Seat0
  PYTHONPATH = /home/melis/projects/mvq-py-git:/home/melis/software/pywebkitgtk/lib64/python2.7/site-packages/:
  TERM = xterm-color
  CUDA_PATH = /opt/cuda
  MOZ_PLUGIN_PATH = /usr/lib/mozilla/plugins
  PATH = /home/melis/software/miniconda3/condabin:/home/melis/projects/packages:/home/melis/.local/bin:/home/melis/software/ParaView-5.5.2-Qt5-MPI-Linux-64bit/bin:/home/melis/scripts:/usr/local/bin:/usr/bin:/bin:/usr/local/sbin:/opt/cuda/bin:/usr/lib/emscripten:/usr/lib/jvm/default/bin:/usr/bin/site_perl:/usr/bin/vendor_perl:/usr/bin/core_perl
[ Info: Julia executable: /usr/bin/julia
[ Info: Trying to import PyCall...
┌ Warning: PyCall is already installed.  However, you may have trouble using
│ this Python executable because it is statically linked to libpython.
│ 
│ For more information, see:
│     https://pyjulia.readthedocs.io/en/latest/troubleshooting.html
│ 
│ Python executable:
│     /home/melis/software/blender-2.90.0-linux64/blender
│ Julia executable:
│     /usr/bin/julia
└ @ Main ~/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/julia/install.jl:90

So the fact that Blender links statically to libpython is definitly a difference that might be of influence here.

[paulmelis]

With a locally compiled version of Python 3.7.7 statically linked to libpython and NumPy 1.19.1 it works correctly:

melis@juggle 09:04:~/concepts/blender-julia-test/test$ ~/software/python-3.7.7-static/bin/python3.7m -m pip install -U numpy --user
Collecting numpy
  Downloading https://files.pythonhosted.org/packages/50/8f/29d5688614f9bba59931683d5d353d738d4a3007833219ee19c455732753/numpy-1.19.1-cp37-cp37m-manylinux2010_x86_64.whl (14.5MB)
     |████████████████████████████████| 14.5MB 5.7MB/s 
Installing collected packages: numpy
Successfully installed numpy-1.19.1
melis@juggle 09:04:~/concepts/blender-julia-test/test$ ~/software/python-3.7.7-static/bin/python3.7m -m pip install -U julia --user
Collecting julia
  Using cached https://files.pythonhosted.org/packages/af/d2/15aeddffc2d6799fef41552a9949b72983f59659cf68e94add1c6a3e2631/julia-0.5.4-py2.py3-none-any.whl
Installing collected packages: julia
Successfully installed julia-0.5.4
melis@juggle 09:04:~/concepts/blender-julia-test/test$ ~/software/python-3.7.7-static/bin/python3.7m subdiv.py 
/home/melis/.local/lib/python3.7/site-packages/numpy/__init__.py
1.19.1
/home/melis/.local/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.531ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 22.324ms
<class 'numpy.ndarray'>
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ]
<class 'numpy.ndarray'>
[ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93]
<class 'numpy.ndarray'>
[4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4]
<class 'numpy.ndarray'>
[15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]

So could be a NumPy version-related issue?

[paulmelis]

Python 3.8.5 (system version) with older NumPy 1.17.5 WORKS:

melis@juggle 09:15:/tmp$ virtualenv env
created virtual environment CPython3.8.5.final.0-64 in 3219ms
  creator CPython3Posix(dest=/tmp/env, clear=False, global=False)
  seeder FromAppData(download=False, pip=latest, setuptools=latest, wheel=latest, via=copy, app_data_dir=/home/melis/.local/share/virtualenv/seed-app-data/v1.0.1)
  activators BashActivator,CShellActivator,FishActivator,PowerShellActivator,PythonActivator,XonshActivator
melis@juggle 09:16:/tmp$ source env/bin/activate
(env) melis@juggle 09:16:/tmp$ which python
/tmp/env/bin/python
(env) melis@juggle 09:16:/tmp$ python
Python 3.8.5 (default, Jul 27 2020, 08:42:51) 
[GCC 10.1.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ModuleNotFoundError: No module named 'numpy'
>>> 
(env) melis@juggle 09:16:/tmp$ which pip
/tmp/env/bin/pip
(env) melis@juggle 09:16:/tmp$ pip install numpy==1.17.5
Collecting numpy==1.17.5
  Downloading numpy-1.17.5-cp38-cp38-manylinux1_x86_64.whl (20.5 MB)
     |████████████████████████████████| 20.5 MB 6.6 MB/s 
Installing collected packages: numpy
Successfully installed numpy-1.17.5
(env) melis@juggle 09:18:/tmp$ pip install julia
Collecting julia
  Using cached julia-0.5.4-py2.py3-none-any.whl (66 kB)
Installing collected packages: julia
Successfully installed julia-0.5.4

(env) melis@juggle 09:18:~$ cd concepts/blender-julia-test/test
(env) melis@juggle 09:18:~/concepts/blender-julia-test/test$ python subdiv.py 
/tmp/env/lib/python3.8/site-packages/numpy/__init__.py
1.17.5
/tmp/env/lib/python3.8/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.535ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 21.894ms
<class 'numpy.ndarray'>
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ]
<class 'numpy.ndarray'>
[ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93]
<class 'numpy.ndarray'>
[4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4]
<class 'numpy.ndarray'>
[15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]

Dang, so it's not just the older NumPy

[paulmelis]

Static Python 3.7.7 with older Numpy 1.17.5 WORKS:

melis@juggle 09:24:~/concepts/blender-julia-test/test$ ~/software/python-3.7.7-static/bin/python3.7m -m pip install -U numpy==1.17.5
Collecting numpy==1.17.5
  Downloading https://files.pythonhosted.org/packages/b1/51/20098150b6108061cb7542af3de7bfcfe0182bca21613697153e49dc4adc/numpy-1.17.5-cp37-cp37m-manylinux1_x86_64.whl (20.0MB)
     |████████████████████████████████| 20.0MB 6.5MB/s 
Installing collected packages: numpy
  Found existing installation: numpy 1.19.1
    Uninstalling numpy-1.19.1:
      Successfully uninstalled numpy-1.19.1
Successfully installed numpy-1.17.5
melis@juggle 09:24:~/concepts/blender-julia-test/test$ ~/software/python-3.7.7-static/bin/python3.7m 
Python 3.7.7 (default, Sep  4 2020, 09:00:20) 
[GCC 10.1.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy
>>> numpy.__version__
'1.17.5'
>>> 
melis@juggle 09:24:~/concepts/blender-julia-test/test$ ~/software/python-3.7.7-static/bin/python3.7m subdiv.py 
/home/melis/software/python-3.7.7-static/lib/python3.7/site-packages/numpy/__init__.py
1.17.5
/home/melis/.local/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.527ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 22.234ms
<class 'numpy.ndarray'>
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ]
<class 'numpy.ndarray'>
[ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93]
<class 'numpy.ndarray'>
[4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4]
<class 'numpy.ndarray'>
[15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]

[paulmelis]

Using the python3.7m interpreter included in Blender (3.7.7, statically linked to libpython, NumPy 1.17.5) FAILS:

melis@juggle 09:25:~/concepts/blender-julia-test/test$ ldd ~/software/blender-2.90.0-linux64/2.90/python/bin/python3.7m
	linux-vdso.so.1 (0x00007ffc432d7000)
	libcrypt.so.1 => /usr/lib/libcrypt.so.1 (0x00007f63760b8000)
	libpthread.so.0 => /usr/lib/libpthread.so.0 (0x00007f6376096000)
	libdl.so.2 => /usr/lib/libdl.so.2 (0x00007f6376090000)
	libutil.so.1 => /usr/lib/libutil.so.1 (0x00007f637608b000)
	libm.so.6 => /usr/lib/libm.so.6 (0x00007f6375f46000)
	libc.so.6 => /usr/lib/libc.so.6 (0x00007f6375d7f000)
	/lib64/ld-linux-x86-64.so.2 => /usr/lib64/ld-linux-x86-64.so.2 (0x00007f6376172000)
melis@juggle 09:25:~/concepts/blender-julia-test/test$ ~/software/blender-2.90.0-linux64/2.90/python/bin/python3.7m 
Python 3.7.7 (default, Jul  9 2020, 12:52:49) 
[GCC 9.3.1 20200408 (Red Hat 9.3.1-2)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy
>>> numpy.__version__
'1.17.5'
>>> 
melis@juggle 09:27:~/concepts/blender-julia-test/test$ ~/software/blender-2.90.0-linux64/2.90/python/bin/python3.7m subdiv.py 
/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/numpy/__init__.py
1.17.5
/home/melis/.local/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.547ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 22.337ms
<class 'list'>
[-0.5555555820465088, -0.5555555820465088, -0.5555555820465088, -0.5061728358268738, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.6049383282661438, -0.5061728358268738, -0.5061728358268738, 0.5665295124053955, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.5665295124053955, -0.44581618905067444, 0.5208048224449158, 0.520804762840271, 0.520804762840271, -1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 1.0, -0.75, -0.75, 0.0, -0.75, 0.0, 0.75, -0.75, 0.75, 0.0, -0.75, 0.0, -0.75, 0.0, 0.75, 0.75, 0.75, 0.75, 0.0, 0.0, 0.75, -0.75, 0.75, 0.0, 0.75, 0.75, -0.75, 0.0, 0.75, 0.0, -0.75, 0.0, -0.75, 0.75, 0.0, -0.75, -0.75]
<class 'list'>
[1, 5, 9, 13, 17, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 61, 65, 69, 73, 77, 81, 85, 89, 93]
<class 'list'>
[4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4]
<class 'list'>
[15, 2, 16, 9, 16, 4, 17, 9, 17, 3, 18, 9, 18, 1, 15, 9, 17, 4, 19, 10, 19, 8, 20, 10, 20, 7, 21, 10, 21, 3, 17, 10, 20, 8, 22, 11, 22, 6, 23, 11, 23, 5, 24, 11, 24, 7, 20, 11, 23, 6, 25, 12, 25, 2, 15, 12, 15, 1, 26, 12, 26, 5, 23, 12, 21, 7, 24, 13, 24, 5, 26, 13, 26, 1, 18, 13, 18, 3, 21, 13, 19, 4, 16, 14, 16, 2, 25, 14, 25, 6, 22, 14, 22, 8, 19, 14]

[paulmelis]

Comparing python binaries linking statically to libpython:

# Blender
melis@juggle 09:28:~/concepts/blender-julia-test/test$ ldd ~/software/blender-2.90.0-linux64/2.90/python/bin/python3.7m
	linux-vdso.so.1 (0x00007ffc432d7000)
	libcrypt.so.1 => /usr/lib/libcrypt.so.1 (0x00007f63760b8000)
	libpthread.so.0 => /usr/lib/libpthread.so.0 (0x00007f6376096000)
	libdl.so.2 => /usr/lib/libdl.so.2 (0x00007f6376090000)
	libutil.so.1 => /usr/lib/libutil.so.1 (0x00007f637608b000)
	libm.so.6 => /usr/lib/libm.so.6 (0x00007f6375f46000)
	libc.so.6 => /usr/lib/libc.so.6 (0x00007f6375d7f000)
	/lib64/ld-linux-x86-64.so.2 => /usr/lib64/ld-linux-x86-64.so.2 (0x00007f6376172000)
# Manually built
melis@juggle 09:28:~/concepts/blender-julia-test/test$ ldd ~/software/python-3.7.7-static/bin/python3.7m
	linux-vdso.so.1 (0x00007ffddf1f5000)
	libcrypt.so.1 => /usr/lib/libcrypt.so.1 (0x00007f37fe5b4000)
	libpthread.so.0 => /usr/lib/libpthread.so.0 (0x00007f37fe592000)
	libdl.so.2 => /usr/lib/libdl.so.2 (0x00007f37fe58c000)
	libutil.so.1 => /usr/lib/libutil.so.1 (0x00007f37fe587000)
	libm.so.6 => /usr/lib/libm.so.6 (0x00007f37fe442000)
	libc.so.6 => /usr/lib/libc.so.6 (0x00007f37fe27b000)
	/lib64/ld-linux-x86-64.so.2 => /usr/lib64/ld-linux-x86-64.so.2 (0x00007f37fe9ae000)

I.e. same set of dynamic libs used

[paulmelis]

Manually built Python 3.7.7 using Blender's site-packages (i.e. numpy and pyjulia) FAILS:

melis@juggle 09:31:~/concepts/blender-julia-test/test$ PYTHONPATH=$HOME/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages ~/software/python-3.7.7-static/bin/python3.7m
Python 3.7.7 (default, Sep  4 2020, 09:00:20) 
[GCC 10.1.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy
>>> numpy.__file__
'/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/numpy/__init__.py'
>>> 
melis@juggle 09:34:~/concepts/blender-julia-test/test$ PYTHONPATH=$HOME/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages ~/software/python-3.7.7-static/bin/python3.7m subdiv.py 
/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/numpy/__init__.py
1.17.5
/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.487ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 21.800ms
<class 'list'>
[-0.5555555820465088, -0.5555555820465088, -0.5555555820465088, -0.5061728358268738, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.6049383282661438, -0.5061728358268738, -0.5061728358268738, 0.5665295124053955, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.5665295124053955, -0.44581618905067444, 0.5208048224449158, 0.520804762840271, 0.520804762840271, -1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 1.0, -0.75, -0.75, 0.0, -0.75, 0.0, 0.75, -0.75, 0.75, 0.0, -0.75, 0.0, -0.75, 0.0, 0.75, 0.75, 0.75, 0.75, 0.0, 0.0, 0.75, -0.75, 0.75, 0.0, 0.75, 0.75, -0.75, 0.0, 0.75, 0.0, -0.75, 0.0, -0.75, 0.75, 0.0, -0.75, -0.75]
<class 'list'>
[1, 5, 9, 13, 17, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 61, 65, 69, 73, 77, 81, 85, 89, 93]
<class 'list'>
[4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4]
<class 'list'>
[15, 2, 16, 9, 16, 4, 17, 9, 17, 3, 18, 9, 18, 1, 15, 9, 17, 4, 19, 10, 19, 8, 20, 10, 20, 7, 21, 10, 21, 3, 17, 10, 20, 8, 22, 11, 22, 6, 23, 11, 23, 5, 24, 11, 24, 7, 20, 11, 23, 6, 25, 12, 25, 2, 15, 12, 15, 1, 26, 12, 26, 5, 23, 12, 21, 7, 24, 13, 24, 5, 26, 13, 26, 1, 18, 13, 18, 3, 21, 13, 19, 4, 16, 14, 16, 2, 25, 14, 25, 6, 22, 14, 22, 8, 19, 14]

So there really appears to be something different in the numpy included with Blender (as pyjulia is the same).

[paulmelis]

Blender's numpy is patched, but nothing of importance here.

Numpy build script in Blender also doesn't reveal anything noteworthy.

[paulmelis]

Running Blender's python3.7m binary with the site-packages of the manually built Python 3.7.7 WORKS:

melis@juggle 09:40:~/concepts/blender-julia-test/test$ PYTHONPATH=$HOME/software/python-3.7.7-static/lib/python3.7/site-packages/ ~/software/blender-2.90.0-linux64/2.90/python/bin/python3.7m subdiv.py 
/home/melis/software/python-3.7.7-static/lib/python3.7/site-packages/numpy/__init__.py
1.17.5
/home/melis/.local/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.494ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 22.240ms
<class 'numpy.ndarray'>
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ]
<class 'numpy.ndarray'>
[ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93]
<class 'numpy.ndarray'>
[4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4]
<class 'numpy.ndarray'>
[15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]

So that's pretty definitive, it is the numpy in Blender that's causing this.

[paulmelis]

Doing the same trick with Blender itself (i.e. override PYTHONPATH to point to the site-packages of the manually built Python 3.7.7) WORKS!

melis@juggle 09:44:~/concepts/blender-julia-test$ PYTHONPATH=$HOME/software/python-3.7.7-static/lib/python3.7/site-packages/ ~/software/blender-2.90.0-linux64/blender --python-use-system-env
Read prefs: /home/melis/.config/blender/2.90/config/userpref.blend
found bundled python: /home/melis/software/blender-2.90.0-linux64/2.90/python
Read blend: /home/melis/concepts/blender-julia-test/test.blend
Initializing Julia (this might take a moment the first time)...
Done!
(Blender) Preparing data took 0.499ms
(Julia) Building half edges done in 1.479ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 21.253ms
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ] [ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93] [4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4] [15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]
<class 'numpy.ndarray'>
(Blender) Call to Julia subdivision took 2110.277ms
(Blender) Updating subdivided mesh took 0.733ms
(Blender) Total time: 2111.508ms
Info: Deleted 1 object(s)

Initializing Julia (this might take a moment the first time)...
Done!
(Blender) Preparing data took 0.042ms
(Julia) Building half edges done in 0.019ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 0.158ms
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ] [ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93] [4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4] [15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]
<class 'numpy.ndarray'>
(Blender) Call to Julia subdivision took 211.427ms
(Blender) Updating subdivided mesh took 0.602ms
(Blender) Total time: 212.071ms

Getting back numpy arrays instead of lists from Julia also provides (as expected) a massive speedup. Here's the Bunny model:

Initializing Julia (this might take a moment the first time)...
Done!
(Blender) Preparing data took 1.863ms
(Julia) Building half edges done in 79.709ms
(Julia) Input: 35947 vertices, 69451 polygons, 104288 polygon edges
(Julia) Output: 209686 vertices, 208353 quads
(Julia) Subdivision done in 54.399ms
[-0.26612237  0.04188797  0.94303054 ... -0.5553862  -0.2588799
  0.0315131 ] [     1      5      9 ... 833401 833405 833409] [4 4 4 ... 4 4 4] [105399  21216 105400 ...  17278 209686 105398]
<class 'numpy.ndarray'>
(Blender) Call to Julia subdivision took 351.701ms
(Blender) Updating subdivided mesh took 269.674ms
(Blender) Total time: 623.238ms

So 623.674ms total instead of 1379.705ms as reported in the readme. Another factor of 2.2x faster, wow. So compared to the subsurf modifier time of 13583.849ms the Julia version would be 21.8x faster...

[paulmelis]

Hmm, this might still be something with the julia package as installed in Blender. Doing the subdiv.py test from Blender uses julia from .local and WORKS:

/home/melis/software/python-3.7.7-static/lib/python3.7/site-packages/numpy/__init__.py
1.17.5
/home/melis/.local/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 0.018ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 0.150ms
<class 'numpy.ndarray'>
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ]
<class 'numpy.ndarray'>
[ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93]
<class 'numpy.ndarray'>
[4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4]
<class 'numpy.ndarray'>
[15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]

Removing julia from Blender's site-packages are retesting from within Blender using the site-packages from the manually built Python WORKS:

melis@juggle 09:55:~/concepts/blender-julia-test$ rm -rf ~/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/julia*
melis@juggle 09:55:~/concepts/blender-julia-test$ PYTHONPATH=$HOME/software/python-3.7.7-static/lib/python3.7/site-packages/ ~/software/blender-2.90.0-linux64/blender --python-use-system-env
Read prefs: /home/melis/.config/blender/2.90/config/userpref.blend
found bundled python: /home/melis/software/blender-2.90.0-linux64/2.90/python
Read blend: /home/melis/concepts/blender-julia-test/test.blend
/home/melis/software/python-3.7.7-static/lib/python3.7/site-packages/numpy/__init__.py
1.17.5
/home/melis/.local/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.488ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 21.158ms
<class 'numpy.ndarray'>
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ]
<class 'numpy.ndarray'>
[ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93]
<class 'numpy.ndarray'>
[4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4]
<class 'numpy.ndarray'>
[15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]

So the julia package isn't of influence. Hmmm, just to make sure running subdiv.py test from within Blender but allowing Blender to pick up the julia package from .local (but still using Blender's own numpy) FAILS:

melis@juggle 09:57:~/concepts/blender-julia-test$ ~/software/blender-2.90.0-linux64/blender --python-use-system-env
Read prefs: /home/melis/.config/blender/2.90/config/userpref.blend
found bundled python: /home/melis/software/blender-2.90.0-linux64/2.90/python
Read blend: /home/melis/concepts/blender-julia-test/test.blend
/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/numpy/__init__.py
1.17.5
/home/melis/.local/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.578ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 22.457ms
<class 'list'>
[-0.5555555820465088, -0.5555555820465088, -0.5555555820465088, -0.5061728358268738, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.6049383282661438, -0.5061728358268738, -0.5061728358268738, 0.5665295124053955, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.5665295124053955, -0.44581618905067444, 0.5208048224449158, 0.520804762840271, 0.520804762840271, -1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 1.0, -0.75, -0.75, 0.0, -0.75, 0.0, 0.75, -0.75, 0.75, 0.0, -0.75, 0.0, -0.75, 0.0, 0.75, 0.75, 0.75, 0.75, 0.0, 0.0, 0.75, -0.75, 0.75, 0.0, 0.75, 0.75, -0.75, 0.0, 0.75, 0.0, -0.75, 0.0, -0.75, 0.75, 0.0, -0.75, -0.75]
<class 'list'>
[1, 5, 9, 13, 17, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 61, 65, 69, 73, 77, 81, 85, 89, 93]
<class 'list'>
[4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4]
<class 'list'>
[15, 2, 16, 9, 16, 4, 17, 9, 17, 3, 18, 9, 18, 1, 15, 9, 17, 4, 19, 10, 19, 8, 20, 10, 20, 7, 21, 10, 21, 3, 17, 10, 20, 8, 22, 11, 22, 6, 23, 11, 23, 5, 24, 11, 24, 7, 20, 11, 23, 6, 25, 12, 25, 2, 15, 12, 15, 1, 26, 12, 26, 5, 23, 12, 21, 7, 24, 13, 24, 5, 26, 13, 26, 1, 18, 13, 18, 3, 21, 13, 19, 4, 16, 14, 16, 2, 25, 14, 25, 6, 22, 14, 22, 8, 19, 14]

So still points to Blender's numpy being at fault.

[paulmelis]

Upgrading Blender's numpy using pip (and reinstall a local copy of pyjulia) fixes this for now:

melis@juggle 09:59:~/concepts/blender-julia-test$ ~/software/blender-2.90.0-linux64/2.90/python/bin/python3.7m -m pip install numpy
Collecting numpy
  Using cached https://files.pythonhosted.org/packages/50/8f/29d5688614f9bba59931683d5d353d738d4a3007833219ee19c455732753/numpy-1.19.1-cp37-cp37m-manylinux2010_x86_64.whl
Installing collected packages: numpy
Successfully installed numpy-1.19.1
melis@juggle 10:01:~/concepts/blender-julia-test$ ~/software/blender-2.90.0-linux64/blender
Read prefs: /home/melis/.config/blender/2.90/config/userpref.blend
found bundled python: /home/melis/software/blender-2.90.0-linux64/2.90/python
Read blend: /home/melis/concepts/blender-julia-test/test.blend
/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/numpy/__init__.py
1.19.1
Traceback (most recent call last):
  File "/home/melis/concepts/blender-julia-test/test.blend/(test)", line 6, in <module>
ModuleNotFoundError: No module named 'julia'
Error: Python script failed, check the message in the system console

Saved session recovery to '/tmp/quit.blend'

Blender quit
melis@juggle 10:01:~/concepts/blender-julia-test$ ~/software/blender-2.90.0-linux64/2.90/python/bin/python3.7m -m pip install julia
Requirement already satisfied: julia in /home/melis/.local/lib/python3.7/site-packages (0.5.4)
melis@juggle 10:01:~/concepts/blender-julia-test$ ~/software/blender-2.90.0-linux64/2.90/python/bin/python3.7m -m pip install -U julia
Requirement already up-to-date: julia in /home/melis/.local/lib/python3.7/site-packages (0.5.4)
melis@juggle 10:02:~/concepts/blender-julia-test$ ~/software/blender-2.90.0-linux64/2.90/python/bin/python3.7m -m pip install -U julia --force-reinstall
Collecting julia
  Using cached https://files.pythonhosted.org/packages/af/d2/15aeddffc2d6799fef41552a9949b72983f59659cf68e94add1c6a3e2631/julia-0.5.4-py2.py3-none-any.whl
Installing collected packages: julia
  Found existing installation: julia 0.5.4
    Uninstalling julia-0.5.4:
      Successfully uninstalled julia-0.5.4
Successfully installed julia-0.5.4
melis@juggle 10:02:~/concepts/blender-julia-test$ ls ~/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/
certifi  chardet  idna  julia  julia-0.5.4.dist-info  libextern_draco.so  numpy  numpy-1.19.1.dist-info  numpy.libs  requests  urllib3
melis@juggle 10:02:~/concepts/blender-julia-test$ ~/software/blender-2.90.0-linux64/blender
Read prefs: /home/melis/.config/blender/2.90/config/userpref.blend
found bundled python: /home/melis/software/blender-2.90.0-linux64/2.90/python
Read blend: /home/melis/concepts/blender-julia-test/test.blend
/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/numpy/__init__.py
1.19.1
/home/melis/software/blender-2.90.0-linux64/2.90/python/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.445ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 21.566ms
<class 'numpy.ndarray'>
[-0.5555556  -0.5555556  -0.5555556  -0.50617284 -0.50617284  0.6049383
 -0.50617284  0.6049383  -0.50617284 -0.4458162   0.5665295   0.5665295
  0.6049383  -0.50617284 -0.50617284  0.5665295  -0.4458162   0.5665295
  0.5665295   0.5665295  -0.4458162   0.5208048   0.52080476  0.52080476
 -1.          0.          0.          0.          1.          0.
  1.          0.          0.          0.         -1.          0.
  0.          0.         -1.          0.          0.          1.
 -0.75       -0.75        0.         -0.75        0.          0.75
 -0.75        0.75        0.         -0.75        0.         -0.75
  0.          0.75        0.75        0.75        0.75        0.
  0.          0.75       -0.75        0.75        0.          0.75
  0.75       -0.75        0.          0.75        0.         -0.75
  0.         -0.75        0.75        0.         -0.75       -0.75      ]
<class 'numpy.ndarray'>
[ 1  5  9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93]
<class 'numpy.ndarray'>
[4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4]
<class 'numpy.ndarray'>
[15  2 16  9 16  4 17  9 17  3 18  9 18  1 15  9 17  4 19 10 19  8 20 10
 20  7 21 10 21  3 17 10 20  8 22 11 22  6 23 11 23  5 24 11 24  7 20 11
 23  6 25 12 25  2 15 12 15  1 26 12 26  5 23 12 21  7 24 13 24  5 26 13
 26  1 18 13 18  3 21 13 19  4 16 14 16  2 25 14 25  6 22 14 22  8 19 14]

But has the downside of using a non-standard installation of Blender.

[paulmelis]

Well, what do you know? Manually building NumPy 1.17.5 with the manually built static Python 3.3.7 FAILS:

melis@juggle 20:48:~/c$ wget https://github.com/numpy/numpy/archive/v1.17.5.tar.gz -O numpy-1.17.5.tar.gz
melis@juggle 20:49:~/c$ tar xf numpy-1.17.5.tar.gz 
melis@juggle 20:49:~/c$ cd numpy-1.17.5
melis@juggle 20:49:~/c/numpy-1.17.5$ ~/software/python-3.7.7-static/bin/python3 setup.py build
...
gcc -pthread -shared build/temp.linux-x86_64-3.7/numpy/random/mtrand.o build/temp.linux-x86_64-3.7/numpy/random/src/legacy/legacy-distributions.o build/temp.linux-x86_64-3.7/numpy/random/src/distributions/logfactorial.o build/temp.linux-x86_64-3.7/numpy/random/src/distributions/distributions.o -Lbuild/temp.linux-x86_64-3.7 -lm -o build/lib.linux-x86_64-3.7/numpy/random/mtrand.cpython-37m-x86_64-linux-gnu.so
...
melis@juggle 20:52:~/c/numpy-1.17.5$ cd build/lib.linux-x86_64-3.7/
melis@juggle 20:53:~/c/numpy-1.17.5/build/lib.linux-x86_64-3.7$ ~/software/python-3.7.7-static/bin/python3
Python 3.7.7 (default, Sep  4 2020, 09:00:20) 
[GCC 10.1.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy
>>> numpy.__file__
'/home/melis/c/numpy-1.17.5/build/lib.linux-x86_64-3.7/numpy/__init__.py'
>>> 
melis@juggle 20:53:~/c/numpy-1.17.5/build/lib.linux-x86_64-3.7$ ~/software/python-3.7.7-static/bin/python3 -m pip install -U julia
melis@juggle 20:53:~/c/numpy-1.17.5/build/lib.linux-x86_64-3.7$ ls ~/software/python-3.7.7-static/lib/python3.7/site-packages/
julia  julia-0.5.4.dist-info  numpy  numpy-1.17.5.dist-info  README.txt

melis@juggle 20:54:~/concepts/blender-julia-test/test$ PYTHONPATH=~/c/numpy-1.17.5/build/lib.linux-x86_64-3.7 ~/software/python-3.7.7-static/bin/python3.7m subdiv.py 
/home/melis/c/numpy-1.17.5/build/lib.linux-x86_64-3.7/numpy/__init__.py
1.17.5
/home/melis/software/python-3.7.7-static/lib/python3.7/site-packages/julia/__init__.py
0.5.4
Initializing Julia (this might take a moment the first time)...
Done!
(Julia) Building half edges done in 1.513ms
(Julia) Input: 8 vertices, 6 polygons, 12 polygon edges
(Julia) Output: 26 vertices, 24 quads
(Julia) Subdivision done in 22.505ms
<class 'list'>
[-0.5555555820465088, -0.5555555820465088, -0.5555555820465088, -0.5061728358268738, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, 0.6049383282661438, -0.5061728358268738, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.6049383282661438, -0.5061728358268738, -0.5061728358268738, 0.5665295124053955, -0.44581618905067444, 0.5665295124053955, 0.5665295124053955, 0.5665295124053955, -0.44581618905067444, 0.5208048224449158, 0.520804762840271, 0.520804762840271, -1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 1.0, -0.75, -0.75, 0.0, -0.75, 0.0, 0.75, -0.75, 0.75, 0.0, -0.75, 0.0, -0.75, 0.0, 0.75, 0.75, 0.75, 0.75, 0.0, 0.0, 0.75, -0.75, 0.75, 0.0, 0.75, 0.75, -0.75, 0.0, 0.75, 0.0, -0.75, 0.0, -0.75, 0.75, 0.0, -0.75, -0.75]
<class 'list'>
[1, 5, 9, 13, 17, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 61, 65, 69, 73, 77, 81, 85, 89, 93]
<class 'list'>
[4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4]
<class 'list'>
[15, 2, 16, 9, 16, 4, 17, 9, 17, 3, 18, 9, 18, 1, 15, 9, 17, 4, 19, 10, 19, 8, 20, 10, 20, 7, 21, 10, 21, 3, 17, 10, 20, 8, 22, 11, 22, 6, 23, 11, 23, 5, 24, 11, 24, 7, 20, 11, 23, 6, 25, 12, 25, 2, 15, 12, 15, 1, 26, 12, 26, 5, 23, 12, 21, 7, 24, 13, 24, 5, 26, 13, 26, 1, 18, 13, 18, 3, 21, 13, 19, 4, 16, 14, 16, 2, 25, 14, 25, 6, 22, 14, 22, 8, 19, 14]