Is it safe to update numpy or install matplotlib?

lIlIl

I have two questions:

Is it safe to install packages in the Python installation of Code_Aster?
Installing matplotlib may trigger an update of NumPy: would Code_Aster still work with an updated version of NumPy (since most of its code is in Fortran)? I could install a downgraded version of matplotlib==3.4 that depends on NumPy 1.16.

Here are some things I've tried so far:

I ran this .comm file:

import os
os.system(r'where python > "C:\Users\<username>\Downloads\where_python.txt"')

The content of where_python.txt is (I installed the Windows version from code-aster-windows.com):

C:\Users\AppData\Local\code_aster\v2021\Python37\python.exe

then I ran

import subprocess
import sys
subprocess.check_call([sys.executable, "-m", "pip", "list"])

which printed

Package    Version
---------- -------
astk       2021.0 
numpy      1.16.1 
pip        18.1   
psutil     5.5.0  
scipy      1.2.0  
setuptools 40.6.2 
sip        4.19.8

So my installation of Code_Aster uses numpy 1.16.1. In a separate virtual environment I installed matplotlib to reveal its dependencies (as of February 2024):

Collecting numpy<2,>=1.21 (from matplotlib)
  Downloading numpy-1.26.3-cp311-cp311-win_amd64.whl.metadata (61 kB)

I don't know if there have been breaking changes between NumPy 1.16 and 1.21.

mf

Hello,

there were a lot of releases of Numpy between 1.16 and 1.21. You can look up any changes in Numpy in the release notes

numpy.org/devdocs/release/1.21.0-notes.html

From the little Python I do, I have the feeling that matplotlib is always behind Numpy (it depends on an older version).

However, I am using a quite new version of CA (16.4) in Linux and it uses Numpy 1.15.1 (this is in the container). You didn't mention your CA version, but I know the Windows version is always behind. So, my assumption is, you are already using a more modern version of Numpy than your CA depends on!

Also, it comes down to what is 'safe'? You can always go back to the old version in case you encounter trouble. Just try, also Windows is not dependent on a Python installation, meaning it's not possible to break the OS with a faulty Python installation. This is not true in Debian and Ubuntu, messing up the system installation of Python can break the OS. My guess is, that is also a good reason to put SM and CA in a container, right? :-)

Mario.

lIlIl

Thank you for the link. It would take a lot of time to skim through all the changes that happened in NumPy between the two versions.

I was afraid to try, but here we go:

# in the .comm file
subprocess.check_call([sys.executable, "-m", "pip", "install", "--upgrade", "pip"])

This worked perfectly. Then, I don't want to take too many risks so I'll install the older version of matplotlib:

>>> subprocess.check_call([sys.executable, "-m", "pip", "install", "matplotlib==3.4"])
...
Requirement already satisfied: numpy>=1.16 in c:\users\<username>\appdata\local\code_aster\v2021\python37\lib\site-packages (from matplotlib==3.4) (1.16.1)
...
Installing collected packages: typing-extensions, six, pyparsing, pillow, cycler, python-dateutil, kiwisolver, matplotlib
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
anyio 3.1.0 requires idna>=2.8, which is not installed.
jupyter-server 1.8.0 requires ipython-genutils, which is not installed.
...
Successfully installed cycler-0.11.0 kiwisolver-1.4.5 matplotlib-3.4.0 pillow-9.5.0 pyparsing-3.1.1 python-dateutil-2.8.2 six-1.16.0 typing-extensions-4.7.1

Despite the ERROR which only seems to be a warning, it worked. I found online that it is advisable to use --use-deprecated=legacy-resolver or --use-feature=2020-resolver. After trying to call matplotlib, this appears in the .mess file:

fort.1: UserWarning: Matplotlib is currently using agg, which is a non-GUI backend, so cannot show the figure.

Which has an easy fix since Tkinter is included by default. This works:

import matplotlib
matplotlib.use('TkAgg') #Agg, Qt5Agg and WXAgg didn't work
import matplotlib.pyplot as plt
plt.plot([0,1],[0,1],'.-',color='orange')
plt.show()

I can now display results directly from Code_Aster, which is great! Let's hope I don't run into problems because of this. I'll keep you updated.

EDIT: when plotting scatter() in 3D, the colors of the markers are incoherent (they are randomized each time you rotate the view). It works in 2D. This is probably due to the TkAgg backend.

lIlIl

After installing medcoupling 9.11.0 today, which may have affected the numpy install, this appeared in my message file:

RuntimeError: module compiled against API version 0xe but this version of numpy is 0xd
EXECUTION_CODE_ASTER_EXIT_1250=0

But everything worked and nothing seemed to be broken.

faelannm

Hi,

It’s generally not recommended to update or install packages directly into the Python environment used by Code_Aster, as updates can introduce incompatibilities that might break functionality, especially with dependencies like NumPy, which is heavily used in Code_Aster.

lIlIl

Here are all the packages I have installed in versions 15 and 16 and everything still works:

Code_Aster Windows v2021 (15.4.0)

Package           Version
----------------- -----------
astk              2021.0
cycler            0.11.0
freetype-py       2.5.1
hsluv             5.0.4
kiwisolver        1.4.5
matplotlib        3.4.0
medcoupling       9.7.0.post2
numpy             1.21.6
packaging         24.0
Pillow            9.5.0
pip               24.0
psutil            5.5.0
PyOpenGL          3.1.7
pyopengltk        0.0.4
pyparsing         3.1.1
python-dateutil   2.8.2
scipy             1.2.0
setuptools        40.6.2
sip               4.19.8
six               1.16.0
typing_extensions 4.7.1
vispy             0.13.0
windows-toasts    0.3.3
winsdk            1.0.0b9

Code_Aster Windows v2023 (16.5.0)

Package           Version
----------------- -----------
astk              2021.0
cycler            0.11.0
fonttools         4.38.0
kiwisolver        1.4.5
matplotlib        3.5.3
medcoupling       9.7.0.post2
mpi4py            3.1.3
numpy             1.21.6
packaging         24.0
Pillow            9.5.0
pip               24.0
psutil            5.5.0
pyparsing         3.1.2
python-dateutil   2.9.0.post0
scipy             1.2.0
setuptools        40.6.2
sip               4.19.8
six               1.16.0
typing_extensions 4.7.1
windows-toasts    0.3.3
winsdk            1.0.0b9