fix running in bash environment on windows

[lainme]

With this fix, aiida verdi commands can run on bash environment on windows, such as git bash

[unkcpz]

Thanks @lainme for the PR. I am not familiar with windows, @edan-bainglass could you give this a look. I guess the cmdline is not the only problem that we cannot support aiida natively on windows?

[lainme]

I have sucessfully run a simple example from aiida tutorial on windows. Here is what I do,

• Install git bash
• Install python3, patched aiida-core, graphviz, optionally install erlang and rabbitmq
• Run verdi presto under git bash
• Run python test.py as shown below
• Run verdi process list -a to track the status

rom aiida import load_profile, engine, orm, plugins
from aiida.manage.configuration import get_config
from aiida.storage.sqlite_temp import SqliteTempBackend

load_profile('presto')

node = orm.Int(2)
node.store()

@engine.calcfunction
def multiply(x, y):
    return x * y

x = orm.load_node(pk=1)
y = orm.Int(3)
z = multiply(x,y)

print(x)
print(y)
print(z)

from aiida.tools.visualization import Graph
graph = Graph()
calc_node = orm.load_node(3)
graph.add_incoming(calc_node, annotate_links="both")
graph.add_outgoing(calc_node, annotate_links="both")
graph.graphviz.render('test.gv')

Here is some screenshot

To run aiida in jupyter on windows platform

1. Install jupyter with pip install jupyter
2. In .ipynb file, add the following on top of any other code, which will fix the asyncio error

import nest_asyncio
nest_asyncio.apply()

Here is the screenshot of running in jupyter notebook

[codecov]

Codecov Report

Attention: Patch coverage is 80.30303% with 13 lines in your changes missing coverage. Please review.

Project coverage is 78.12%. Comparing base (d2fbf21) to head (ae25b13).

Files with missing lines	Patch %	Lines
src/aiida/transports/transport.py	20.00%	8 Missing ⚠️
src/aiida/manage/configuration/profile.py	0.00%	3 Missing ⚠️
src/aiida/engine/daemon/client.py	66.67%	1 Missing ⚠️
src/aiida/transports/plugins/local.py	75.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6715      +/-   ##
==========================================
- Coverage   78.13%   78.12%   -0.00%     
==========================================
  Files         564      564              
  Lines       42526    42549      +23     
==========================================
+ Hits        33222    33236      +14     
- Misses       9304     9313       +9     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[lainme]

The previous modification only fixes the running of verdi commands and internal python code. New commits fix the running of simple external code (such as aiida-diff) under windows, and more verdi commands can run under powershell as well.

To make external code works, one should also make some modification to git bash (or other bash system on windows):

• bash.exe should be added to PATH
• ps --version should show it belongs to procps-ng, not cygwin.

In git bash, to use ps from procps-ng:

• Download procps-ng package from https://repo.msys2.org/msys/x86_64/, unpackage the .zst format to .tar (you may use (https://www.tc4shell.com/en/7zip/modern7z/)
• Open git bash as administrator, and run tar -xvf XXX.pkg.tar -C / usr
• Make symlink ln -sf /usr/bin/procps /usr/bin/ps

[agoscinski]

Could you embed the suggested change in the pyproject.toml? Then we can run the CI

[agoscinski]

We just pinned most recent version

Suggested change

	'chardet'
	'chardet~=5.2.0 ; platform_system == "Windows"'

and then you need to run uv lock --upgrade-package chardet to put this into the uv.lock file.

[agoscinski]

When you do this we can run the CI

[agoscinski]

I the current state of this PR on the CI on windows and more than a third of the tests now pass, before this PR nothing would pass. See test suite run https://github.com/aiidateam/aiida-core/actions/runs/12950900268/job/36124705490?pr=6729 The test suite seems to crash (at least the errors are not properly reported after the run finishes), because of

Exception ignored in atexit callback: <bound method InteractiveShell.atexit_operations of <IPython.terminal.interactiveshell.TerminalInteractiveShell object at 0x0000019912B26CD0>>
Traceback (most recent call last):
  File "D:\a\aiida-core\aiida-core\.venv\Lib\site-packages\IPython\core\interactiveshell.py", line 3957, in atexit_operations
    tfile.unlink()
  File "C:\hostedtoolcache\windows\Python\3.11.9\x64\Lib\pathlib.py", line 1147, in unlink
    os.unlink(self)
PermissionError: [WinError 32] The process cannot access the file because it is being used by another process: 'C:\\Users\\RUNNER~1\\AppData\\Local\\Temp\\tmpasaqzo87test_hist.sqlite'

Will need more time to actually mark the tests that are passing so we only run these ones in the CI. In any case I would not block this PR till we have something in the CI running windows.

[lainme]

I also makes the daemonization in background works with a patched version of circus-0.18.0, but I don't have time to make the patch work with the master branch of circus.

The basic idea is parent->call vbs->call bat->call child, then the child can run in background continuously.

It's possible to implement this in aiida instead of circus, but the circus patch also fixed other stuff such as signal handling, etc.

Here is the patched circus-0.18.0: https://github.com/lainme/circus/tree/0.18.0-fix-windows

[agoscinski]

Thanks, I will have a look at your fork and see if we can monkey patch it. We also contribute to circus, so I can have a look to merge it there.

Regarding this PR I think we just need to add chardet to the conda environment.yml. I'll try to push something

[agoscinski]

Conda does not seem to have for the moment a way to make dependencies OS specific. There is the selector mechanic which some tools related to conda support like conda-lock which uses the same syntax conda uses to define metapackages. So I will just go with this to mark the package so one can in principle extract it out in the future.

For reference, in conda there is an ongoing discussion about supporting selectors suhttps://github.com/conda/conda/issues/8089

[agoscinski]

Will just have final look on Monday and then merge