Missing PyTorch dependencies in `pyproject.toml`

[Lex-DRL]

While debugging #46 , I discovered that on attempt to run existing tests in main, I get:

ImportError while loading conftest 'E:\GIT\ComfyUI-dev\custom_nodes\basic_data_handling\tests\conftest.py'.
..\__init__.py:15: in <module>
    from src.basic_data_handling import NODE_CLASS_MAPPINGS
..\src\basic_data_handling\__init__.py:1: in <module>
    from . import (boolean_nodes, casting_nodes, comparison_nodes, control_flow_nodes,
..\src\basic_data_handling\tensor_nodes.py:1: in <module>
    import torch
E   ModuleNotFoundError: No module named 'torch'

Process finished with exit code 4

Empty suite

[set-soft]

Hmmm ... I don't think listing it is a good idea.

1. This is a ComfyUI plugin, so is safe to assume ComfyUI is there. As ComfyUI depends on PyTorch there is no need to list it.
2. Even when ComfyUI filters PyTorch as dependency in plugins, is quite dangerous to list it, I saw pip pulling extra copies of PyTorch (various GB)
3. Is always advised to install PyTorch manually, this is because the package you need to install depends on the hardware

For testing purposes I suggest to also install it manually, most probably the CPU version for general CI/CD where no GPU is available.

[Lex-DRL]

To the best of my knowledge, ComfyUI manager (both the old standalone "node pack" and the new built-in extension manager) uses requirements.txt for the actual dependencies installation. Your repo lists the dependencies required for testing as optional dev set of dependencies. So the proposed change wouldn't affect regular users.
Also, the current tests import the actual code by absolute paths - i.e., in such a way that this repo is supposed to be in PATH, so by default it's assumed to be the root project folder for IDE, NOT a subfolder under ComfyUI/custom_nodes. So, to run tests locally without any manual setup, you DO need this node pack to be opened in IDE as its own project and you DO need it to have its own venv.
Yes, a contributor can dig into your github workflow to see what the dev environment needs. But those are extra steps. If it was right there in pyproject.toml, they'll immediately know how to setup the dev venv. If torch is already installed, pip would just skip it.

For testing purposes I suggest to also install it manually, most probably the CPU version for general CI/CD where no GPU is available.

If you stick to this decision, it would be nice to at least have a CONTRIBUTING.md and mention it there.
But I still believe that updating pyproject.toml is actually a better way.

[Lex-DRL]

On a complete side note, when should we expect the next release with #46 ?

[StableLlama]

I was quite busy and can now start to look into it.

And TBH, I don't know how #46 got pulled. All I did was to let the tests run.
Anyway, I'll review it and depending on that I'll keep it like it is or might revert it and ask for refinement.

Once that is done I'll trigger the next release.

[StableLlama]

Version 1.5.0 is released now.

The torch dependency is automatically fulfilled by Comfy and thus not included in the config. But to help out of tree testing (why would you want to do that, except when running test automation? For developing it is best to do it in the comfy directory tree as that allows you to live test in the target environment) I have added in #48 tests that fail with an error message hinting the user about how to fix it.

[Lex-DRL]

@StableLlama
Thanks for the improvement... but FYI: people use IDEs other than VSCode, too. (e.g., I use PyCharm) To make pytest work there, you have two options:

• opening the repo dir as its own project (probably, with its own venv)
• manually configure the global ComfyUI project you have - to include this repo's dir into PATH. It's done somewhat easily... as long as you know where to configure it. And it's buried VERY deeply inside project settings.

That was the whole reason why I raised the issue.

[StableLlama]

These nodes were initially created by me using IntelliJ and I only recently switched to VSCode.
So I'm pretty sure you can run the tests in PyCharm as well, as I could at that time. :)