Molten CPU usage when doing nothing

[medwatt]

make sure to do the following:
read the README
check existing issues (the config problem tag is helpful)
try with the latest version of molten and image.nvim, latest releases and then main/master branches
run :UpdateRemotePlugins

• OS: EndevourOS (Linux)
• NeoVim Version: 0.10.2
• Python Version: 3.12
• Python is installed with: pacman

`:checkhealth molten`

molten-nvim ~
- OK NeoVim >=0.9
- OK Python >=3.10
- OK Python module pynvim found
- OK Python module jupyter-client found
- WARNING Optional python module cairosvg not found
  - ADVICE:
    - pip install cairosvg
- WARNING Optional python module pnglatex not found
  - ADVICE:
    - pip install pnglatex
- WARNING Optional python module plotly not found
  - ADVICE:
    - pip install plotly
- WARNING Optional python module kaleido not found
  - ADVICE:
    - pip install kaleido
- OK Python module pyperclip found
- OK Python module nbformat found
- OK Python module pillow found

`:checkhealth provider` (the python parts)

provider.python: require("provider.python.health").check()

Python 3 provider (optional) ~
- `g:python3_host_prog` is not set. Searching for python3 in the environment.
- Executable: /usr/bin/python3
- Python version: 3.12.6
- pynvim version: 0.5.0
- OK Latest pynvim is installed.

Python virtualenv ~
- OK no $VIRTUAL_ENV

Description

There's a non-negligible CPU usage when using molten even when doing nothing.
The CPU usage drops to zero when the focus shifts to something else.
See video below.

molten_cpu_usage.mp4

Reproduction Steps

Steps to reproduce the behavior.
ie. open this file, run this code and wait for the output window to open, then do x

Optionally you can include a minimal config to reproduce the issue. This will help me figure things out much more quickly. You can find a sample minimal config here. If you include one, please also include the output of pip freeze from the python3 host program that you specify in the config.

Expected Behavior

A clear and concise description of what you expected to happen.

[benlubas]

I see similar spikes when in a buffer that molten is attached to. I suspect that this is b/c we're constantly polling the kernel in the background.

Fixing requires a large rewrite. Something I just haven't really had time to do. In an ideal world, this plugin would be written in rust, and I could ship it with luarocks.

I might never seriously start to work on that though, I don't have much motivation to do it.

[medwatt]

@benlubas, thanks for letting me know. The thing is, most of the time I just have a jupyter notebook opened in the background. I was thinking of switching my notebook activities to neovim (for obvious reasons). However, consuming 4-5 % of CPU usage while doing nothing is too much. I guess I'll stick to jupyter notebooks for now, until something more optimized comes along.

[medwatt]

I believe this is the function (in runtime.py) where the constant polling is done. Is there any reason why this code wasn't written with asyncio ?

def tick(self, output: Optional[Output]) -> bool:
    did_stuff = False

    assert isinstance(
        self.kernel_client,
        jupyter_client.blocking.client.BlockingKernelClient,
    )

    if not self.is_ready():
        try:
            self.kernel_client.wait_for_ready(timeout=0)
            self.state = RuntimeState.IDLE
            did_stuff = True
        except RuntimeError:
            return False

    if output is None:
        return did_stuff

    while True:
        try:
            message = self.kernel_client.get_iopub_msg(timeout=0)

            if "content" not in message or "msg_type" not in message:
                continue

            did_stuff_now = self._tick_one(output, message["msg_type"], message["content"])
            did_stuff = did_stuff or did_stuff_now

            if output.status == OutputStatus.DONE:
                break
        except EmptyQueueException:
            break

    return did_stuff

[benlubas]

The reason is I didn't write it. It came from magma.

The two things you could look into if you're willing to contribute are switching from the blocking kernel client to a non blocking one and switching from polling to not polling.

Potentially the issue with async is communicating back to the running nvim process, might have/want to totally change the way that's done. But it's definitely gotta be doable.

[benlubas]

Btw there's more than just that function iirc

[benlubas]

Also also. You might bc interested in neopyter. Pretty cool plugin too