Editing a cell does not mark its dependents as stale

[zmbc]

Describe the bug

I may be misinterpreting how this is supposed to work, but it is quite surprising to me that when you edit the contents of a Python cell, and even when you then de-focus that cell, the only cell that is marked stale is the one you edited. You then have to press "run all stale cells" twice to get everything to update.

Will you submit a PR?

• Yes

Environment

Details

{
  "marimo": "0.19.7",
  "editable": false,
  "location": "/mnt/share/homes/zmbc/mambaforge/envs/vivarium_gates_mncnh_artifact/lib/python3.11/site-packages/marimo",
  "OS": "Linux",
  "OS Version": "5.4.0-216-generic",
  "Processor": "x86_64",
  "Python Version": "3.11.14",
  "Locale": "en_US",
  "Binaries": {
    "Browser": "--",
    "Node": "--"
  },
  "Dependencies": {
    "click": "8.3.1",
    "docutils": "0.22.4",
    "itsdangerous": "2.2.0",
    "jedi": "0.19.2",
    "markdown": "3.10.1",
    "narwhals": "2.15.0",
    "packaging": "25.0",
    "psutil": "7.2.1",
    "pygments": "2.19.2",
    "pymdown-extensions": "10.20.1",
    "pyyaml": "6.0.3",
    "starlette": "0.52.1",
    "tomlkit": "0.14.0",
    "typing-extensions": "4.15.0",
    "uvicorn": "0.40.0",
    "websockets": "16.0"
  },
  "Optional Dependencies": {
    "altair": "6.0.0",
    "duckdb": "1.4.3",
    "ibis-framework": "11.0.0",
    "loro": "1.10.3",
    "nbformat": "5.10.4",
    "openai": "2.16.0",
    "pandas": "1.5.3",
    "polars": "1.37.1",
    "pyarrow": "23.0.0",
    "pytest": "9.0.2",
    "ruff": "0.14.14",
    "sqlglot": "27.29.0"
  },
  "Experimental Flags": {}
}

Code to reproduce

No response

[mscolnick]

@zmbc i assume you have on-cell-change off like below?

I can see your argument for sure. I'm not sure what feels more correct.

[zmbc]

Yes -- this is with everything set to "lazy." I tried to add a video but was having some trouble recording my screen.

My thinking is that the "run all stale cells" action should always result in there being no remaining stale cells.

[mscolnick]

We likely won't color yellow the cells downstream of an edited cell, but I think you are right that you should not need to click twice to run all stale cells.

Chatted with the team. We are planning to stick with the current way. In lazy mode, we want to be careful re-running un-intended cells. And this change would catch a wider net and can be more distruptive.

[zmbc]

I don't understand what you mean by this. I wasn't asking for any additional cells to be run, only marked as stale. And they will be, anyway, as soon as you run the edited cell.

What is a use case in which this change would cause too many cells to be run?

[akshayka]

I think many of our users habitually use Ctrl-Shift-R to run cells (I do). When switching to lazy mode, I'd probably hit that reflexively, without intending to run all affected cells. Just one thought. I also think highlighting all descendants could add visual noise.

Maybe your suggestion makes sense, I am not entirely sure. But it would be a big change for our many users, who are accustomed to the current behavior. So I would want high conviction and feedback from others before making such a change.

[zmbc]

When switching to lazy mode, I'd probably hit that reflexively, without intending to run all affected cells.

The part I don't understand is that, if you use ctrl-shift-R to run cells, then the next time you use it (to run whatever else you change next), it will run these downstream cells regardless. Or is there some way to manually mark cells as "non-stale"?

[akshayka]

No, you are right, there is no way to mark cells as non-stale. Might it feel noisy to have edits mark so many cells as stale, and undos possibly mark many cells as not stale? I don't dispute your logic, I suppose I just don't know how it will feel and how our community members will react.

[zmbc]

Interesting question. I don't know. In general I think doing something in-between, like marking the dependents as stale on the edited cell losing focus, could be reasonable. It just feels very strange to tie it to the run action.

[akshayka]

I'm inclined to keep this the way that it is, since the point of the lazy executor is to be extra careful not to run expensive cells

[mscolnick]

(moving to a discussion)