refactor: consolidate Runner and app ScriptRunner

[dmadisetti]

📝 Summary

This PR cleans up dataflow::Runner into functions that directly utilize the evaluator pipeline introduced in #9616.
Additionally, it consolidates AppScriptRunner to use this logic as well.

A note, this refactor prevents AppScriptRunner from classifying mo.stop as an exception. The previous behavior would throw and cause a total script stop- the revised behavior (and desired), is that the dependency tree is cancelled while the rest of the notebook can compute.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
marimo-docs	Ready	Preview, Comment	May 21, 2026 1:46am

[cubic-dev-ai]

3 issues found across 10 files

Architecture diagram

sequenceDiagram
    participant Client as App/Cell API
    participant by_kwargs as by_kwargs module
    participant Evaluator as Evaluator
    participant Executor as DefaultExecutor
    participant Scheduler as SequentialScheduler
    participant Graph as DirectedGraph

    Note over Client,Graph: NEW: Cell.run(**kwargs) public API path

    Client->>by_kwargs: run_cell_sync(graph, cellId, kwargs)
    by_kwargs->>Graph: cells[cellId].is_coroutine()
    alt Cell is coroutine
        by_kwargs->>by_kwargs: Raise RuntimeError
    else Sync cell
        by_kwargs->>by_kwargs: _validate_kwargs(cell, kwargs)
        by_kwargs->>Graph: Get ancestor closure (pruned by substitutions)
        by_kwargs->>by_kwargs: topological_sort(graph, ancestors)
        by_kwargs->>Evaluator: evaluate_sync(ancestorCell, glbls)
        Evaluator->>Executor: execute_cell(ancestorCell, glbls)
        loop For each ancestor
            Executor-->>Evaluator: glbls updated
        end
        by_kwargs->>by_kwargs: _substitute_refs(cell, glbls, kwargs)
        by_kwargs->>Evaluator: evaluate_sync(targetCell, glbls)
        Evaluator->>Executor: execute_cell(targetCell, glbls)
        Executor-->>Evaluator: output
        by_kwargs-->>Client: (output, definitions)
    end

    Note over Client,Scheduler: NEW: ScriptRunner consolidated flow

    Client->>Scheduler: __init__(cells_to_run, graph)
    Scheduler->>Scheduler: Pop next cell in order
    loop For each cell
        Scheduler-->>Client: pop_cell()
        alt Cell is cancelled
            Client->>Scheduler: cancelled(cid) == True
            Client->>Client: Skip execution
        else Cell not cancelled
            Client->>Evaluator: evaluate_sync(cell, glbls)
            Evaluator->>Executor: execute_cell(cell, glbls)
            alt Execution succeeds
                Executor-->>Evaluator: output
                Evaluator-->>Client: RunResult(output=..., exception=None)
                Client->>Client: _handle_run_result: record output
            else Execution raises MarimoRuntimeException
                Evaluator-->>Client: RunResult(exception=MarimoRuntimeException)
                Client->>Client: unwrap_user_exception()
                alt MarimoStopError
                    Client->>Client: Record stop output
                    Client->>Scheduler: cancel(cid)
                    Scheduler->>Scheduler: Mark descendants cancelled
                else Other exception
                    Client->>Client: Re-raise exception
                end
            end
        end
    end
    Client-->>Client: Return (outputs, glbls)

    Note over Evaluator,Executor: NEW: evaluate_sync & evaluate_interruptible added

    Client->>Evaluator: evaluate_interruptible(cell, glbls)
    alt Cell is coroutine & main thread
        Evaluator->>Evaluator: Create future, wrap with SIGINT handler
        Evaluator->>Evaluator: asyncio.ensure_future(evaluate(...))
        alt SIGINT received
            Evaluator->>Evaluator: Cancel future → CancelledError
        else Normal completion
            Evaluator-->>Client: RunResult
        end
    else Sync cell or not main thread
        Evaluator->>Evaluator: evaluate(cell, glbls) directly
        Evaluator-->>Client: RunResult
    end

    Note over Evaluator,Executor: CHANGED: lifecycle chain extracted to helper methods

    Client->>Evaluator: evaluate/evaluate_sync(cell, glbls)
    Evaluator->>Evaluator: _setup_chain(cell, glbls)
    loop For each lifecycle
        alt Lifecycle returns Skip
            Evaluator->>Evaluator: Record skip decision
        else Lifecycle completes
            Evaluator->>Evaluator: Append to completed list
        end
    end
    alt Skip was returned
        Evaluator->>Evaluator: value = skip.value
    else No skip
        Evaluator->>Executor: execute_cell / execute_cell_async(cell, glbls)
        Executor-->>Evaluator: output value
    end
    Evaluator->>Evaluator: _teardown_chain(cell, glbls, completed, value, exception)
    loop Reversed completed lifecycles
        Evaluator->>Evaluator: teardown(cell, glbls, result)
    end
    Evaluator-->>Client: RunResult

Loading

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[mscolnick]

is there ever a plain MarimoStopError that is not wrapped? maybe we can be defensive?

[mscolnick]

should we check if e.__traceback__.tb_next is not none either?
maybe we can pull out to a util up_one_traceback(e)

[mscolnick]

this should be tested

[mscolnick]

this is no longer cached. intentional?

[cubic-dev-ai]

1 issue found across 4 files (changes from recent commits).

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>