fix(data): skip chart for nested/unknown column types

[kirangadhave]

Problem

Hovering a Polars Struct or List column in the datasets panel surfaced a confusing error and logged a spurious warning. These dtypes resolve to a FieldType of "unknown", but chart generation still ran the full pipeline and failed deep in Altair's dataframe-interchange serializer with ValueError: Unexpected DtypeKind: Struct(...). The caller caught the exception, logged a WARNING (which reads like a real failure), and shipped the raw serializer error to the frontend as ColumnPreview.error, as if charting had genuinely broken.

Fix

Short-circuit _get_altair_chart for "unknown" field types, returning no chart and no error. The dtype is known to be unchartable up front, so there is nothing to attempt and nothing to report. The check lives right after the field-type lookup that the function already performs, so it adds no extra work. The stats path is on a separate branch and is unaffected, so aggregation stats still compute and display for these columns.

Before:

After:

Closes MO-6157

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
marimo-docs	Ready	Preview, Comment	Jun 13, 2026 1:55am

[cubic-dev-ai]

No issues found across 2 files

Architecture diagram

sequenceDiagram
    participant UI as Frontend UI
    participant Preview as get_column_preview_dataset()
    participant Chart as _get_altair_chart()
    participant Table as Table Manager
    participant Stats as Statistics Engine
    participant Export as Altair Export

    Note over UI,Export: Column Preview Flow

    UI->>Preview: Request column preview
    Preview->>Table: get_field_type(column_name)
    Table-->>Preview: column_type, external_type
    
    alt column_type == "unknown" (e.g., Polars Struct/List)
        Preview->>Chart: _get_altair_chart(column_name)
        Chart->>Table: get_field_type(column_name)
        Table-->>Chart: column_type = "unknown"
        Chart-->>Preview: Return (None, None, None, None)
        Preview-->>UI: result(chart_spec=None, chart_code=None, error=None)
    else column_type != "unknown"
        Preview->>Chart: _get_altair_chart(column_name)
        Chart->>Table: get_field_type(column_name)
        Table-->>Chart: column_type (known type)
        
        alt stats.total == 0
            Chart-->>Preview: Return (None, None, "Table is empty", None)
        else stats.total > 0
            Chart->>Export: Serialize to Altair chart
            alt Serialization succeeds
                Export-->>Chart: Chart spec/code
                Chart-->>Preview: Return chart_spec, chart_code
            else Serialization fails
                Export-->>Chart: ValueError
                Chart-->>Preview: Return error
            end
        end
        
        Preview->>Stats: Compute aggregation stats
        Stats-->>Preview: Statistics data
        
        Preview-->>UI: result(chart_spec, chart_code, error, stats)
    end

    Preview->>Stats: Compute aggregation stats (always)
    Stats-->>Preview: Statistics data
    Preview-->>UI: result(stats=statistics)

Loading

_{Re-trigger cubic}

[Copilot]

Pull request overview

This PR improves dataset column previews by avoiding Altair chart generation for nested/unchartable column dtypes that currently map to FieldType == "unknown" (e.g. Polars Struct/List), preventing confusing backend warnings and serializer errors from being surfaced to the frontend.

Changes:

• Short-circuit _get_altair_chart when the resolved FieldType is "unknown", returning no chart and no error.
• Add a regression test ensuring Polars Struct columns produce stats but no chart and do not log warnings.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
marimo/_data/preview_column.py	Skips Altair chart generation entirely for "unknown" field types to avoid serializer failures.
tests/_data/test_preview_column.py	Adds coverage to ensure struct/nested columns return no chart/error and don’t emit warnings.

[Light2Dark]

Should we return a "No chart available" message. There may be users confused why there isn't a chart here.