fix column summary chart warnings and edge cases for null cols

[Light2Dark]

Summary

Fixes #10303, the Vega console warnings and hover flicker in table column-summary charts reported in, and cleans up several related edge cases for null/NaN columns.

Root causes addressed:

• Hover flicker + Dropping ... aggregate max warning. The invisible tooltip/hover layers encoded y: { aggregate: "max" }. Without a field Vega drops the encoding (console warning); with a field the hit area shrinks to the (often tiny) bar height, causing flicker on short integer bars. Replaced with full-height pixel hit targets (y: { value: 0 }, y2: { value: <height> }) in both the current and legacy numeric/temporal specs.
• Infinite extent warnings on all-NaN / all-null columns. All-NaN bins produced empty quantitative histograms and fed NaN axis tick values into Vega. Bins are now filtered through a shared isValidBinValue guard (rejects null and NaN), all-null columns render a dedicated null bar only, and axis tick values are NaN-safe.
• State mutation bug. getVegaSpec mutated the stored bin values on every call by push-ing a synthetic null bin, so bins grew across renders. Chart data now derives from a fresh array and never mutates stored state.

Also refactors the duplicated null-bar + full-height-tooltip construction into a single buildNullBarSpec helper, and expands the null/NaN column-summary smoke tests (pandas_nan.py, column-header-chart.py).

Test plan

• pnpm test src/components/data-table/__tests__/chart-spec-model.test.ts (45 passing, incl. new all-null numeric/temporal, NaN-bin no-mutation, and empty-bins cases)
• make fe-check (typecheck + lint clean)
• make py-check (format + mypy clean)
• Manual: open marimo/_smoke_tests/tables/column-header-chart.py, confirm no Vega Infinite extent / Dropping ... aggregate max warnings in the browser console and no hover flicker on short bars

Made with Cursor

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
marimo-docs	Ready	Preview, Comment	Jul 24, 2026 5:40am

[cubic-dev-ai]

All reported issues were addressed across 8 files

Architecture diagram

sequenceDiagram
    participant UI as Table UI
    participant Model as ColumnChartSpecModel
    participant Bins as BinValues Store
    participant Utils as isValidBinValue / calculateBinStep
    participant SpecBuilder as buildNullBarSpec / FullHeightTooltip
    participant Vega as Vega Rendering Engine

    Note over UI,Vega: Column Summary Chart Rendering Flow

    UI->>Model: getHeaderSummary(column)
    
    Model->>Bins: get column bin values
    Bins-->>Model: binValues (array)
    
    Model->>Bins: get column stats (nulls count)
    Bins-->>Model: nullCount
    
    Model->>Utils: isValidBinValue(binValue)
    Utils-->>Model: true/false
    
    alt All bins invalid (all-NaN/all-null)
        Model->>Model: Filter valid bins → empty array
        
        alt nullCount > 0
            Model->>SpecBuilder: buildNullBarSpec(width=CONCAT_CHART_WIDTH)
            SpecBuilder-->>Model: null-only chart spec (vacant bar + invisible tooltip)
            Model->>Vega: Render null bar only
            
            Note over SpecBuilder: Uses y: { value: 0 }, y2: { value: 30 }
            Note over SpecBuilder: No aggregate: max encoding
        else No nulls
            Model-->>UI: null spec (no chart)
        end
        
    else Valid bins exist
        Model->>Model: Create fresh bins array (no mutation)
        
        opt nullCount > 0
            Model->>Model: Append synthetic null bin to fresh array
        end
        
        alt Numeric column
            Model->>SpecBuilder: buildNullBarSpec(width=CONCAT_NULL_BAR_WIDTH, filterNullOnly)
            SpecBuilder-->>Model: null bar spec
            
            Model->>Model: Build histogram with transform filter (exclude null bins)
            Model->>Model: Calculate axisTickValues (filter NaN)
            Note over Model: Uses FULL_HEIGHT_TOOLTIP_Y instead of aggregate: max
            
            Model->>Vega: Render hconcat(nullBar, histogram)
            
        else Temporal column
            Model->>Utils: getPartialTimeTooltip(validBins)
            Model->>Model: Build temporal chart spec
            
            alt Single non-null bin + no nulls
                Model->>Vega: Full-width bar (no hconcat)
            else Multiple bins
                Model->>Vega: Layered chart with null bar + temporal bars
            end
        end
    end
    
    Note over Model,Vega: Key Design Decisions
    Note over Model: Always creates fresh bins array - no stored state mutation
    Note over SpecBuilder: Tooltip layers use pixel y/y2 values
    Note over SpecBuilder: No aggregate: max encoding prevents hover flicker
    Note over Utils: isValidBinValue rejects null, NaN, and string/Date NaN

Loading

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

Re-trigger cubic</sub>

[Copilot]

Pull request overview

Fixes Vega console warnings and hover flicker in data-table column-summary charts for all-null / all-NaN columns by making hover hit targets full-height, filtering invalid bin values, and preventing bin-state mutation across renders.

Changes:

• Replace tooltip/hover layers that used aggregate: "max" with full-height pixel hit targets (y/y2 with value) in both current and legacy specs.
• Filter out invalid histogram bins (null/NaN) via shared isValidBinValue, add NaN-safe axis tick values, and avoid mutating stored bin arrays when injecting a synthetic null bin.
• Refactor null-bar construction into a helper and expand unit tests + smoke tests for null/NaN column scenarios.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
marimo/_smoke_tests/tables/column-header-chart.py	Expands visual smoke coverage for null/NaN numeric + temporal columns and threshold behavior.
marimo/_smoke_tests/pandas_smoke_tests/pandas_nan.py	Adds smoke scenarios for all-NaN and mixed NaN columns above/below chart threshold.
frontend/src/components/data-table/types.ts	Exports BinValue type so it can be referenced in shared bin-validation helpers.
frontend/src/components/data-table/column-summary/utils.ts	Adds isValidBinValue and uses it to avoid null/NaN-derived temporal tooltip/bin-step issues.
frontend/src/components/data-table/column-summary/legacy-chart-spec.ts	Updates legacy tooltip hit-target encodings to avoid Vega “Dropping aggregate max” warnings and hover flicker.
frontend/src/components/data-table/column-summary/chart-spec-model.tsx	Centralizes null-bar spec creation, filters invalid bins, avoids bin-array mutation, and makes tooltip hit areas full-height.
frontend/src/components/data-table/tests/chart-spec-model.test.ts	Adds/updates tests to cover all-null numeric/temporal, NaN-bin filtering, and non-mutation behavior.
frontend/src/components/data-table/tests/snapshots/chart-spec-model.test.ts.snap	Updates snapshots for new tooltip-layer encodings and new null-only chart specs.

Comments suppressed due to low confidence (1)

frontend/src/components/data-table/column-summary/legacy-chart-spec.ts:158

• y2: { value: 100 } is hard-coded to match the legacy chart height, but the actual height is controlled by base.height. Deriving y2 from base.height avoids the tooltip hit area drifting if the legacy chart height changes.

          },
          y: { value: 0 },
          y2: { value: 100 },
          tooltip: [

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (2)

frontend/src/components/data-table/column-summary/utils.ts:12

• The JSDoc says this guard rejects “nulls and NaN”, but the implementation also rejects ±Infinity via Number.isFinite. Update the comment so it matches the actual behavior (and keeps the “finite” wording consistent).

/**
 * True when a bin has finite, non-null start/end suitable for quantitative /
 * temporal encodings. Rejects nulls and NaN (all-NaN hist garbage).
 */

marimo/_smoke_tests/tables/column-header-chart.py:262

• requests.get(...) without a timeout can hang indefinitely during manual smoke testing, and failures currently surface later as a zip parsing error. Consider adding a timeout and raise_for_status() so failures are fast and explicit.

    response = requests.get(train_parquet_link)
    zip_data = io.BytesIO(response.content)