nordisk/nornir

🧪 The shining Test pane — per-repo × per-aspect health matrix (EPIC L5)

Design + guide for the flagship nornir viz Test tab (src/viz/test_tab.rs, Tab::Test). EPIC L5 turned the old flat run/case list into a gorgeous repo × aspect matrix painted on the facett palette, with metric badges, per-aspect trend sparklines, per-repo health scores, a summary strip, and click-to-drill-down. THE_BIG_PLAN LAWs apply: everything rendered is exposed as data (state_json()["test"]["grid"] + $NORNIR_VIZ_TRACE), the test inject-and-asserts the real grid, and a wgpu PNG snapshot is the acceptance artifact.

What it shows

A grid over the warehouse test_results table (which since EPIC L carries aspect: String + metric: f64 per row):

Health score

health_score(aspects, cells) (in src/viz/test_matrix_grid.rs) blends, over the aspects that actually ran in the latest run:

• green ratio (green / (green+red+skip)) — the dominant term,
• coverage % when measured (+25 pts weight; redistributed to green-ratio when there is no coverage signal),
• a penalty for clippy warnings (−4 each) and audit advisories (−6 each), each capped, so one noisy aspect can't drive the score negative.

Result is clamped to 0–100. All-green + high coverage + no warnings → ~95–100; a build fail + low coverage + several warnings → well under 50.

Architecture — pure model + thin renderer

The LAW "components expose readable data" is met by splitting the pane:

• src/viz/test_matrix_grid.rs — pure, no egui. Matrix::build(&[TestResultRow]) groups rows by repo, picks each repo's latest run, builds one Cell per aspect column (rolling the unit aspect's per-test rows into pass/fail counts + drill-down RedCases), computes the health score, and the per-aspect sparkline series. Unit-tested in isolation; folded verbatim into state_json.
• src/viz/facett_theme.rs — the facett palette family, vendored. nornir's viz is on egui 0.33 while upstream facett-core is on egui 0.34; a direct dependency would link two incompatible egui Color32/Ui types into one binary. So the palette (names, colours, ALL order) is a verbatim copy of facett_core::theme::Theme, plus the semantic status ramp (status_fill) and health ramp (health_color). When facett ships on egui 0.33 this collapses to a re-export. (Blocker recorded: the egui-version skew is why we vendor rather than depend.)
• src/viz/test_tab.rs — the egui renderer. Paints the grid on a canvas (painter) honouring the active Theme: zebra rows, hover highlight, selection accent, rounded status tiles, health pills, sparklines. Emits the grid as state_json()["test"]["grid"] and traces test.matrix.{load,build,end} + test.matrix.{palette,drilldown} events.

state_json shape (["test"]["grid"])

{
  "aspects": ["build","unit","doctest","clippy","fmt","audit","coverage"],
  "repos":   ["nornir","skade","znippy"],
  "rows": [
    {
      "repo": "znippy",
      "run_id": "znippy-new",
      "health": 49.0,
      "cells": [
        {
          "aspect": "clippy", "status": "fail",
          "ran": true, "red": true, "green": false,
          "metric": 3.0, "badge": "⚠3",
          "count": 1, "passed": 0, "failed": 1,
          "red_cases": [{ "test_name": "clippy", "suite": "znippy",
                          "status": "fail", "message": "3 clippy warning(s)" }],
          "spark": [0.0, 40.0]           // inverted: 5→3 warnings reads as rising
        }
        // … one cell per aspect column (blank cells: ran=false, badge="")
      ]
    }
  ],
  "totals": { "green": 15, "red": 2, "skip": 1, "blank": 3 }
}

The legacy keys (runs, cases, selected_run, selected_repo, selected_green, plus the new palette + selected_cell) remain on ["test"] for back-compat with the run-summary consumers.

Tests (LAW #1 + LAW #4)

• cargo test --lib viz::test_matrix_grid — 9 pure unit tests: rows-are-repos / cols-are-aspects, metric badges, unit roll-up, latest-run-wins, not-run blanks, health high/low, coverage sparkline rising, warning sparkline inversion.
• cargo test --lib viz::facett_theme — palette uniqueness/lookup, status ramp, health ramp.

• tests/viz_test_pane.rs:

  • test_pane_exposes_run_summaries_and_cases — legacy run shape still correct.
  • test_pane_exposes_repo_x_aspect_grid — injects a 3-repo × 7-aspect × multi-run matrix and asserts the grid: repo/aspect ordering, per-cell status + metric + badge (88%, ⚠3, 🛡0, 1p 1f), blank not-run cells, health ordering (skade ≥ nornir > znippy), sparkline series (coverage 71→88, clippy warnings inverted), and summary totals (2 red, 1 skip).
  • viz_snapshot_test_pane — wgpu render of the populated, colourful matrix to tests/snapshots/viz_test_pane.png (the acceptance artifact), plus a functional assert that 3 repos × 7 aspects with ≥1 red cell are on screen.

How to drive it live

nornir test all                 # populate test_results across repos
nornir viz                      # open nornir-ui → 🧪 Test tab
# pick a palette from the top strip; click any cell to drill into its failures.
NORNIR_VIZ_TRACE=/tmp/t.jsonl nornir viz   # machine-readable IN/OUT/END twin