chart: scalar per-point mapping + vline/hline ctors; de-flake [62] capture (#828, #876)

[InauguralPhysicist]

Closes #828. Closes #876.

#828.1 — per-point allocation in the chart render loop

Reproduced the consumer's measurement locally (4,000-point points-style series, four vline markers, hover off, SDL_VIDEODRIVER=dummy, n=5 medians): 62.7 ms/frame baseline. A semantics-breaking ceiling probe (mapping replaced by constants) measured 39.6 ms/frame — the _chart_map list-per-point was ~37% of the frame, confirming the report before building the fix.

The shipped fix inlines the mapping as scalar arithmetic in the hot loop with the view factors hoisted (_chart_map stays for the cold sites: ticks, markers, hover, chart_to_pixel). Shipped: 42.6 ms/frame (n=5 median) — 87% of the probed ceiling, 1.47×. Also removes the #827 amplifier (per-point lists pinned under armed history: +3.9 MB/frame in the dynamics consumer).

Correctness pin: a new [63] check renders a 2-point series and asserts the recorded dot rects sit at exactly chart_to_pixel's floored pixels — the inline and the readable form cannot drift apart silently.

#828.2 — dead marker coordinate

chart_vline(x, label, color) / chart_hline(y, label, color) — the same marker dict without the coordinate the kind ignores. Docs row + constructor checks.

#876 — [62] audio capture flake (folded, found while landing this)

Two full-suite runs today reported 38 failures with no code cause: a real capture device that OPENS but delivers no samples in the 2s bounded poll fails two delivery checks, and [62] lump-counts all 38. Reproduced 2-in-3 against the real driver; 3/3 green under SDL_AUDIODRIVER=dummy. The delivery-dependent checks now SKIP with the count held constant (the file's existing no-device convention), keeping full real-driver coverage on boxes where capture delivers. [62] verified 53/53 in all three modes.

Validation

• Full suite vs gfx build: 3815/3815.
• Bench n=5 before/after/ceiling as above; bench driver in the PR discussion if needed.
• --lint on touched files: no new warnings.

Closes #828. Closes #876.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR addresses two user-facing issues in the UI chart and test suite: it removes per-point list allocation from the chart render hot loop (improving high-point-count throughput) and deflakes the audio capture test section by treating “opened but silent” capture devices as an environment skip rather than a runtime failure.

Changes:

• Inlined chart data→pixel mapping as scalar arithmetic in _render_chart’s per-point loop, hoisting view factors out of the inner loop to avoid per-point allocations.
• Added chart_vline / chart_hline constructors to avoid requiring callers to pass an ignored coordinate for vline/hline markers.
• Updated tests/docs/changelog: pinned mapping correctness against chart_to_pixel, and adjusted audio capture tests to skip delivery-dependent checks when a device opens but delivers no samples.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/test_ui.eigs	Adds a regression check that recorded point-dot rects match chart_to_pixel’s mapped pixels; covers new marker constructors.
tests/test_audio.eigs	Deflakes capture tests by skipping delivery-dependent assertions when capture opens but yields no samples (environment condition).
lib/ui_w_viz.eigs	Implements scalar per-point mapping in the chart render loop and adds chart_vline/chart_hline.
docs/STDLIB.md	Documents the new marker constructors.
CHANGELOG.md	Records the chart performance improvement and the audio capture flake fix.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.