Routing export: extreme-aspect utility runs can overflow fixed-pad PDF viewport

[SomSamantray]

Problem

'routing' admits the longest/thinnest builders (pipe/duct/liquid-line/HVAC) whose bbox can be near-zero in one axis. The export viewport is a fixed 20%+1m pad computed once and fit preserving aspect; an extreme utility-run aspect yields a poor fit or geometry bleeding off-page when label collision shifts push toward the viewport edge. Reachable only because 'routing' exists; the old scopes are unaffected.

Suggested fix

Re-measure the viewport against the routed (post-filter) geometry, clamp the pad on the short axis, and/or re-fit after annotation collision resolution; add an end-to-end export test with a collinear pipe run.

Evidence

packages/editor/src/lib/floorplan/floorplan-export.tsx:521-524 -- const padding = Math.max(MIN_PLAN_PADDING_M, Math.max(modelBounds.width, modelBounds.height) * PLAN_PADDING_RATIO)

Source

Branch feat/floorplan-export-routing-scope (head f2d3051) — planned PR for #619.

Metadata

Severity: P3 | Confidence: 75 | Reviewer(s): adversarial | Finding ID: floorplan-export-tsx-521-extreme-aspect-viewport

[SomSamantray]

Carried by PR: #636

[Aymericr]

Closing — I checked the padding math and the mechanism is inverted. padding = max(1m, max(width, height) * 0.2) is derived from the larger dimension and applied symmetrically, so a near-zero-thickness pipe run gets proportionally more padding on its short axis, not less; clamping the short axis would make the fit worse. And fitPlanToBox preserves aspect while shrinking to fit, so it can't exceed the page box.

The concrete fix requested — "re-measure the viewport against the post-filter geometry" — is also already what happens: the viewport handed to the PDF renderer is mounted.viewport, measured from the mounted SVG after scope filtering, not from pre-filter bounds. Annotation collision shifts pushing a label past the edge is a real generic concern, but it's scope-independent and would need its own reproduction.