canvas

rpic extension. canvas from <position> to <position> fixes the output page to the rectangle spanned by the two corners, independent of the drawn content.

Without it, the viewBox hugs the content’s bounding box — so moving the leftmost or topmost object shifts the whole drawing on screen. With a fixed canvas, the frame’s origin and size are stable: dragging one object moves only that object, which is exactly what a visual editor needs.

.PS
canvas from (0,0) to (4,2)
box "stays put" fit at (1,1)
circle rad 0.3 at (3,1.3)
.PE

Anchoring to geometry

The corners are ordinary pic positions, so the page can be pinned to an invisible frame:

.PS
F: box wid 3.5 ht 1.6 at (1.75,0.8) invis
canvas from F.sw to F.ne
box "framed" fit at (0.8,0.8)
.PE

Rules

• The rectangle must have positive width and height; corners may be given in either order. The last canvas statement wins.
• Content outside the canvas is clipped by the viewBox (the SVG default; the PNG/PDF renderers behave the same way).
• It composes with the framing pipeline: coordinates are user units (they follow scale), the margin variables add whitespace around the fixed page, and .PS width sizing plus maxpswid/maxpsht clamping scale page and content together.
• Contextual keyword: only the exact canvas from … spelling triggers. canvas = 2, a macro named canvas, or any other use of the name keeps its classic meaning — unused, output is byte-for-byte classic pic.

For editors

The per-object geometry export (compile() → objects, with each shape’s bbox in viewBox units and its source span) uses the same frame. Together they close the loop for bidirectional editing: stable coordinates out, precise source positions back.