A vector graphics path — a line, curve, rectangle, or filled shape — extracted from a PDF page, with its zero-based page index and bounding box.

The :operations field holds the path's drawing commands as a list of flat tagged tuples, in stream order:

• {:move_to, x, y} — start a new subpath at (x, y)
• {:line_to, x, y} — straight line to (x, y)
• {:curve_to, c1x, c1y, c2x, c2y, ex, ey} — cubic Bézier curve (the two control points and the endpoint)
• {:rectangle, x, y, width, height} — a complete rectangle subpath
• :close_path — close the current subpath

A path may be stroked, filled, or both: :stroke_color and :fill_color are each a PdfElixide.Color.RGB or nil. Upstream resolves every colorspace to DeviceRGB during extraction, so these are always RGB — never the other PdfElixide.Color structs.

Rectangles and straight lines

PdfElixide.Document.rects/2 and PdfElixide.Document.lines/2 return these same structs, narrowed by a test of :operations alone — of the drawing commands, not of the geometry they describe:

• a rectangle is a lone {:rectangle, …}, or a {:move_to, …} followed by exactly three {:line_to, …}, optionally then :close_path;
• a straight line is a {:move_to, …} followed by exactly one {:line_to, …}, optionally then :close_path.

(Those are vector graphics; PdfElixide.Document.text_lines/2 is the unrelated text extractor.) The two sets are therefore disjoint subsets of paths/2, with four consequences:

• A rectangle drawn back to its starting corner is classified as neither — one {:line_to, …} too many. Boxes drawn with the re operator, the usual case, are unaffected.
• Four points on one straight line are classified as a rectangle, and arrive with a zero-height or zero-width :bbox.
• Degenerate shapes are kept, so filter on :bbox if you need shapes with extent.
• A curve disqualifies a path entirely. A rounded rectangle is neither.

Each call runs a full extraction of the page, so asking one page for both its rectangles and its lines costs twice what asking for either does.

One shape reaches none of the three: a path painted with B, B* or b* (fill and stroke together) is not extracted at all, and its commands are prepended to the next path — changing that path's operation count, so a rectangle or line drawn right after one goes missing too.