Everything happens in the browser
When you add a component, R only stores its settings in the map. The component itself is a small piece of JavaScript that runs in the reader’s browser: it reads the layer’s data once, computes the figures for the whole map, and recomputes the figures for the view each time the map moves. That is why the components work in static HTML and Quarto documents, without a Shiny server.
What counts as “in view”
A polygon counts as in view when any part of it is on screen; a point, when the point is on screen. Features outside the layer’s zoom range are not counted.
Relying only on what the map has drawn, with MapLibre’s
queryRenderedFeatures(), is not enough:
- features too small to be drawn at low zoom are missed;
- on the globe projection, some features far from the center are missed, and features on the far side of the globe can be returned as if they were visible.
So maplore combines both: a feature is in view if it was drawn
or a geometric test finds it on screen, and on the
globe only features on the visible side count. For points, the geometric
test alone is used. The counts have been checked against sf
and against pixel positions at several zoom levels, with rotation and
pitch.
When the layer has a filter, only features drawn on
screen are counted, since the geometric test cannot apply the filter.
Very small features may then be missed at low zoom.
Performance
Updating takes a few milliseconds for hundreds of polygons or a few thousand points. With 50,000 points, a summary and a ranking together take about 20 ms with all of them on screen and still update while panning; searching them takes about 2 ms. If an update takes longer than 25 ms (around 200,000 points), the components update only when the map stops moving, so panning stays smooth.
Several components on the same layer share the work of finding the features in view.