format function

[calvinmetcalf]

add the ability to pass a function to the format argument, something like a function that returns a bbox given an object

[mourner]

The reason I didn't do this is that item sorting functions (sort by minX, minY, etc.) heavily affect the performance of the library, and they need to be as simple and direct as possible to perform well. If I accepted a format function that returns a bbox from an item, I would be forced to call it from inside the sorting functions, and performance would drop dramatically. Function compilation performs much better.

[calvinmetcalf]

What I was thinking was a function that would be called when the feature was added not one to be used internally.

[mourner]

I see... This has some implications as well, as you would have to keep all item bboxes separately from items. Is there a really important use case for this that I'm missing, or is it just more of a nice to have for convenience?

[calvinmetcalf]

off the top of my head the use case was consuming geojson that doesn't have the bbox property set but also coordinate transformations would be another conceivable use

[mourner]

I think I'd suggest data pre-processing in this case. E.g. go over all features and set a bbox property. Also, next GeoJSON spec will restrict coordinates to WGS84, so no need for coords convertions. geojson/draft-geojson#6

[calvinmetcalf]

Those are two separate thoughts btw but I was actually thinking in terms of someone wanting to have the tree be in a projected coordinate system not the other way around.

Pre-processing would be the optimal way but that can be applied to having the data in any other format besides the default. If done right (and I can send a pull request) it shouldn't slow down anyone who doesn't use it.

[mourner]

But if it indeed does slow down the algorithm if used, I'd rather encourage data preprocessing than allow using RBush with expectedly subpar performance. And so far I couldn't come up with a way to introduce a format function without affecting the performance much due to the way sorting is used in the algorithm. If you do, please let me know.

[mourner]

Fixed in a way by #14