summary tables

[Fil]

just a prototype… there's still a lot to do

This works in user land, which makes it easy to hack on and also means that anyone will be able to substitute whatever we come up with with their own code.

I'm beginning to think that this (when it's ready) should be the default inspector for Arrow tables, and not only tied to the sql fenced code blocks. (I don't think anyone will miss the current inspector which shows a Ho object with cryptic properties—and if you need it to debug something, you can just put the Arrow table inside an Array.)

I don't use DuckDB (yet) but maybe it will become necessary when we add filtering. But there will be a trade-off because setting up a DuckDB instance can only be slower than not doing it. We'll see what's easiest and fastest.

I also use Plot for faster prototyping, but the actual implementation will likely use d3. We'll need some kind of preprocessing of the channels, so we don't compute the bins twice.

~~The bin width is also often too small. I'm "hacking" the rects so that they are always visible, but a better implementation will be needed…~~ (and maybe that's a flaw in the default strategy we're using in Plot to decide the number of bins).

closes #23

[mbostock]

default inspector for Arrow tables

Yes, I like this idea. 👍

[Fil]

TODO:

• [ ] detect dates since they are numbers as of apache-arrow@16 (see #1375)
  • [x] solved in https://github.com/observablehq/inputs/pull/263