Tom MacWright

:+1: http://theunitedstates.io/citation/

This is a really interesting thread: we (mapbox) haven't done nearly enough to document how tilelive parts fit together, despite that system being the backbone of tile serving. I'll try...

> Wild, likely wrong guess, does this has to do with missing combinators? No, this is more about additional logic that would need to be implemented in `filterset.js`, which is...

This is going to be a deep one. Right now the entire filter system is written with the assumption that there will be a single `key=*` in each filterset. Also...

I'll open up an experimental branch that doesn't do as much de-duping, but it'll both require the same kind of re-testing that we will need with #20 and might worsen...

Experimental branch at https://github.com/mapbox/carto/tree/natural-filters - it's quite a departure.

By MapBox you mean TileMill, right? This is indeed a bug, and in Carto, due to testing for `testname =` two different values.

Doesn't look like there's anything that hits exactly this, and you've given a good test case, thanks.

This is probably unavoidable. We can make judgments about zoom level filters and other relative filters, like `=`, `

> @tmcw - are you positive this bug is "unavoidable" ? I'm positive that the regex-makes-more-rules bug is unavoidable unless carto or mapnik changes its approach, but it sounds like...