merge GeoJSON vector tiles work

[incanus]

Picking up from #507 (comment), we'll need to merge the geojsonvt branch, which will require it interfacing better with the library.

• match vector tile object structure with main library
• allow for live reindexing when features are added/removed
• handle non-GeoJSON (in-memory features)
• nailing some hopefully-basic performance kinks
• MultiGeometry support
• other API merging (e.g. LatLng struct)
• figuring out how independent of or reliant on GL we want the library to be

/cc @mourner @1ec5

[incanus]

The library consists of geojsonvt.hpp and geojsonvt.cpp.

[incanus]

@kkaefer @mourner Do you guys think this should be a part of GL itself, or should we try to keep it a separate library for general use of geo data -> vector tiles?

[mourner]

@incanus it would be nice to have it as a separate library — easier to maintain this way, with separate tests etc. It could also grow into a useful standalone tool.

[incanus]

Related to #904, we can't subclass Source to make e.g. a LiveSource (my first approach), so instead I am thinking:

• Use Source, possibly with manually-created or passed SourceInfo related to the data (GeoJSON or other) setup.
• Adapt Source::addTile() to handle SourceType:GeoJSON (currently unimplemented) as a first measure.
• Above relies on a new GeoJSONTileData class which works similar to VectorTileData, but which pulls tiles from the GeoJSONVT structure.
• A Source initialized in this way (perhaps with passed GeojSON) would async run the GeoJSONVT clipping & splitting algorithm upon creation so that it would be ready for these tile queries. This shouldn't take long, but maybe it will block addTile() calls until it's done?

[incanus]

FYI above I was speaking of GeoJSON only because that's what GeoJSONVT currently takes (a GeoJSON string) in its conversion, but ideally GL will go right at this conversion regardless of the format or of runtime-added features. We don't really care about GeoJSON file support right now in the iOS beta.

[incanus]

Current status: as of 988e64e, I have some semblance of an ever-present annotations layer atop the style layers, which generates tile coverage properly. Then, instead of kicking off async web loads for tile data, it queries a passed GeoJSONVT object. I can pan and zoom and observe the proper tiles being requested out of the object, including it auto-drilling down to slice up the GeoJSON.

Next steps, roughly:

• Bridge the GeoJSONVT Tile and TileFeature format into VectorTileData that we can render.
• Cut my sample GeoJSON (or any passed GeoJSON, for that matter) out for now.
• Write an API to expose direct writes/conversions into Tile and TileFeature.

[incanus]

Still making slow progress here. Current structure is a LiveTileData peer to VectorTileData & RasterTileData, which will support programmatic annotation adding as well as (eventually) GeoJSON file retrieval/parsing. Just as VectorTileData passes to TileParser, LiveTileData passes to LiveTileParser, which will get refactored using large parts of TileParser such as render bucket creation, but first I'm just getting through the functionality.

[incanus]

As of 645aeda I'm getting things to render a bit.

[incanus]

Next up is some careful looking into the GeoJSONVT library and my use of move/copy and references. I got to the bottom of an issue that was causing modification to throwaway stack variables and thus losing fidelity. So... back to the porting drawing board a bit for some cleanups, then:

• Get all shapes to render on first view.
• Figure out what's causing pieces of PA and all of NJ to not show up (some sort of collision thing or possibly actually losing data unintentionally during conversion).
• Cut out the hardcoded GeoJSON load and go right at the GeoJSONVT conversion routines with live-added data.
• Add APIs to stick stuff into said conversion routines.
• Layer iOS API on top of that plus add API for currently hardcoded annotation styling.

That should keep me busy a while longer.

[incanus]

Get all shapes to render on first view.

As of d36caee this now works, at least within the limits of:

Figure out what's causing pieces of PA and all of NJ to not show up

That's next.

[incanus]

(some sort of collision thing or possibly actually losing data unintentionally during conversion)

It's the latter, as merely drawing background plus the data is no better.