get OSM Bright style working

[incanus]

From https://github.com/mapbox/osm-bright-gl/blob/master/style.js

/cc @edenh

[incanus]

Working with @edenh on this yields success on OS X and iOS:

[edenh]

🎆
some notes:

• dasharray property ignored
• translate property ignored
• road labels appear intermittently. in general, labels are very sparse
• have not successfully used sprite

[incanus]

Sprite's working for me and in general I have seen labels to be sparse overall previously. I haven't looked into that yet but @kkaefer might have some insight.

[incanus]

iPhone 5s video: https://dl.dropboxusercontent.com/u/575564/osm-native.mp4

[incanus]

POIs aren't coming through, either.

[incanus]

• dasharray: 9e8ace2
• translate: fb581fa

@edenh Try grabbing the latest from the stylequest branch, which includes OSM Bright as the style, and see if the sprite works for you. It does for me (no POIs still, though).

[edenh]

@incanus sprite works. looking good 👍

[incanus]

POIs now work as of c379c9b. Just something odd about (I think) translucency that I still have to look into. Coloring is working ok, though.

(Translation building shadow, also.)

[incanus]

POIs are ok, I think. Bigger problem right now that I'm working on is country/admin/marine line features not showing up.

[incanus]

These line features are related to parsing of vector tiles, not rendering. Getting close, but this part of the code is pretty new to me.

[incanus]

Argghhhh... finally solved.

Turns out this std::find:

https://github.com/mapbox/llmr-native/blob/797f590cfb846e978561f2ff1e6921e37c2d4aff/src/map/vector_tile.cpp#L242

Does not work for numeric feature values (such as when looking for admin_level = 2 and maritime = 1) but does work for string values (most of the other layers).

This happens because the type of number parsed from the protobuf in value.cpp (https://github.com/mapbox/llmr-native/blob/797f590cfb846e978561f2ff1e6921e37c2d4aff/src/style/value.cpp#L6) doesn't necessarily correspond to the type of number iterated out of bucket_desc.source_value (https://github.com/mapbox/llmr-native/blob/797f590cfb846e978561f2ff1e6921e37c2d4aff/src/map/vector_tile.cpp#L241), i.e. it could be comparing 32-bit and 64-bit floats and finding inequality.

So subtle, so much pain.

[incanus]

With this I think stylequest is done and we can merge it into master with OSM Bright as the new default theme for now. Label sparseness is tracked in #162 and nested layers in #140, so I think we're in good shape.

@kkaefer thoughts on merging to master?

[springmeyer]

nice work getting to the bottom of this @incanus. Hope you have a warm dinner or tasty beverage in your future to help unwind this bug from your head.

[kkaefer]

Hah, dang. I meant to fix that by implementing cross-type compare functions a long time ago (I already have an implementation somewhere (i.e. we shouldn't convert everything to a string, but instead cast strings to numbers if possible)

[mikemorris]

Damn you static typing and strict comparisons!

[incanus]

Yeah, I should have guessed at something like this (20/20) but it took me a while to get down to that level, stepping through buckets and manually parsing out vector tile features (as well as learning the vector tile format) but in the end, the way I was debugging at first was outputting things as integers, which the values essentially were, and the opacity of pbf types contributed, too. Anyway, onward! I now know both the tile format and the basics of pbf a lot better.

@kkaefer I've ticketed that in #166 to improve if you can dig up that code.

[incanus]

Stylequest is now merged and OSM Bright is the default style in master.