New Features

• Add new built-in rendering styles for all combinations of blend and base #719
  • This feature eases another common pain point with blend modes, as a follow-up to enabling blend_order to be set at the draw level in v0.18.0.
  • New built-in rendering styles are now included, to make it easier to use the different blend modes without having to write boilerplate styles. For each combination of blend mode (opaque, overlay, etc.) and style base (polygons, lines, etc.), a style is automatically created at scene load-time with the naming scheme blend_base, e.g. translucent_polygons, overlay_lines, etc. Each of these is a minimal style definition, such as (if expressed in YAML):

  translucent_polygons:
    base: polygons
    blend: translucent
  

  • To maintain backwards compatibility, any of these new "bootstrapped" styles is skipped if the user has already defined one with the same name in their scene.
• Add support for parsing MVT feature properties as JSON #715
  • Some MVT data sources include properties with a richer object format than just strings or numbers -- e.g. arrays, nested objects, etc. The MVT spec doesn't prescribe explicit behavior for these cases, but notes that common tools such as Tippecanoe and Mapnik will encode these properties as stringified JSON.
  • Client-side support for parsing these fields is added through a new parse_json property:
    • If parse_json is true, then each property will be checked to see if it "looks like" stringified JSON (defined as a string with first character being { or [); if so, it is parsed as JSON (with JSON.parse).
    • If parse_json is an array of property names, only those specific properties are checked for JSON parsing. This is preferred to the above, because it limits the parsing impact to only the fields that need it.
    • If parse_json is undefined/null/false, no special parsing of properties is done (e.g. the current behavior).
  • Example usage:

  sources:
    tiles:
      type: MVT
      url: https://...
      parse_json: [prop_a, prop_b] # treat feature properties 'prop_a' and 'prop_b' as stringified JSON
  

• Add native filter syntax for accessing nested feature properties #715
  • Dot notation with . can be used to access nested feature properties (these properties could have been encoded in a GeoJSON/TopoJSON source, or parsed from MVT stringified JSON properties with the feature introduced above). Previously, these properties could only be accessed through custom JS filter functions.
  • Given a feature property a: { b: { c: 'test' } }, this filter will match:
    • filter: { a.b.c: test }
  • Feature property names that include a . can be escaped with \., e.g. a feature property named 'd.e.f': 'escaped' will match with:
    • filter: { d\.e\.f: escaped }
  • These could be mixed, e.g. a property { 'a.b': { c: 'mixed' } would match with:
    • filter: { a\.b.c: mixed }
• Add native filter syntax for array-to-array operations #715
  • As part of the new support for "complex" property values that may be arrays, we can expand our filter syntax to support querying these fields more easily. Two keywords are added for this, extending the same filter object pattern we use for min/max range filter syntax.
  • includes_any: check if an array a contains one or more of the values p, q, r:
    • filter: { a: { includes_any: [p, q, r] } }
  • includes_all: check if an array a contains the values p, q, AND r:
    • filter: { a: { includes_all: [p, q, r] } }
• Add all_layers: true wildcard for data source layer matching #713
  • Introduces a "wildcard" syntax for the source layer in the data block in layers. When the parameter all_layers: true is included, it will match ALL layers in the data source. This is useful for easily creating wireframe-like views, without knowing or needing to specify all the layers in the data source.

Improvements

• More robust support for feature.id #720
  • All data formats supported by Tangram have a feature.id property (at the top feature level, outside feature.properties; some data sets may separately include an id or similar property within feature.properties), but we haven't had full support for it. This version fixes that.
  • Ensures feature.id persists wherever features are internally copied/synthesized/etc. internally
  • Provides access to feature.id in user-authored JS scene functions with a new $id variable
  • Includes feature.id in feature object results returned from scene.queryFeatures()
  • Includes feature.id in the default uniqueness check for scene.queryFeatures(), with $id syntax used if specifying a list of properties to uniqueify on (see #720 for details).
• More intuitive default repeat_distance for point feature labels #718
  • The default repeat_distance for labels has always been 256px across the board. This is a sensible default for cases such as street labels (the original use case for repeat label culling) and road shields, but can generate unexpected results particularly when applied to point labels. For example, when labelling data with a property that has a finite set of values, such as category names ("bar", "restaurant", "hardware store", etc.) or numeric values, it's often surprising/undesirable that only a few labels are plotted.
  • This version changes this behavior such that:
    • For labels generated from point features, the default repeat_distance is 0px.
    • For all other labels, including those generated from line or polygon features (e.g. street names, road shields, park names, etc.), the default repeat_distance continues to be 256px. Note that in these examples, the labels themselves may still be point labels -- it is the underlying geometry type that matters, e.g. road shields are point labels generated from line geometries.
  • See #718 for example.

Bug Fixes

• Handle tiles with no data in scene.queryFeatures() 7946554
• Workaround for obscure bug seen with small (<28px) SVG images encoded as data URLs e1f2869

Performance/Internal

• Add a collision grid for better performance of high density point/label scenes #722
  • Adds a simple collision grid system to drastically reduce the number of collisions performed for dense data sets. In a collision grid, the labels are divided (in this case in screen-space) into a grid of a given size; each label is added to the one or more grid cells that it intersects. When we need to know which labels a given label intersects, we only need to test the "local" labels that are in the same grid cells. See #722 for details.
• Default to WebGL "high performance" mode #721
  • Defaults to using high-performance for WebGL's powerPreference setting. This will ask for use of the discrete GPU where possible. The user can override this default, as with other GL context options:

  const layer = Tangram.leafletLayer({ scene, { webGLContextOptions: { powerPreference: 'low-power' } } });