Remove "fill-antialias" property

[lucaswoj]

From @ansis on June 11, 2015 19:39

fill-antialias is useless. Theoretically you could disable antialiasing to get a tiny performance improvement if the fill edges are covered by something else (a thick line outline). In practice the performance improvement is insignificant.

Copied from original issue: mapbox/mapbox-gl-style-spec#289

[lucaswoj]

From @nickidlugash on June 15, 2015 15:59

@ansis Want to add to v8?

[lucaswoj]

From @ansis on June 15, 2015 18:25

added

[lucaswoj]

From @kkaefer on June 16, 2015 11:24

In practice the performance improvement is insignificant.

Not sure about this. Do you have any numbers? I've seen performance boosts for detailed polygons, like forest areas.

[lucaswoj]

From @ansis on June 16, 2015 15:37

nope, I don't have any numbers. I was guessing that since the property could be used without visual defects in only a very small number of cases it wouldn't have any big impact on total performance.

I have a hard time imagining anyone actually using this property in a style, so I don't think we should have it in the editor or spec.

I'm fine with keeping it if you think people will actually use it.

[lucaswoj]

From @jfirebaugh on August 6, 2015 18:47

This doesn't block app, right? If so let's not change it in v8.

[lucaswoj]

From @ajashton on January 25, 2017 2:4

When dealing with transparent overlapping shapes and subtle color shifts the aliased polygons tend to actually look better (especially on hidpi). We have fill-antialias disabled in hillshading layers for several styles to avoid problems with overlapping transparency (note both the seams due to lack of buffers and the darker outline around the full outline of each hillshade polygon):

I would also like to see some numbers re: performance on some dense layers such as hillshading and landuse.

[xabbu42]

We also disable antialiasing on zoomed out tiles for our forest polygons as with antialiasing they tend to all blend together to one big area (compare the lower left corner of the following tiles):

By the way is this stark visual difference from antialiasing expected or should I open a new issue about it?