Kai, can you take a look at this?

I was looking into this during the F2F as well when Brandon mentioned it. I'm actually very confused about the stats, I may need to ask Florian about them. Specifically:

Windows Chrome + Android Chrome, shows a sharp drop:
https://webglstats.com/webgl/extension/EXT_texture_filter_anisotropic?platforms=000002000010000000

Windows Chrome, no drop:
https://webglstats.com/webgl/extension/EXT_texture_filter_anisotropic?platforms=000002000000000000

Android Chrome, no drop:
https://webglstats.com/webgl/extension/EXT_texture_filter_anisotropic?platforms=000000000010000000

I think we should add histograms to extensions/caps in Chrome to get more accurate data. Florian's sample size is small and biased.

I just noticed something I missed before: Along with the percentages there's a number reported in parens that I think is the total number of samples. This number jumps significantly at the same point as the support falloff. Looking at the daily graph tt reports 95% (162.8k) on 6/15 and 68% (540.8k) on 6/17. It's possible that Florian simply gained a new and more popular embedder at that time.

Both desktop and mobile see a jump in usage after that date, but mobile's increase is larger relative to it's prior levels so it has and oversized impact on the total percentage. That's why neither mobile or desktop alone show a sharp drop (though mobile is trending down worryingly) but together there's a cliff.

So it seems like this isn't anything to worry about. Sorry for the noise! Moral of the story is to be sure to understand all factors of the graphs you're looking at. I agree with zmo@ that our own histograms for these type of data would be nice.

More details that Brandon and I just looked at together:

On Chrome on Android:
https://webglstats.com/webgl/extension/EXT_texture_filter_anisotropic?platforms=000000000010000000

the number of samples on 2017-06-05 was 48.3K. On 2017-06-19 that number was 1.8M. This is nearly a factor of *40* increase in the amount of data -- which is awesome!

However, on Chrome on Windows:
https://webglstats.com/webgl/extension/EXT_texture_filter_anisotropic?platforms=000002000000000000

the number of samples on 2017-06-05 was 681.5K, and on 2017-06-19 was 1.5M -- a factor of 2.2 increase.

This means that the Android numbers dominate relatively more in the more recent data -- specifically, after that cliff. Florian's numbers indicate 52% support on Chrome on Android, which is why the number decreased at that time.

Closing as WontFix -- not a bug.

Awesome, thanks Brandon and Ken for figuring this out. I assume this is also the source of the drop in draw_buffers support that I think we saw during the F2F.