Scattergl selection / clustering fix #2593
Conversation
|
Sweet! @dy can you explain what changes you needed to make to fix the problem? |
|
Amazing. I'll run a few performance tests tomorrow morning to see these changes have a side-effect. |
|
In the meantime @dy, can you confirm that this branch works in IE11? |
|
Sure |
|
Moreover @dy, can you test out the other regl components (regl-line2d, regl-error2d, and sploms) in Safari? |
|
@dy I'm noticing differences in behavior using 1.36.1 and this branch off https://codepen.io/anon/pen/zjoWpO?editors=0010 in 1.36.1:
this branch:
where it looks like 1.36.1 clustered things a little more aggressively making dimmed points appears when selection on a 1e5 pt graph. Is this expected? |
|
Yes, that is the result of additional levels of clustering rendered due to #2586 mentioned by Alex |
|
In terms of performance, comparing 1.36.1 and this branch:
1.36.1:
this branch:
I'd even say panning off this branch perform a little worse than in 1.35.2: https://codepen.io/etpinard/pen/MGpRdW?editors=0010
So, @dy can you explain why panning is affected by your patches here? Does it have to? |
|
Oh and can you add a mock that lock down the fix for #2334? Thanks! |
|
I imagine there might be a way of adaptive opacity based on the density of a cluster, but that better be addressed as a separate issue. |
Ok, but why does a selection fix affect panning? |
|
Slow panning is the direct result of increased number of points caused by #2334 fix. Although it should not be significantly slower than snap-points-2d in 1.35, but possibly can. I did not notice fps drop though. That is the result of z-ordering a point-cluster tree instead of sorting by x, which saves ~100ms of sorting routine on init by displaying extra 10-15% points in runtime. |
|
Although there might be another reason - non-square clustering. It needs a bit of experiment though. |
|
Thanks for the info @dy ! Can you describe what will happen in turns of performance improvement once (if 😉 ) regl-project/regl#474 is fixed? |
|
@etpinard 1k 1-point scattergl traces will be rendered ~6 times faster. If number of points per trace more than 1, it is not that noticeable. |
|
Ok. Is there anything we can do to improve panning performance at 1e6 pts? |
|
Off this branch, zooming in repetitively
leads to:
|
|
Very strong panning perf off dbd6049:
the bug from #2593 (comment) appears to be fixed too 🎉 |
| @@ -549,6 +549,21 @@ describe('@gl Test splom interactions:', function() { | |||
| .catch(failTest) | |||
| .then(done); | |||
| }); | |||
|
|
|||
| it('should work with typed arrays', function(done) { | |||
There was a problem hiding this comment.
locking a fix for #2585 from the new regl-splom version
|
All right. Tests are ✅ Let's merge this thing! 💃 🔈 Big shoutout to @dy for closing 5 issues in one PR. Huge! 🔈 1.37.0 will be out in the next few hours. |
Fixes #2586 (both issues), fixes #2334, fixes #2543, fixes #2457
Changes:
regl._refresh()stub to workaround Constant attribute does not update regl-project/regl#474regl.bufferdata