Closed jdemeyer closed 6 years ago
Doesn't strike me as a release blocker bug...
It is. It yields a massive (orders of magnitude) slowdown on most platforms (x86_64 is the exception here), causing make ptest
to timeout consistently and it is a recent regression.
Author: Jeroen Demeyer
Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:
4c5a373 | Remove vertices containing a NaN or infinity |
Commit: 4c5a373
I thought all along that if there was a problem it had to be with the code that does the triangulation, not with the inputs. It's good that someone was capable of sorting through the code. I see no NaN
s in the output and the surfaces appear to be drawn over the full convex hull. It may be interesting to see what are the CPU savings? But I do see at least one failure
sage -t --long src/sage/plot/plot3d/parametric_surface.pyx
**********************************************************************
File "src/sage/plot/plot3d/parametric_surface.pyx", line 257, in sage.plot.plot3d.parametric_surface.ParametricSurface.obj_repr
Failed example:
s[:2]+s[2][:3]+s[3][:3]
Expected:
['g obj_1',
'usemtl texture91',
'v -2 -2 0',
'v -2 -1.89744 0.399737',
'v -1.89744 -1.89744 0',
'f 1 2 3 4',
'f 2 5 6 3',
'f 5 7 8 6']
Got:
['g obj_1',
'usemtl texture111',
'v -2 -2 0',
'v -2 -1.89744 0.399737',
'v -1.89744 -1.89744 0',
'f 1 2 3 4',
'f 2 5 6 3',
'f 5 7 8 6']
**********************************************************************
1 item had failures:
1 of 5 in sage.plot.plot3d.parametric_surface.ParametricSurface.obj_repr
Replying to @strogdon:
But I do see at least one failure
I cannot reproduce that failure but the original test had texture...
so I'll change that.
Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:
4d17a73 | Remove vertices containing a NaN or infinity |
On my very old, although x86_64, computer
sage -t --long --warn-long 10.0 src/sage/plot/plot3d/list_plot3d.py
[45 tests, 23.20 s]
----------------------------------------------------------------------
All tests passed!
----------------------------------------------------------------------
Total time for all tests: 23.3 seconds
cpu time: 4.4 seconds
cumulative wall time: 23.2 seconds
The results do visually look to be correct. I can't test on anything but x86_64 but if it is slow elsewhere it's not now due to the NaN
s.
Reviewer: Steven Trogdon
Changed branch from u/jdemeyer/list_plot3d_py_should_not_generate_nan_coordinates to 4d17a73
Both #12798 and the matplotlib upgrade (#23696) cause this:
on hardware which is slow in computing with
NaN
values.This plot is done by tachyon but in the input to tachyon, there are many
NaN
s. This can be seen by applying attachment: nancount.patch. The result isCC: @strogdon @kiwifb
Component: graphics
Author: Jeroen Demeyer
Branch/Commit:
4d17a73
Reviewer: Steven Trogdon
Issue created by migration from https://trac.sagemath.org/ticket/13135