"could not find triangle" crash when exporting a grid from SERIAL results

[duncan]

from ticket#160; running \production\onslow_2006\export_results.py for sww file stored in output directory 20060704_063005 produces a similar error - could not find triangle for point. What is important in this example, is that the sww file here was created in sequential mode. The example I ran was to export the grid for the entire model area. It was successful when the cell size was 250, but failed when it was 150. The sww file is large 750Mb. Based on this, Ole suggested reassigning the ticket to you to do the same troubleshooting for this sequential case.

running \production\onslow_2006\test_export_results.py revision 4521 should exibit this ticket

[duncan]

Results from running \production\onslow_2006\test_export_results.py

FitInterpolate: Building mesh
FitInterpolate: Building quad tree
Interpolating
/d/cit/1/dgray/inundation/anuga_core/source/anuga/geospatial_data/geospatial_data.py:602: DeprecationWarning: .xya format is deprecated.  Please use .txt.
  warn(msg, DeprecationWarning)
Traceback (most recent call last):
  File "test_export_results.py", line 47, in ?
    format = 'asc')
  File "/d/cit/1/dgray/inundation/anuga_core/source/anuga/shallow_water/data_manager.py", line 1945, in sww2dem
    grid_values = interp.interpolate(q, grid_points).flat
  File "/d/cit/1/dgray/inundation/anuga_core/source/anuga/fit_interpolate/interpolate.py", line 156, in interpolate
    verbose=verbose)
  File "/d/cit/1/dgray/inundation/anuga_core/source/anuga/fit_interpolate/interpolate.py", line 202, in interpolate_block
    verbose=verbose)
  File "/d/cit/1/dgray/inundation/anuga_core/source/anuga/fit_interpolate/interpolate.py", line 295, in _build_interpolation_matrix_A
    raise Exception(msg)
Exception: ('Could not find triangle for point', array([ 22050.,  59700.]))

[duncan]

the point is very close to the mesh edge. see attachments. Error could be due to slight differences in different algorithms working out if a point is in.

[duncan]

x [ 22050. 59700.] data point

triangle vertices; xi0 [ 22500. 60000.] xi1 [ 22031.25 59687.5 ] xi2 [ 22350.31640625 59716.71484375]

sigma's have to be greater than -epsilon for a point to be in the triangle. sigma0 0.04 sigma1 0.96 sigma2 -3.72309217755e-16 epsilon 2.22044604925e-16

I'll doulble epsilon so this points like this are in.

[ole]

Merely changing epsilon is not a very good fix. See test in changeset:6538 which has a point that is still causing the algorithm to fail.

I am working on a better and more robust solution, separating the decision whether a point is in from the computation of the sigmas. See ticket:314