More GeometryInfo work, the questionable part. #4430
Conversation
|
/rebuild |
There was a problem hiding this comment.
Maybe the tester was a bit overzealous when comparing the numbers in the other PR. I remember that the difference was something like 1e-8 and I thought our numdiff tolerance was coarser anyway. But could you replace the 1 << x line with what I suggested earlier: (dim == 3) ? 4 : 2;? I think it is just easier to understand.
|
Bizarre that it works now. I've made the change you requested (just the other way around :-) ) |
source/geometry_model/chunk.cc
Outdated
| const auto min_distance_to_first_vertex = | ||
| std::min_element(distances_to_first_vertex.begin(), | ||
| distances_to_first_vertex.end()); | ||
| distances_to_first_vertex.begin()+face->n_vertices()); |
There was a problem hiding this comment.
Oh, I think I see what is happening, another random test failure due to access to non-initialized memory. Does this need to be the following? (see the size of the array above)
| distances_to_first_vertex.begin()+face->n_vertices()); | |
| distances_to_first_vertex.begin()+face->n_vertices()-1); |
There was a problem hiding this comment.
Oh, you're totally right! How insidious! How did this work before?
source/geometry_model/chunk.cc
Outdated
| const auto min_distance_to_first_vertex = | ||
| std::min_element(distances_to_first_vertex.begin(), | ||
| distances_to_first_vertex.end()); | ||
| distances_to_first_vertex.begin()+face->n_vertices()); |
There was a problem hiding this comment.
Oh, you're totally right! How insidious! How did this work before?
This is the part of #4425 that didn't work yesterday. I checked that at least on my system, there is no difference between the before/after outputs of the
merged_chunks_2Dtest that failed yesterday, though the output is (expectedly) slightly different than what's stored./rebuild