Refspace fixes - #4538
Conversation
I don't think this is a slow one but it's worth logging.
I'm not sure how we had the wrong data here.
Rochi found bugs in our higher-order Pyramids reference elements, and these tests catch them.
|
I think we may have a threading race condition problem with |
| {-0.5, -0.5, 0.5}, | ||
| {0.5, -0.5, 0.5}, | ||
| {0.5, 0.5, 0.5}, | ||
| {-0.5, 0.5, 0.5}, | ||
| {0, 0, 0}, | ||
| {0, -Real(2)/3, Real(1)/3}, | ||
| { Real(2)/3, 0, Real(1)/3}, | ||
| {0, Real(2)/3, Real(1)/3}, | ||
| {-Real(2)/3, 0, Real(1)/3} |
There was a problem hiding this comment.
Hmm... I don't think I realized that the base class's _master_points array was meant to have entries for all possible subtypes that might exist? Maybe we should document that better somehow?
There was a problem hiding this comment.
I'll add docs for that (to this PR, which will end up being a "just consolidate reference point definitions" after I rebase it post-#4539 and fix the race condition), yeah ... and in the comments in the .C I'll probably explicitly mention the types that are currently supported, because IIRC I tried to make sure I was finding all the code that needed changing by e.g. searching for "Pyramid13" and "PYRAMID13", and those searches totally missed these spots.
3 participants
In one of @rochi00's branches his new test coverage discovered some utterly broken master point coordinate definitions for Pyramid13, Pyramid14, and especially Pyramid18. I fixed those, added some tests for the fix, and it turns out Tri7/Prism20/Prism21 also had issues. All fixed and in test coverage now.