SCOREC / core

parallel finite element unstructured meshes
Other
183 stars 62 forks source link

adaptation fails due to transferParametric's default value? #106

Open cwsmith opened 7 years ago

cwsmith commented 7 years ago

posting this for @KennethEJansen

I can also add the phastaChef fails and probably all chef runs that attempt to do an adaptation while using a dmg file.

I don’t have the time to trace back when this change happened but my limited knowledge of this part of the code says that setting in.transferParmetric=1 as a default seems guaranteed to break dmg adapt. Here is a simple diff of what restored it back to working

diff --git a/phasta/phInput.cc b/phasta/phInput.cc
index 4148ad4..c7c2ea6 100644
--- a/phasta/phInput.cc
+++ b/phasta/phInput.cc
@@ -52,7 +52,7 @@ static void setDefaults(Input& in)
   in.restartFileName = "restart";
   in.phastaIO = 1;
   in.snap = 0;
-  in.transferParametric = 1;
+  in.transferParametric = 0;
   in.splitAllLayerEdges = 0;
   in.filterMatches = 0;
   in.axisymmetry = 0;
lines 1-13/13 (END)
cwsmith commented 7 years ago

As suggested, this commit https://github.com/SCOREC/core/commit/776376a2a896f56af5f15e78d9c35f335e1a219e switched the default from off to on and added a .inp control transferParametric for the user to disable it.

So, we should revert transferParametric to off by default and add a phastachef or chef test case that adapts on a dmg model. I think we can convert one of the existing test cases to do this...

yangf4 commented 7 years ago

https://github.com/SCOREC/core/commit/20d360ff472b338206671ba0b55840e0f4fadd66 please let me know if this commit fixes that.