Code is nondeterministic (still/again)
See Job #1195404.
There are at least two problems:
-
The topological sort, which is part of PolyCSE, uses Python sets (https://i10git.cs.fau.de/he66coqe/poly-cse/-/blob/main/poly-cse-py/python/poly_cse_py/sympy_interface.py?ref_type=heads#L167). -
The u_dof_i
accesses in the NonlinearDiffusionNewtonGalerkin form are ordered randomly. I do not know why.
Edited by Daniel Bauer