Background: The slow tests have been failing for a while now. By careful bisection of the git history, this can be traced to the merge of the overhaul of convergence criteria #1448 (the slow tests were not run as part of that PR, I guess me and @jwboth are to blame for that).
The file tests/functional/test_buoyancy_flow.py contains a single test, which, for the discussion below consider the parametrization (BuoyancyFlowModel2N, 3, 4) on branch fix_slow_tests - please use this one (if on develop, use (BuoyancyFlowModel2N, False, 4)). The model is multiphase (in this case 2 phases) flow and transport.
The test makes a single time step with a given absolute residual tolerance for Newton, and verifies that the mass errors are bound by the Newton tolerance. Since the mass conservation equations are represented in the system to be solved, the mass conservation error should be strictly smaller than the Newton tolerance, thus it should not be possible that Newton considers itself converged and the test then fails. However, this is what happens and why the test fails.
Looking updates to this file related to #1448, I understand the idea was to preserve the convergence criteria, just switching to the new framework. Clearly this did not work as intended, or else the tests would not have failed. It is of course important to figure out whether this has wider ramifications.
PS: There is also an error in a slow test for the geothermal reservoir example. I'll have a look at that and possibly write here or file a separate issue.
Background: The slow tests have been failing for a while now. By careful bisection of the git history, this can be traced to the merge of the overhaul of convergence criteria #1448 (the slow tests were not run as part of that PR, I guess me and @jwboth are to blame for that).
The file
tests/functional/test_buoyancy_flow.pycontains a single test, which, for the discussion below consider the parametrization(BuoyancyFlowModel2N, 3, 4)on branch fix_slow_tests - please use this one (if on develop, use(BuoyancyFlowModel2N, False, 4)). The model is multiphase (in this case 2 phases) flow and transport.The test makes a single time step with a given absolute residual tolerance for Newton, and verifies that the mass errors are bound by the Newton tolerance. Since the mass conservation equations are represented in the system to be solved, the mass conservation error should be strictly smaller than the Newton tolerance, thus it should not be possible that Newton considers itself converged and the test then fails. However, this is what happens and why the test fails.
Looking updates to this file related to #1448, I understand the idea was to preserve the convergence criteria, just switching to the new framework. Clearly this did not work as intended, or else the tests would not have failed. It is of course important to figure out whether this has wider ramifications.
PS: There is also an error in a slow test for the geothermal reservoir example. I'll have a look at that and possibly write here or file a separate issue.