[Core] Derive Dof from Flags

[matekelemen]

See #11916

Changes

• derive Dof from Flags
• add new flag FIXED
• replace the mIsFixed boolean member with the FIXED flag
• adjust the serialization/deserialization of Dof to include Flags

[loumalouomega]

Fine from my side, but this must be approved by technical committee. In any case I hope the performance penalty is not super high

[RiccardoRossi]

it is actually a good idea. One problem is that how flags are implemented as of now they occupy 2 doubles, so it is a memory increase. Nevertheless the idea is definitely worth exploring

[matekelemen]

This PR will also open the possibility to make the StructuralMechanicsApplication a bit more consistent.

Kratos/applications/StructuralMechanicsApplication/python_scripts/structural_mechanics_solver.py

Lines 198 to 220 in 358979a

	def AddDofs(self):
	# Append formulation-related DOFs and reactions
	dofs_and_reactions_to_add = []
	dofs_and_reactions_to_add.append(["DISPLACEMENT_X", "REACTION_X"])
	dofs_and_reactions_to_add.append(["DISPLACEMENT_Y", "REACTION_Y"])
	dofs_and_reactions_to_add.append(["DISPLACEMENT_Z", "REACTION_Z"])
	if self.settings["rotation_dofs"].GetBool():
	dofs_and_reactions_to_add.append(["ROTATION_X", "REACTION_MOMENT_X"])
	dofs_and_reactions_to_add.append(["ROTATION_Y", "REACTION_MOMENT_Y"])
	dofs_and_reactions_to_add.append(["ROTATION_Z", "REACTION_MOMENT_Z"])
	if self.settings["volumetric_strain_dofs"].GetBool():
	dofs_and_reactions_to_add.append(["VOLUMETRIC_STRAIN", "REACTION_STRAIN"])
	if self.settings["displacement_control"].GetBool():
	dofs_and_reactions_to_add.append(["LOAD_FACTOR", "PRESCRIBED_DISPLACEMENT"])
	# Append user-defined DOFs and reactions in the ProjectParameters
	auxiliary_solver_utilities.AddAuxiliaryDofsToDofsWithReactionsList(
	self.settings["auxiliary_dofs_list"],
	self.settings["auxiliary_reaction_list"],
	dofs_and_reactions_to_add)
	KratosMultiphysics.VariableUtils.AddDofsList(dofs_and_reactions_to_add, self.main_model_part)
	KratosMultiphysics.Logger.PrintInfo("::[MechanicalSolver]:: ", "DOF's ADDED")

StructuralMechanicsSolver always adds 3D DoFs regardless of the problem's domain size, which results in a mismatch between Dofs stored in nodes and those that are constructed by BuilderAndSolver. In 2D problems, the Z component of vector Dofs are added to Nodes, but their equation ID is never set (meaning that they're always 0).

One possible solution to this issue is setting ACTIVE on Dofs that make it into the B&S.

[matekelemen]

One problem is that how flags are implemented as of now they occupy 2 doubles

I was wondering about that, and got quite confused about what the purpose of Flags::mIsDefined is.

Kratos/kratos/containers/flags.h

Lines 466 to 467 in 358979a

	BlockType mIsDefined; /// Bitmask representing defined flags.
	BlockType mFlags; /// Bitmask representing flag values.

I'd usually assume that a flag is either set or it isn't, but kratos Flags can actually have 3 states:

• not defined, not set
• defined, not set
• defined, set

@RiccardoRossi Do we ever make a distinction between flags that are neither defined nor set, and flags that are defined but not set?

[loumalouomega]

One problem is that how flags are implemented as of now they occupy 2 doubles

I was wondering about that, and got quite confused about what the purpose of Flags::mIsDefined is.

Kratos/kratos/containers/flags.h

Lines 466 to 467 in 358979a

	BlockType mIsDefined; /// Bitmask representing defined flags.
	BlockType mFlags; /// Bitmask representing flag values.

I'd usually assume that a flag is either set or it isn't, but kratos Flags can actually have 3 states:

* not defined, not set

* defined, not set

* defined, set

@RiccardoRossi Do we ever make a distinction between flags that are neither defined nor set, and flags that are defined but not set?

If you remove the defined/not defined you will break a lot of code, because someone (like me) assumed that and we did developments assuming that logic.

[ddiezrod]

One problem is that how flags are implemented as of now they occupy 2 doubles

I was wondering about that, and got quite confused about what the purpose of Flags::mIsDefined is.

Kratos/kratos/containers/flags.h

Lines 466 to 467 in 358979a

	BlockType mIsDefined; /// Bitmask representing defined flags.
	BlockType mFlags; /// Bitmask representing flag values.

I'd usually assume that a flag is either set or it isn't, but kratos Flags can actually have 3 states:

• not defined, not set
• defined, not set
• defined, set

@RiccardoRossi Do we ever make a distinction between flags that are neither defined nor set, and flags that are defined but not set?

Yes, if I remember correctly an element is considered ACTIVE (so we have to compute its local matrix and rhs) if the flag is not defined or if it is set to ACTIVE

Kratos/kratos/sources/geometrical_object.cpp

Line 25 in 9bc21c6

return IsDefined(ACTIVE) ? Is(ACTIVE) : true;

(I think this is exactly the case you were talking about)

[loumalouomega]

One problem is that how flags are implemented as of now they occupy 2 doubles

I was wondering about that, and got quite confused about what the purpose of Flags::mIsDefined is.

Kratos/kratos/containers/flags.h

Lines 466 to 467 in 358979a

	BlockType mIsDefined; /// Bitmask representing defined flags.
	BlockType mFlags; /// Bitmask representing flag values.

I'd usually assume that a flag is either set or it isn't, but kratos Flags can actually have 3 states:

• not defined, not set
• defined, not set
• defined, set

@RiccardoRossi Do we ever make a distinction between flags that are neither defined nor set, and flags that are defined but not set?

Yes, if I remember correctly an element is considered ACTIVE (so we have to compute its local matrix and rhs) if the flag is not defined or if it is set to ACTIVE

Kratos/kratos/sources/geometrical_object.cpp

Line 25 in 9bc21c6

return IsDefined(ACTIVE) ? Is(ACTIVE) : true;

(I think this is exactly the case you were talking about)

And not just that, in Contact I use it for an advance logic, for the frictional contact. I guess I can rewrite it, but It woul not be trivial.

[matekelemen]

Ah well ¯\_(ツ)_/¯

Anyway, reducing the footprint of Flags is not necessary for this PR. That said, I should probably shuffle the order of member variables around to improve alignment.

[loumalouomega]

Ah well ¯_(ツ)_/¯

Anyway, reducing the footprint of Flags is not necessary for this PR

Sotty :P

[matekelemen]

I should probably shuffle the order of member variables around to improve alignment.

well nope, it's already the smallest it can be.

@pooyan-dadvand can you please review?