TranformationFactory('core.fix_discrete') doesn't work for cloned models

[RomeoV]

Summary

I have noticed that often times TransformationFactory('core.fix_discrete') doesn't work for a model, i.e. it doesn't actually fix the variables. I have found out that it happens to cloned models (whereas 'normal' models work as expected).

Minimal working example

from pyomo.environ import *
m = ConcreteModel()
m.y = Var(domain=Integers, bounds=(0,10), initialize=0)
m_clone = m.clone()
TransformationFactory('core.fix_discrete').apply_to(m_clone)
m_clone.pprint()

yields

1 Var Declarations
    y : Size=1, Index=None
        Key  : Lower : Value : Upper : Fixed : Stale : Domain
        None :     0 :     0 :    10 : False : False : Integers
1 Suffix Declarations
    _fixed_discrete_vars : Direction=Suffix.LOCAL, Datatype=Suffix.FLOAT
        Key  : Value
        None :    []
2 Declarations: y _fixed_discrete_vars

Notice that the variable y didn't actually get fixed. If the TranformationFactory is intead applied to model m, the behaviour is as expected.

Looking into the code

When diving into the code, the problem ends up being when the domain of the variable y is checked against a list of domains here

The expression var.domain in _discrete_relaxation_map will yield False, even though var is an Integer.

[qtothec]

I think this is related to a discrepancy between is and == in the comparison due to the fact that Integer in the clone will have a different id than in the original.

[jsiirola]

This is a fundamental problem with the current Set system (continuous sets are not first-class Set components, and as such do not follow the block scoping rules within Block.clone())

This will be addressed by #326.

[RomeoV]

What can I do until #326 is fixed? It seems like it might be a while. I don't want to 'hand-code' the functionality again for cloned models.

[RomeoV]

I have noticed that the expression type(m.y) is type(m_clone.y) is True, so that has lead me to the following 'dirty' fix:
Replace

var.domain in _discrete_relaxation_map

with

any(type(var.domain) is type(domain) for domain in _discrete_relaxation_map)

This will be slightly slower (although not by that much), but the transformation shouldn't be critical for performance. Additionally, imo it doesn't sacrifice readibility - instead it indicates that we are looking for domains in _discrete_relaxation_map, which was not very clear before.

I can submitt a PR if you think this is acceptable (for now).

[qtothec]

This issue can be seen as a consequence of #165.