Review of how fixed variables are managed in problem writers

[whart222]

This is prompted by #619 and the discussion related to the resolution of an error in the LP writer when fixed variables are processed. It is not obvious how to extend the tests in pyomo.repn to handle fixed variables, which, more generally, relates to some implicit assumptions regarding how fixed variables arise and managed.

There are a number of questions regarding the writers:

• Do they generate StandardRepn for all objectives/constraints by default, even if repn data is cached?
• Should repn data be cached on the model, or managed by the solver that is leveraging this during resolves?
• Are fixed variables removed? Why or why not?
• What writer options are necessary for managing writing with fixed variables?

[whart222]

Given answers to these questions, an action to resolve this ticket is to (a) confirm that the writers are consistent, and (b) add tests with models that include repn data with fixed variables.

[mrmundt]

@jsiirola - Is this still a valid discussion given the changes to the writers recently?

[jsiirola]

This is still waiting for the BAR and GAMS writer updates, and probably for the overall deprecation of the original implementation of generate_standard_repn (or at least porting that function over to use the new Linear / Quadratic walkers).