Re: [Help-glpk] MILP and CP

[Andrew Makhorin]

> 
> Andrew, in other way: you have ever considered (or someone suggested
> or asked for) add some CP syntax to GMPL? Like implications os
> allDiferent constraints? Ok, it will extrapolate AMPL, but can it be
> considered?
> 

In glpk it would be useless, because the package itself and so gmpl
translator are designed for ordinary numeric variables and linear forms.
On the other hand, it is possible to mark some constraints as "special"
ones (say, by using some special tags in constraint names), e.g.

   foo$all_different${i in I}: sum{j in S[i]} x[j];

or

   foo$sos1${i in I}: sum{j in S[i]} x[j];

and then treat such constraints in an appropriate way assuming that the
(non-glpk) solver is able to understand that. In this case no changes in
the language and translator are ever needed.

On developing the gmpl translator there was an idea to allow some
classes of specific variables and constraints (e.g. semi-continuous
variables, sos1 and sos2 constraints, disjunctive constraints, etc.),
However, the problem arised was that such constraints should be
transformed to ordinary linear constraints, and since it can be done in
numerous ways, it would be impossible to automatically choose a
preferable way for a particular model, so I decided not to implement
such a feature.


Andrew Makhorin