issue when not adding a constraint directly to a model

[sdementen]

If one wants to build constraints without adding the directly to a model, the CyLPExpr is "corrupted".

Indeed, each operation on a CyLPExpr has the strong side effect of modifying in place the CyLPExpr.expr (for instance https://github.com/coin-or/CyLP/blob/master/cylp/py/modeling/CyLPModel.py#L191). It is only at the end of the CyLPExpr.evaluate (called from the CyLPModel.addConstraint) that the CyLPExpr.expr is reset.
This side effect is most probably there to support the a <= x <= b notation which is translated by python in (a <= x) and (x <= b) requiring the last element (x <= b) to evaluate to the CyLPExpr.expr representing a <= x <= b.

However, it introduces the quite non pythonic behavior that a CyLPExpr cannot be instantiated without directly assigning to a model.
i.e, the following

model = CyLPModel()

y = model.addVariable("y", 1)
z = model.addVariable("z", 1)

constraints = [
    z - y == 0,
    z >= -5, 
]
for c in constraints:
    model += c

is not equivalent to

model = CyLPModel()

y = model.addVariable("y", 1)
z = model.addVariable("z", 1)

model += z - y == 0
model += z >= -5

Indeed, in

constraints = [
    z - y == 0,
    z >= -5, 
]

when the second constraint z >= -5 is evaluated, the CyLPVar.expr of z has been changed when the first constraint z - y == 0 was defined and is equal now to ´z - y´ leading to the wrongly evaluated expression z - y >= -5

This bug has bitten me hard ;-) and forces the user to assign the constraint directly to the model to avoid the bug.

In other LP modeling framework using python syntax, the alternative solutions to express a <= x <= b are:

• either to disallow the ability to see both bounds at once (the user should go with a <= x and x <= b)
• use the notation x == (a,b) to mean a <= x <= b (my preferred option)