Description
Describe the bug
Cannot pass own solver to class Model constructor because of incorrect imports.
To Reproduce
To reproduce:
import mip
from mip.cbc import SolverCbc
class MySolver(SolverCbc):
pass
class MyModel(mip.Model):
def __init__(self, name: str = "", sense: str = mip.MINIMIZE, *args, **kwargs):
if kwargs["solver_name"].upper() in ("MYSOLVER"):
kwargs["solver"] = MySolver(self, name, sense)
super().__init__(name, sense, *args, **kwargs)
if __name__ == "__main__":
solver = MyModel(solver_name="MYSOLVER") File "C:\nano_sources\shop\shop\models\t_solver.py", line 17, in <module>
solver = MyModel(solver_name="MYSOLVER")
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "...\t_solver.py", line 13, in __init__
super().__init__(name, sense, *args, **kwargs)
File "...\Lib\site-packages\mip\model.py", line 107, in __init__
self.constrs = mip.ConstrList(self)
^^^
UnboundLocalError: cannot access local variable 'mip' where it is not associated with a value
Expected behavior
Basically it is not possible to pass any argument in solver and not get the error above. Please see my bugreport for cpython for explanation why this happens and what is wrong
Desktop (please complete the following information):
- Operating System, version: Windows
- Python version: Probably all of them, tested on 3.11 and 3.12
- Python-MIP version (we recommend you to test with the latest version): 1.15
Additional context
The reason I want to do this are some inconsistencies in handling nans in LinExpr between CBC and Gurobi. We have created model exploiting the CBC behaviour, which is throwing away anything containing nan. This is useful when defining models with LinExprTensor where some of the elements does not contain any variable, leading e.g. to constraints like 1==1. I wanted to create my own solver to make it consistent.