Changing the way PuLP reports the status

[pchtsp]

Hello everyone,

There's lately been several comments on the way PuLP handles solution statuses when solving. The fact is that I've never been happy with how it handles them. Last year I added a not-very-documented LpProblem.sol_status property that gives the actual status of the solution but there's still the LpProblem.status ambiguous codes. My objective is to report the following two status types in PuLP:

1. The first one, obtained via LpProblem.getSolverStatus(), will return the reason why the solver stopped: optimal, time limit, not solved, node limit, memory limit, undefined, unbounded, unfeasible, etc.
2. The second one, obtained via LpProblem.getSolutionStatus() will return the status of the solution when the solver stopped: optimal, feasible, unfeasible, no solution found, etc.

I hope this change will make it more clear and it will also better match the codes returned by the solvers.

This implies:

• Deprecating the querying of LpProblem.status with a warning.
• Changing all the solver interfaces to get the precise solver status and solution status.
• Changing all the documentation and examples so they match this new way of reporting.

Despite the soft deprecating of LpProblem.status, this will probably break some things in people's code because LpProblem.status will now return slightly different information (now it returns 1 = "Optimal" when it found at least one feasible solution).

[estellederrien]

Hello,
I am a pulp fan, I use it in my (unofficial) app solvgraph .
AnyOne who wanna chat is welcome, And I wanna pay and support too, someday

[wookenny]

I'm a giant fan of the proposed change! 🥇
There was a lot of confusion for me initially to understand that

(now it returns 1 = "Optimal" when it found at least one feasible solution).

Having a status feasible if a solution was found, but not necessarily an optimal one is much better than optimal.
Yes, we would need to change code, but in particular code that was creating a logic similar to what you describe here.

[wookenny]

IMHO: If pulp uses a new version, it seems okay to break old code.
We had to change for the jump to version 2.0 but it was not a big issue.

[ElCrisLWH]

That would be much more intuitive.
I lost a few hours thinking that my code might have a bug because it returned Optimal when it was stopped by the time limit.
Another option that doesn't include breaking old code could be to document the not well-documented status and sol_status as you said.