tests related to cplex / gurobi

[vdelecroix]

• File "numerical/backends/generic_backend.pyx", line 1153, the following doctest is wrong if gurobi is available

  sage: default_mip_solver("GUROBI")
  Traceback (most recent call last):
  ...
  ValueError: Gurobi is not available. Please refer to the
  documentation to install it.
  

• File "numerical/backends/cplex_backend.pyx", line 861

  Failed example:
      p.solve()                                                      # optional - CPLEX
  Expected:
      Traceback (most recent call last):
      ...
      MIPSolverException: 'CPLEX: The primal has no feasible solution'
  Got:
      Traceback (most recent call last):
      ...
      MIPSolverException: CPLEX: The primal has no feasible solution
  

• File "numerical/backends/gurobi_backend.pyx", line 686

  Failed example:
      p.solve()                                                       # optional - Gurobi
  Expected:
      Traceback (most recent call last):
      ...
      MIPSolverException: 'Gurobi: The problem is infeasible'
  Got:
      Traceback (most recent call last):
      ...
      MIPSolverException: Gurobi: The problem is infeasible
  

• File "numerical/backends/gurobi_backend.pyx", line 449

  Failed example:
      p.get_values([x,y])                          # optional - Gurobi
  Expected:
      [0.0, 3.0]
  Got:
      [-0.0, 3.0]
  

• File "numerical/backends/gurobi_backend.pyx", line 484

  Failed example:
      p.row(0)                                                                # optional - Gurobi
  Expected:
      ([0, 1, 2, 3, 4], [0.0, 1.0, 2.0, 3.0, 4.0])
  Got:
      ([1, 2, 3, 4], [1.0, 2.0, 3.0, 4.0])
  

• File "numerical/backends/gurobi_backend.pyx", line 554

  Failed example:
      p.row(0)                                                         # optional - Gurobi
  Expected:
      ([0, 1, 2, 3, 4], [0.0, 1.0, 2.0, 3.0, 4.0])
  Got:
      ([1, 2, 3, 4], [1.0, 2.0, 3.0, 4.0])
  

• File "numerical/backends/gurobi_backend.pyx", line 607

  Failed example:
      p.row(0)                                                          # optional - Gurobi
  Expected:
      ([0, 1, 2, 3, 4], [0.0, 1.0, 2.0, 3.0, 4.0])
  Got:
      ([1, 2, 3, 4], [1.0, 2.0, 3.0, 4.0])
  

[mkoeppe]

Some of these were fixed in #19525.

[mkoeppe]

Here's a fix for the remaining failures with Gurobi.

mip.pyx is no longer testing with the default MIP solver; now always uses GLPK. Testing with the default MIP solver seems like a bad idea. We'll ensure consistency of the backends using the new backend testsuite introduced in #20323, #20424.

---

New commits:

​658972c	Fix doctests when Gurobi is installed
​cfaf157	Instead of running MixedIntegerLinearProgram doctests with the default solver, use GLPK

[mkoeppe]

I've tested with current Gurobi and current CPLEX on Mac OS X. All tests pass.

[mkoeppe]

Follow-up: #20600