GLPK: glp_exact ignores verbosity level

[Matthias Köppe]

The GLPK exact solver ignores msg_lev and prints a lot of things on the screen while solving (see tests in #20424).

sage: p = MixedIntegerLinearProgram(solver="GLPK/exact")
sage: p.get_backend().set_verbosity(0)
sage: TestSuite(p.get_backend()).run(skip="_test_pickling")
glp_exact: 5 rows, 1 columns, 4 non-zeros
GNU MP bignum library is being used
...

Reported upstream:

1. ​http://lists.gnu.org/archive/html/help-glpk/2011-10/msg00037.html

2. ​http://lists.gnu.org/archive/html/bug-glpk/2017-08/msg00013.html

[François Bissey]

I am ok with patching but I prefer stuff that is also submitted upstream so we have hope that the functionality will eventually be there (cross fingers).

[David Coudert]

another example.

sage: p = MixedIntegerLinearProgram(solver="GLPK/exact")
sage: b = p.new_variable(nonnegative=True)
sage: p.add_constraint(b[0]+b[1] == 1)
sage: p.solve(log=0)
glp_exact: 1 rows, 2 columns, 2 non-zeros
GNU MP bignum library is being used
      0:   infsum =                      1   (0)
      1:   infsum =                      0   (0)
*     1:   objval =                      0   (0)
*     1:   objval =                      0   (0)
OPTIMAL SOLUTION FOUND
0.0

However, it's not the only issue with this solver:

sage: for solver in ['Cplex', 'Gurobi', 'Coin', 'PPL', 'GLPK']:
....:     p = MixedIntegerLinearProgram(solver=solver)
....:     _ = p.solve()
....:     
sage: p = MixedIntegerLinearProgram(solver="GLPK/exact")
sage: p.solve()
glp_exact: problem has no rows/columns
---------------------------------------------------------------------------
MIPSolverException                        Traceback (most recent call last)
...
MIPSolverException: GLPK: Solver failure

We also have this with another solver...

sage: p = MixedIntegerLinearProgram(solver="CVXOPT")
sage: p.solve()
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
...
TypeError: 'c' must be a dense column matrix

[François Bissey]

Re-reading the summary, is this a deliberate behavior from upstream? If not, ignoring the verbosity level sounds like an upstream bug. In that case I have no problem with going forward with a fix.

[David Coudert]

I had a look at the code of the backends, but I'm not confident with inheritance. The backend of GLPK/exact inherits from the backend of GLPK and has no specific method modifying the verbosity level. So I don't understand what's going on. However, the doc of the backend suggests that the default behavior is verbose (see after cdef class GLPKExactBackend(GLPKBackend)).

[Matthias Köppe]

Replying to fbissey:

Re-reading the summary, is this a deliberate behavior from upstream? If not, ignoring the verbosity level sounds like an upstream bug. In that case I have no problem with going forward with a fix.

I think it is an upstream bug that should be fixed.

Replying to dcoudert: The problem is not in our sage code.

[Jeroen Demeyer]

Replying to fbissey:

Re-reading the summary, is this a deliberate behavior from upstream? If not, ignoring the verbosity level sounds like an upstream bug. In that case I have no problem with going forward with a fix.

+1 to adding a patch in Sage and submitting that patch upstream.

[Jeroen Demeyer]

New commits:

​23102fe	version bump. Slight improvement on skg-src. Fix broken doctest.
​a72d1af	Merge branch 'develop' into glpk4.63
​30c496f	Remove old libraries before installing. This potential enable downrade in the future.
​cc13c4b	Use verbosity level in glp_exact

[git]

Branch pushed to git repo; I updated commit sha1 and set ticket back to needs_review. This was a forced push. New commits:

​fecaf01	version bump. Slight improvement on skg-src. Fix broken doctest.
​baacd6c	Remove old libraries before installing
​4f4fa2c	Use verbosity level in glp_exact

[Jeroen Demeyer]

Rebased on new version of #23596.