1 doctest failing in combinat/designs/incidence_structures.py

[seblabbe]

On Ubuntu 18.04 with 9.4.beta3 with the following installed optional packages:

Using --optional=4ti2,build,ccache,cryptominisat,debian,dochtml,dot2tex,
e_antic,fricas,glucose,latte_int,lidia,normaliz,notedown,
pandoc_attributes,pip,pycosat,pynormaliz,rst2ipynb,sage,
sage_numerical_backends_coin,sage_spkg

the command

sage -t src/sage/combinat/designs/incidence_structures.py

gives

sage -t --random-seed=0 src/sage/combinat/designs/incidence_structures.py
**********************************************************************
File "src/sage/combinat/designs/incidence_structures.py", line 2027, in sage.combinat.designs.incidence_structures.IncidenceStructure.coloring
Failed example:
    len(designs.steiner_triple_system(7).coloring())
Expected:
    3
Got:
    2
**********************************************************************
1 item had failures:
   1 of   6 in sage.combinat.designs.incidence_structures.IncidenceStructure.coloring
    [338 tests, 1 failure, 0.93 s]
----------------------------------------------------------------------
sage -t --random-seed=0 src/sage/combinat/designs/incidence_structures.py  # 1 doctest failed
----------------------------------------------------------------------

• Depends on Deprecate sage_numerical_backends_coin in favor of cvxpy+cylp, upgrade cbc to 2.10.10 #35368

CC: @mkoeppe @slel

Component: doctest coverage

Keywords: coin

Branch/Commit: u/mkoeppe/1_doctest_failing_in_combinat_designs_incidence_structures_py @ b07d203

Issue created by migration from https://trac.sagemath.org/ticket/32053

[slel]

comment:1

The coloring method of Steiner triple systems
calls MixedIntegerLinearProgram, which takes
a solver argument, defaulting to None, in
which case default_mip_solver is used.

Testing on two different Sage installations:

sage: default_mip_solver()
'Glpk'
sage: len(designs.steiner_triple_system(7).coloring())
3

sage: default_mip_solver()
'Coin'
sage: len(designs.steiner_triple_system(7).coloring())
2

A previous report was at #30637.

[slel]

Changed keywords from none to coin

[mkoeppe]

comment:2

This will need testing in the next beta, which has many fixes for the use of MIP (see Meta-ticket #32191)

[sheerluck]

comment:3

bug is in line 2048 of src/sage/combinat/designs/incidence_structures.py

return self.coloring(k)

Recursive call loses value of solver:

sage: designs.steiner_triple_system(7).coloring(solver="literally whatever")
[[1, 2, 3, 4], [0, 5, 6]]

fix would be

return self.coloring(k, solver=solver)

That wolud lead to

sage: designs.steiner_triple_system(7).coloring(solver="GLPK")
[[0, 1], [3], [2, 4, 5, 6]]
sage: designs.steiner_triple_system(7).coloring(solver="Coin")
[[1, 2, 3, 4], [0, 5, 6]]
sage: designs.steiner_triple_system(7).coloring(solver="PPL")
[[0, 1, 2, 5], [3], [4, 6]]

[sheerluck]

comment:4

Even better:

return self.coloring(k, solver=solver, verbose=verbose)

[mkoeppe]

Dependencies: #32218

[mkoeppe]

Branch: u/mkoeppe/1_doctest_failing_in_combinat_designs_incidence_structures_py

[mkoeppe]

Commit: b07d203

[mkoeppe]

comment:7

Did this on top of #32218.

---

Last 10 new commits:

5a74113	MixedIntegerLinearProgram._backend_variable_value: Add docstring
aa4eed5	MixedIntegerLinearProgram._backend_variable_value*: Add docstrings, examples
00b34aa	trac #32197: review commit
db83853	trac #32197: another review commit
2038cc7	BalancedIncompleteBlockDesign.arc: Use get_values(..., convert=bool, tolerance=integrality_tolerance)
7ef58e9	sage.combinat.designs: Fix remaining uses of MixedIntegerLinearProgram.get_values
5af993f	trac #32218: review commit
66c5918	Merge #32218
c77191b	IncidenceStructure.coloring: Change doctest from # not tested to # random
b07d203	IncidenceStructure.coloring: In self-call, pass on parameters

[mkoeppe]

comment:8

I cannot reproduce this behavior - which is clearly a bug

sage: designs.steiner_triple_system(7).coloring(solver="Coin")
[[1, 2, 3, 4], [0, 5, 6]]

[mkoeppe]

comment:9

With which version of CBC do you get this result? We observed similar trouble in #30635

[sheerluck]

comment:10

master branch of https://github.com/coin-or/Cbc with --enable-cbc-parallel because coinor-cbc-2.10.5.ebuild in Gentoo does not have --enable-cbc-parallel and that leads to failed assert(masterThread_); in CbcModel::doOneNode

[sheerluck]

comment:11

I mean stable/2.10 not master. It is 2.10.5 + 43 commits

[mkoeppe]

comment:12

Yes, I also saw these problems in #30635 with the upgrade to cbc 2.10.5 in #30644.

I think the coin interface (https://github.com/sagemath/sage-numerical-backends-coin) needs to be fixed. See #30635 comment:41 for a possible fix.

I won't have time to work on this before mid August unfortunately.

[sheerluck]

comment:14

I noticed some activity at https://github.com/coin-or/Cbc.git then I rebuilded [Cbc, Cgl, Clp, CoinUtils, Osi] from master, saw same behavior for solver="Coin" and decided to ping here (I failed to fix sage-numerical-backends-coin)

[fchapoton]

comment:15

bump to 9.6

[sheerluck]

comment:16

for current master branches of coinor-utils coinor-osi coinor-cgl coinor-clp coinor-cbc one can build sage_numerical_backends_coin only without "CbcSolver" in coin_backend.pyx:

--- a/sage_numerical_backends_coin/coin_backend.pyx
+++ b/sage_numerical_backends_coin/coin_backend.pyx
@@ -1,5 +1,5 @@
 # distutils: language = c++
-# distutils: libraries = Cbc CbcSolver Cgl Clp CoinUtils OsiCbc OsiClp Osi
+# distutils: libraries = Cbc Cgl Clp CoinUtils OsiCbc OsiClp Osi
 """
 COIN Backend
 """

[sheerluck]

comment:17

I was hoping to use #31962 this way:

from sage.numerical.backends.generic_sdp_backend import get_solver
designs.steiner_triple_system(7).coloring(solver=lambda: get_solver("cvxpy/CBC"))

but I got

TypeError: Cannot convert CVXPYSDPBackend to sage.numerical.backends.generic_backend.GenericBackend

[mkoeppe]

comment:18

So far #31962 only provides an (unfinished) SDP backend, not an LP/MIP backend. This would have to be written separately. Any help is welcome on #31962!

[sheerluck]

comment:19

when #31962 is merged and closed, "Coin" should not have a chance to be default_mip_solver.

Only if "Coin" is replaced with "cvxpy/CBC" then len(designs.steiner_triple_system(7).coloring()) will be 3 on every Sage installations again.

[sheerluck]

I have a feeling this issue is solved in master branch