Add copy/__copy__ methods to CVXOPT, PPL, InteractiveLP backends, and __deepcopy__ to MixedIntegerLinearProgram and backends

[mkoeppe]

The COIN, CPLEX, GLPK, and Gurobi backends supply a copy method.
But it's not documented as a backend interface method in GenericBackend, and the CVXOPT, PPL, InteractiveLP backends do not implement it.
It is used by MixedIntegerLinearProgram.__copy__.

The backend method should actually probably be called __copy__ as well -- see https://docs.python.org/2/library/copy.html

The branch on the ticket does this.

We also add __deepcopy__ methods to MixedIntegerLinearProgram and the backends.
This fixes #15159, though the semantics of copy and deepcopy of a MixedIntegerLinearProgram remains questionable due to its interaction with MIPVariable; see #15159 and #19523.

See also #20323 (in which copying a backend could be the basis for more diverse tests).

CC: @videlec @vbraun @dimpase

Component: numerical

Author: Matthias Koeppe

Branch/Commit: 3117b01

Reviewer: Dima Pasechnik

Issue created by migration from https://trac.sagemath.org/ticket/20414

[mkoeppe]

Branch: u/mkoeppe/add_copy___copy___methods_to_cvxopt__ppl__interactivelp_backends

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. Last 10 new commits:

0bd7a87	GenericBackend._test_add_variables(): Add tests from CVXOPTBackend
6b55e16	Add a classmethod _test_solve
e14afd3	GenericBackend._test_solve: Finish
48b9fe5	Change from mutating instance _test methods to class methods
5394729	New method _test_ncols_nonnegative
7138fa0	GenericBackend: Remove failing _test methods from this ticket to make the patchbot and its friends happy
452b406	Merge branch 'u/mkoeppe/common_testsuite_for_mip_backends' into t/20414/add_copy___copy___methods_to_cvxopt__ppl__interactivelp_backends
8f968ed	_test_copy: New
3e26c4d	_test_copy_does_not_share_data: New
cdd159d	test_copy_some_mips: New

[sagetrac-git]

Commit: cdd159d

[sagetrac-git]

Changed commit from cdd159d to 4b9fd05

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

4b9fd05

InteractiveLPBackend.__copy__: New

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

fe1a8b9	PPLBackend: Implement __copy__
d9f1ec8	CVXOPTBackend: Implement __copy__
751d621	InteractiveLPBackend.__copy__: Fix doctest
9d39e0c	CoinBackend: In add_col, don't forget to append to col_names
7c97097	GenericBackend.__copy__: Fix doctest
ce10c70	GenericBackend._test_copy_some_mips: Change test so it does not use 0 coefficients
580e913	GenericBackend._do_test_problem_data: Compare types as well

[sagetrac-git]

Changed commit from 4b9fd05 to 580e913

[mkoeppe]

comment:5

The branch is on top of #20323 (Common TestSuite for MIP backends).
The _test_copy_some_mips method exposed a bug in CoinBackend, which I have fixed on the way.

[sagetrac-git]

Changed commit from 580e913 to 38c6be9

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

38c6be9

MixedIntegerLinearProgram, GenericBackend: Add `__deepcopy__` methods

[mkoeppe]

comment:7

Now also contains a fix for #15159 (see description).
Needs review.

[mkoeppe]

Author: Matthias Koeppe

[dimpase]

comment:10

How reliable is __deepcopy__ here? Is it just working now?

[sagetrac-git]

Changed commit from 38c6be9 to 3117b01

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

3117b01

MixedIntegerLinearProgram.__deepcopy__: Add another doctest

[mkoeppe]

comment:12

Replying to @dimpase:

How reliable is __deepcopy__ here? Is it just working now?

It's working. I've added another test to illustrate deepcopy semantics.

[dimpase]

comment:13

Replying to @mkoeppe:

Replying to @dimpase:

How reliable is __deepcopy__ here? Is it just working now?

It's working. I've added another test to illustrate deepcopy semantics.

was there something about variables that are (not) copied?

[mkoeppe]

comment:14

Replying to @dimpase:

was there something about variables that are (not) copied?

Before this patch, deepcopy didn't even copy the backend of a MIP.
Now it does, simply by doing the same as copy.

Both copy and deepcopy have weird semantics in relation to the MIPVariables in a problems -- because a MIP does not have an API to gain access to its variables; and trying to use the old MIPVariables with the copy is questionable and definitely broken if one creates new components of this variable, see #15159, #19523. Fixing this would require a redesign; a possible stopgap would be to stop all MIPVariables from creating new components when they have been subjected to copy().

[dimpase]

comment:15

ok, fine.

[dimpase]

Reviewer: Dima Pasechnik

[mkoeppe]

comment:16

Replying to @dimpase:

ok, fine.

Thanks for reviewing.

I've created a new ticket, #20461, for solving the issue with the variables of a copied MIP. Please take a look when you have a chance.

[vbraun]

Changed branch from u/mkoeppe/add_copy___copy___methods_to_cvxopt__ppl__interactivelp_backends to 3117b01