Improve GLPK error handling

[jdemeyer]

This ticket adds a proper error handler for GLPK (similar to what we already have for PARI, NTL, ...). It suffices to add sig_on()/sig_off() around a GLPK call to get an error message.

Some more cleanup is also done:

1. change the implementation of MIPSolverException to just inherit from RuntimeError without any custom implementation.
2. use MemoryAllocator in a few places to make some allocations safer.

This branch requires a patch to GLPK, a modified version of it was accepted upstream: ​http://lists.gnu.org/archive/html/help-glpk/2015-11/msg00008.html

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​f808061

Implement GLPK error handling

[git]

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

​f312818

Improve printing of solver exceptions

[git]

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

​18265ee

Rename glp_have_error -> glp_at_error

[jdemeyer]

Upstream claims we are mis-using the GLPK error hook. They are probably right. However, given that I'd like to keep "upgrade GLPK" separated from this ticket, I still propose to review this ticket as-is and then deal with the upgrade on a new ticket.

[dimpase]

Does a reviewer need to create a patched GLPK tarball himself?

[jdemeyer]

Replying to dimpase:

Does a reviewer need to create a patched GLPK tarball himself?

No. This isn't a GLPK upgrade.

[mkoeppe]

Any news on the upstream situation?

[vdelecroix]

does not apply on the latest beta

[jdemeyer]

Any news on the upstream situation?

See 14

[jdemeyer]

Replying to vdelecroix:

does not apply on the latest beta

Are you actually interested in reviewing the ticket? (no offense, but I don't like people complaining that branches don't apply if they have absolutely no interest in reviewing the ticket).

[vdelecroix]

Replying to jdemeyer:

Replying to vdelecroix:

does not apply on the latest beta

Are you actually interested in reviewing the ticket? (no offense, but I don't like people complaining that branches don't apply if they have absolutely no interest in reviewing the ticket).

I will not pronounce about a review. But of course, I was interested in looking at the code.

[git]

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

​5974152

Merge tag '7.2.beta1' into t/19525/improve_glpk_error_handling

[vdelecroix]

hmm

@@ -618,8 +634,14 @@ cdef class GLPKBackend(GenericBackend):
             (2.0, 2.0)
         """
         cdef int n = glp_get_num_cols(self.lp)
+<<<<<<< HEAD
+        cdef MemoryAllocator mem = MemoryAllocator()
+        cdef int * c_indices = <int*>mem.allocarray(n+1, sizeof(int))
+        cdef double * c_values = <double*>mem.allocarray(n+1, sizeof(double))
+=======
         cdef int * c_indices = <int*> sig_malloc((n+1)*sizeof(int))
         cdef double * c_values = <double*> sig_malloc((n+1)*sizeof(double))
+>>>>>>> 7.2.beta1
         cdef list indices = []
         cdef list values = []
         cdef int i,j

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​5211376

Merge tag '7.2.beta1' into t/19525/improve_glpk_error_handling

[vdelecroix]

import "cysignals/signals.pxi" instead of "sage/ext/interrupt.pxi", see #20002 for details ;-)

[git]

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

​3e33342

Use cysignals instead of interrupt.pxi

[vdelecroix]

A commit to fix changes in error messages.

---

New commits:

​71f52a4

Trac 19525: fix error messages for cplex and gurobi

[mkoeppe]

See also #10232

[vdelecroix]

Indeed, in order to fix #10232 there need to be the appropriate sig_on and sig_off around the GLPK calls.

[vdelecroix]

Jeroen, do you prefer another ticket to add the appropriate sig_on/sig_off around GLPK calls?

[vdelecroix]

If yes, then the ticket is good to go from my point of view.

[vbraun]

Fails on 32-bit

sage -t --long src/sage/numerical/backends/glpk_backend.pyx
**********************************************************************
File "src/sage/numerical/backends/glpk_backend.pyx", line 847, in sage.numerical.backends.glpk_backend.GLPKBackend.solve
Failed example:
    lp.solve()
Expected:
    Traceback (most recent call last):
    ...
    GLPKError: Assertion failed: col->lb < col->ub
    Error detected in file glpnpp05.c at line ...
Got:
    <BLANKLINE>
    Traceback (most recent call last):
      File "/home/buildbot/slave/sage_git/build/local/lib/python2.7/site-packages/sage/doctest/forker.py", line 496, in _run
        self.compile_and_execute(example, compiler, test.globs)
      File "/home/buildbot/slave/sage_git/build/local/lib/python2.7/site-packages/sage/doctest/forker.py", line 858, in compile_and_execute
        exec(compiled, globs)
      File "<doctest sage.numerical.backends.glpk_backend.GLPKBackend.solve[24]>", line 1, in <module>
        lp.solve()
      File "sage/numerical/mip.pyx", line 2115, in sage.numerical.mip.MixedIntegerLinearProgram.solve (/home/buildbot/slave/sage_git/build/src/build/cythonized/sage/numerical/mip.c:14757)
        self._backend.solve()
      File "sage/numerical/backends/glpk_backend.pyx", line 1005, in sage.numerical.backends.glpk_backend.GLPKBackend.solve (/home/buildbot/slave/sage_git/build/src/build/cythonized/sage/numerical/backends/glpk_backend.cpp:8406)
        sig_on()
    GLPKError: Assertion failed: q->lb < q->ub
    Error detected in file glpnpp03.c at line 557
**********************************************************************
1 item had failures:
   1 of  76 in sage.numerical.backends.glpk_backend.GLPKBackend.solve

[jdemeyer]

This should do the trick...

---

New commits:

​76a11fd

Use ellipsis in assertion error message