Installing ChomP causes ChainComplex to return incorrect answers

[NathanDunfield]

Installing the ChomP package causes ChainComplex to give different (and basically wrong) answers. Here are two examples from the doctests:

sage: C = ChainComplex([matrix(ZZ, 1, 1, [2]), matrix(ZZ, 1, 1), matrix(1, 1, [3])])
sage: C.torsion_list(5)
[(2, []), (3, [])]   # Should be [(2, [1]), (3, [3])]
sage: S = simplicial_complexes.SumComplex(17, [0, 1, 4])
sage: S.homology(1)
0    # Should be C140183

May be related to #12607 where installing ChomP causes certain 0 dimensional homology groups not to be reported.

Upstream: Reported upstream. No feedback yet.

CC: @jhpalmieri @kini

Component: packages: optional

Keywords: chomp

Issue created by migration from https://trac.sagemath.org/ticket/16364

[NathanDunfield]

comment:2

I poked around a bit. In the first example, the issue seems to be that the wrapper is misparsing the correct output of "homchain". In the second example, "homchain" outputs the following abort message (due to integer overflow, I think) which is ignored by the wrapper.

[Tech info: chain 24, addr 8, intgr 2.]
Reading a chain complex from 'bad.txt'...
Time used so far: 0.00 sec (0.000 min).
The ring of coefficients is the ring of integers.
Computing the homology of the chain complex...
Reducing D_1: 0 +ERROR: Number out of range (*).

[kcrisman]

comment:3

Possibly related: http://ask.sagemath.org/question/26550/make-ptest-best-way-to-fix-errors/

[jhpalmieri]

Upstream: Reported upstream. No feedback yet.

[jhpalmieri]

comment:4

Several comments: first, there is a new version of CHomP: see https://github.com/shaunharker/CHomP, but it does not fix these problems (at least the second one: the first is a problem with Sage's interface with CHomP, so that's our fault). Second, because of the second problem, we obviously need to catch this kind of error to avoid producing nonsensical answers. (It would also be nice to fix CHomP. I'll email the developer.)

[tscrim]

comment:5

We probably should fix #12607 now that optional spkgs are always tested, and we can do a version bump of CHomP (along with making it a new-style spkg).

[jhpalmieri]

comment:6

It would also be good to deal with #15202, but that's a more complicated project.

[fchapoton]

comment:7

The first problem is solved by the branch at #12607, that needs review.

[fchapoton]

Changed keywords from none to chomp

[fchapoton]

comment:9

And the second looks indeed like an overflow error in a product inside CHomP. This works

sage: S = simplicial_complexes.SumComplex(17, [0, 1, 4])
sage: S.homology(algorithm='chomp',verbose=True,base_ring=GF(103))

How should we catch the errors of CHomP ?

[fchapoton]

comment:10

I am taking care of the second problem also in ticket #12607

[fchapoton]

comment:12

Everything should be taken care of in #12607