fix int type usage

[rathann]

Fixes part of #1362 .

Changes made in this Pull Request:

• int type usage fixes to accommodate 32 bit architectures
• slight precision relaxation in analysis/test_psa.py:_BaseHausdorffDistance.test_symmetry comparison

PR Checklist

• Tests?
• Docs?
• CHANGELOG updated?
• Issue raised/referenced?

[rathann]

I'm also wondering if MDAnalysis/core/groups.py:2739 should be changed to np.intp as well.

[kain88-de]

@rathann do you know of a way of testing if this is robust to run on 32bit using a 64bit travis machine?

[orbeckst]

Maybe follow the recipe in travis-ci/travis-ci#5770 and use a docker image with a 32 bit chroot for testing on travis?

[orbeckst]

And obspy/obspy#1580 was apparently a successful attempt to Enable 32bit linux on travis CI without docker, just using compat 32 bit libs and the 32-bit miniconda. And they use Python 2.7 and 3.5.

[orbeckst]

According to numpy/numpy#4384 (comment), np.intp is the recommended dtype for indexing

... use np.intp as dtype whenever things have to do with indexing or are logically related to indexing/array sizes. This is the natural dtype for it and it will normally also be the fastest one.

This seems to imply that we ought to be using it everywhere instead of np.int64 — I don't know if @rathann 's PR here is already replacing all occurrences.

[rathann]

@kain88-de

@rathann do you know of a way of testing if this is robust to run on 32bit using a 64bit travis machine?

I don't know travis, but on Fedora we use 32bit chroot or container to do 32bit builds on a 64bit machine via mock: mock -r fedora-rawhide-i386.

[rathann]

@orbeckst

This seems to imply that we ought to be using it everywhere instead of np.int64 — I don't know if @rathann 's PR here is already replacing all occurrences.

Not everywhere, only for array indexing. And no, I'm not replacing all. Just those that made the tests fail. I'm not familiar with the codebase.

[richardjgowers]

I've made a new PR which extends this a little with some tests. I've changed all the arrays we use for indexing to use np.intp. We don't need to change all integer array dtypes to intp as some integer arrays are data, eg resids, rather than indices.