bug fixes in lib.distances

[zemanj]

Fixes (partially) #2046. This is the fourth of a series of related PRs following PRs #2048, #2062, and #2070.

Changes made in this Pull Request:

• Ensured that none of the functions in MDAnalysis.lib.distances modifies its input coordinate arrays in-place. This fixes a bug I introduced in PR Simplified code in lib.distances a bit #2048 (see updated issue MDAnalysis.lib.distances needs rework #2046).
• Allowed for empty coordinate array input (shape=(0,3)) in all lib.distances functions.
• Fixed failures / wrong distance output of *capped_distance() in several corner cases.
• Added corresponding tests ensuring non-modified input, correct functionality with empty input, and correct return types.
• Improved return type specifications in the docs.

PR Checklist

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

[codecov]

Codecov Report

Merging #2083 into develop will increase coverage by 0.04%.
The diff coverage is 98.63%.

@@             Coverage Diff             @@
##           develop    #2083      +/-   ##
===========================================
+ Coverage    89.37%   89.41%   +0.04%     
===========================================
  Files          159      159              
  Lines        18712    18731      +19     
  Branches      2684     2696      +12     
===========================================
+ Hits         16724    16749      +25     
+ Misses        1384     1381       -3     
+ Partials       604      601       -3

Impacted Files	Coverage Δ
package/MDAnalysis/lib/distances.py	97.46% <98.63%> (+2.19%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8918d42...7bd977e. Read the comment docs.

[zemanj]

This is not ready for review. Since I have to add tests for empty inputs to satisfy the diff coverage, I'll do that for all functions right away.

[richardjgowers]

@zemanj how/where were coordinates being modified in place?

[richardjgowers]

Looks good, few comments would be nice

[zemanj]

@richardjgowers

@zemanj how/where were coordinates being modified in place?

This happened when I introduced the new @check_coords() decorator. For some functions, I falsely assumed that they either make copies of (or only read from) the input arrays internally. I fixed the problem by removing the enforce_copy=False decorator keyword in these cases. In order to guarantee unmodified input arrays, I added the class TestInputUnchanged to lib/test_distances.py.

[zemanj]

@richardjgowers I added the requested comments. Moreover, I removed the obsolete dtype conversion in the return statements of capped_distance() and self_capped_distance(). This has the advantage that the return type is now also tested for the individual methods (not really required IMHO, but nice to have).

[zemanj]

@richardjgowers Do you want to take another look or shall I merge?

[zemanj]

@richardjgowers Thank you for reviewing!