Fix: duplicates with collinear points sometimes not found with shared_coords=False

[theroggy]

In the following case 2 linestrings were not correctly identified as duplicate with with shared_coords=False:
- one of them has extra point(s) that disappear with simplify(0).
- there is another line combination in the dataset that creates junctions for
shared_coords=False.

Practical case where this showed up. The combination of the above ingredients triggered the bug like this:

• At the top right the smaller polygon touches the outer boundary of the largest polygon, so both those polygons need to be split to create the relevant topologies -> this leads to junctions being created.
• The small polygon at the bottom left fills up an "island" in the large polygon contained an extra collinear point compared to the linearring describing the island.
  

[theroggy]

Reference orthoseg/orthoseg#134

[mattijn]

I stared at this for a while, I ran a debugger and I tried to understand why the fix in your PR solves the issue, you basically go from

lines_split.append(remove_collinear_points([np.array(linestring.coords)]))

to

lines_split.append([remove_collinear_points(np.array(linestring.coords))])

While I cannot really explain why this results in another linestring being detected as a duplicate; it resolves the issue that you have and the PR includes a nice test. So thank you!

[theroggy]

Yes, it has taken me quite some hours going from the real-world problem to that specific code change, and the only thing to show for it are 2 square brackets being moved a few characters :-).

The difference is in what remove_collinear_points receives as input:

• before something like this: [ [ [0, 1], [0, 2], [0, 3] ] ]
• versus now it will get this: [ [0, 1], [0, 2], [0, 3] ]

One of the first lines of remove_collinear_points is len(line). For the first array will always be 1... so there are never collinear points to be removed :-(....

    # If only 2 points, no use checking
    if len(line) <= 2:
        return line

[mattijn]

Thanks for the explanation! I just released a new version of topojson, see https://pypi.org/project/topojson/1.7/

[theroggy]

By reading the change notes I noticed that the title of the PR was quite off... so I updated it.