ENH: Added length and perimeter implementation

[JoelJaeschke]

No description provided.

[jorisvandenbossche]

Thanks!

What's broken about the perimeter?

[JoelJaeschke]

I am still getting used to thinking in spherical geometries. The test case I wrote for the perimeter function used a polygon that was rectangular in my head, so I expected its perimeter to be $4\pi / 180°$. However, not only this assumption was wrong (as angular distances between two points with the same latitude at higher latitudes are smaller than those near the equator), but I also did not take into account that all points at the poles are actually coincident, so my result was $3\pi/180°$ which I assumed to be some sort of bug.

Sorry for the confusion, everything was fine from the start!

[jorisvandenbossche]

Looks good!

Do you just want to add it to api.rst as well?

[benbovy]

Thanks @JoelJaeschke!

One question about the API: should we follow BigQuery's API or shapely's API? Or a mix of both? Some options:

1. Expose separate length and perimeter functions, with value zero returned for "incompatible" geography (like BigQuery, currently implemented in this PR)
2. Expose a unique length function (like Shapely)
3. Expose length and perimeter functions where length also returns the perimeter for an input Polygon.

[benbovy]

One question about the API: should we follow BigQuery's API or shapely's API? Or a mix of both?

Any thoughts on this @jorisvandenbossche?

[jorisvandenbossche]

I think I like option 3: we provide a length that can handle both, and then still a perimeter for discoverability for people familiar with other systems like bigquery or postgis (and also perimeter is a more descriptive name in case of polygons).

(we should maybe consider adding a perimeter() in shapely as well)

I am also fine with leaving that for a follow-up, though, either way is fine.

[JoelJaeschke]

I would be happy to change the PR accordingly if option 3 is the way forward!

[benbovy]

Let's choose the easy way and leave length and perimeter as it is in this PR for now.

Thanks @JoelJaeschke !