Improve GEOSGeometry.tuple performance

[David Smith]

In ​PR, David has a proof of concept for improving the performance of the GEOSGeometry.tuple property for geometries with many coordinates, from 1.4x - 46x faster depending on the number of coordinates.

Addressing this will make it more practical to use this property when rearchitecting some other geometry construction paths in GeoDjango.

---

Using one of the fixtures (a Polygon, with 799 coords in one coordinate sequence) from the test suite.

Set up

In [1]: from django.contrib.gis.gdal import DataSource
  ...: from django.contrib.gis.geos import GEOSGeometry
  ...: 
  ...: source = DataSource("tests/gis_tests/data/counties/counties.shp")
  ...: geometry = GEOSGeometry(memoryview(source[0][1].geom.wkb))

Current

In [3]: %timeit geometry.tuple
3.4 ms ± 35.7 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Proposed

In [3]: %timeit geometry.tuple
73.2 μs ± 406 ns per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

A ~46x speedup.

This is a favourable example as it is a single 'large' geometry. The impact will vary depending upon the geometry. A collection of many geometries with each with a small number of coordinates, would perform differently. For example this item from the test suite has 9 sub-geometries with 21 points.

In [1]: import json
  ...: from pathlib import Path
  ...: from django.contrib.gis.geos import GEOSGeometry
  ...: 
  ...: data = json.loads(
  ...:     Path("tests/gis_tests/data/geometries.json").read_text()
  ...: )
  ...: geometry = GEOSGeometry(data["multilinestrings"][1]["wkt"])

Current

In [3]: %timeit geometry.tuple
306 μs ± 3.56 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

Proposed

In [2]: %timeit geometry.tuple
212 μs ± 3.05 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

A 1.4x improvement.