MNT: Add a cache to the Transformer creation in transform_points

[greglucas]

This adds a cache mechanism to creating Transformer objects, which
can be slow when called frequently. Additionally, this requires
adding attributes to the Mercator creation as the x/y limits are
created in the initializer which uses the cache/hash function.

closes #1915

Rationale

Implications

[Bob131]

Yep, this seems to solve the issue I was hitting.

Here's the benchmark I was using, btw. I wrote it thinking the problem was due to transforming polygon points, but tweaking things it seems the important bits are the grid line labels.

import numpy as np
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation
from matplotlib.patches import Polygon

import cartopy.crs as ccrs


class PolygonAnimation:
    params = [
        ( 8, 128),
        (50, 200)
    ]
    param_names = ['n_points', 'n_frames']

    def setup(self, n_points, n_frames):
        self.polygon_points = \
            np.array([[np.cos(x), np.sin(x)]
                      for x in np.linspace(0, 2 * np.pi, n_points)]) \
                          * 180 / np.pi
        self.proj = ccrs.PlateCarree()
        fig, ax = plt.subplots(subplot_kw=dict(projection=self.proj))
        ax.set_global()
        ax.gridlines(crs=self.proj, draw_labels={'left': 'y'}, x_inline=True)
        self.figure = fig
        self.polygon = ax.add_patch(Polygon(self.polygon_points,
                                            transform=self.proj.as_geodetic()))
        self.animation = FuncAnimation(fig, self.update,
                                       frames=np.linspace(0, 360, n_frames),
                                       blit=True)

    def update(self, frame):
        self.polygon.set_xy(self.polygon_points + (frame, 0))
        return self.polygon,

    def time_draw(self, n_points, n_frames):
        self.animation.save('/tmp/cartopy-bench.gif')

➜  cartopy git:(5540d799) ✗  asv --config benchmarks/asv.conf.json run -b polygon_animation 'HEAD~2..HEAD'
· Creating environments
· Discovering benchmarks
·· Uninstalling from virtualenv-py3.10-fiona-matplotlib-numpy-pykdtree-pyproj-scipy
·· Installing 5540d799 into virtualenv-py3.10-fiona-matplotlib-numpy-pykdtree-pyproj-scipy.
· Running 2 total benchmarks (2 commits * 1 environments * 1 benchmarks)
[  0.00%] · For cartopy commit 5540d799:
[  0.00%] ·· Benchmarking virtualenv-py3.10-fiona-matplotlib-numpy-pykdtree-pyproj-scipy
[ 25.00%] ··· Running (polygon_animation.PolygonAnimation.time_draw--).
[ 50.00%] ··· polygon_animation.PolygonAnimation.time_draw                                                                          ok
[ 50.00%] ··· ========== ============ ============
              --                  n_frames        
              ---------- -------------------------
               n_points       50          200     
              ========== ============ ============
                  8       2.46±0.01s   9.10±0.01s 
                 128      2.74±0.09s   10.2±0.04s 
              ========== ============ ============

[ 50.00%] · For cartopy commit 313f02f3 <master>:
[ 50.00%] ·· Building for virtualenv-py3.10-fiona-matplotlib-numpy-pykdtree-pyproj-scipy.
[ 50.00%] ·· Benchmarking virtualenv-py3.10-fiona-matplotlib-numpy-pykdtree-pyproj-scipy
[ 75.00%] ··· Running (polygon_animation.PolygonAnimation.time_draw--).
[100.00%] ··· polygon_animation.PolygonAnimation.time_draw                                                                          ok
[100.00%] ··· ========== ============ ============
              --                  n_frames        
              ---------- -------------------------
               n_points       50          200     
              ========== ============ ============
                  8       7.13±0.03s   28.6±0.06s 
                 128       7.39±0s     29.5±0.06s 
              ========== ============ ============

Decided to hold off.

[dopplershift]

@greglucas Could we instead make our own get_transformer_from_crs (or something) which uses lru_cache that we call? I’d really like to avoid monkeypatching pyproj, which can impact people who are using pyproj outside of cartopy within the same program.

[dopplershift]

To be clear, something like:

@lru_cache
def get_transformer_from_crs(crs):
    return Transformer.from_crs(crs)

[greglucas]

Yep, good suggestion. Should be updated on both branches now.