PERF: Avoid creating GDALEnv for affine transformations.

[groutr]

TransformerBase creates a GDALEnv every time its context is entered. For affine transformations, this is a huge performance hit for nothing.

ref: #2753

Before PR:

In [9]: %%prun
   ...: for lo, la in zip(lons, lats):
   ...:     next(sample.sample_gen(A, [(lo, la)]))
   ...:
         5480066 function calls (5470066 primitive calls) in 9.676 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    10000    3.796    0.000    4.008    0.000 {method 'start' of 'rasterio._env.GDALEnv' objects}
    10000    1.250    0.000    1.874    0.000 {method 'read' of 'rasterio._io.DatasetReaderBase' objects}
    20000    0.651    0.000    9.596    0.000 sample.py:20(sample_gen)

After PR:

In [8]: %%prun
   ...: for lo, la in zip(lons, lats):
   ...:     next(sample.sample_gen(A, [(lo, la)]))
   ...:
         1910066 function calls (1900066 primitive calls) in 2.968 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    10000    0.760    0.000    1.286    0.000 {method 'read' of 'rasterio._io.DatasetReaderBase' objects}
    30000    0.275    0.000    0.362    0.000 stride_tricks.py:340(_broadcast_to)
    20000    0.127    0.000    2.916    0.000 sample.py:45(sample_gen)

[groutr]

Initially, I thought the best solution was to override __enter__ in AffineTransformer. However, TransformerBase should not be responsible for creating a GDALEnv for its all its subclasses. It should be incumbent upon the class inheriting TransformerBase to know if it needs to create a GDALEnv.

[snowman2]

@groutr thanks for this 👍. Thoughts about creating a GDALTranformerBase class for transformers that require a GDAL environment?

[snowman2]

Thanks @groutr 👍

[sgillies]

Thanks @groutr !

[sgillies]

I'm going to cherry pick this one for 1.3.6.