Fix #70 by bypassing lazy loading of datasets (#71)

By default, rioxarray lazyily loads datasets. On Windows, this causes
a PermissionError when the temporary directory containing the downloaded
files is deleted, as it is still open in rioxarray. To avoid that issue,
we now explicitly load the dataset into memory, allowing tempfiles to be
removed. This will potentially increase memory usage in some use cases, but
given the size limitations of datasets downloaded from Earth Engine, that
should never be a practical problem.

wxee/utils.py

@@ -158,7 +158,12 @@ def _dataarray_from_file(file: str, masked: bool, nodata: int) -> xr.DataArray:
 
  The file name must follow the format "{dimension}.{coordinate}.{variable}.{extension}".
  """
- da = rioxarray.open_rasterio(file)
+ with rioxarray.open_rasterio(file) as da:
+ # Load fully into memory rather than reading lazily from disk. This is needed to allow reading from tempfiles
+ # that will be deleted after the function returns. See https://github.com/corteva/rioxarray/issues/485 and
+ # https://github.com/aazuspan/wxee/issues/70.
+ da.load()
+
  dim, coord, var = _parse_filename(file)
 
  da = da.expand_dims({dim: [coord]}).rename(var).squeeze("band").drop_vars("band")