Rasterio

------------------

Bug fixes:

- Secrets kept in GDAL config options could have been leaked via the Python
logger. AWS keys and tokens have always been redacted, but other options like
GDAL_HTTP_USERPWD were not. Logging of GDAL config options has been removed.
- Use of Rasterio with Python threads has not been well tested in
previous versions and a bug (986) involving shared GDAL environment
state was found. Rasterio now uses thread local variables to track
GDAL environment state and uses the appropriate GDAL API functions to
isolate the environments of child threads from each other while
permitting inheritance from the main Python thread (993, 996, 997).
Tests using both `ThreadPoolExecutor` and `ProcessPoolExecutor` have
been added to guard against regresion.

------------------

Roses are red.
Irish say "feck."
Calling conventions
are a pain in the neck

Bug fixes:

- Use __stdcall for GDAL callbacks on Windows (935, 951).
- Tear down Env more carefully (968, 969), taking particular care not to
reinstate default options when a child environment is exited.
- Always close open file handles when exiting context managers returned by
`rasterio.open()` when file objects are given as arguments (973).
- Properly append, not extend, shim module to extension list (975).

------------------

Bug fixes:

- Order of input datasets is preserved by rio-calc (947, 965).

New features:

- `MemoryFile()` can now take a file object parameter or bytes. The file should
be opened in binary mode and its contents are read immediately into bytes.
This form of argument is for convenience only; the provided file is not
accessed in any other way.
- A `ZipMemoryFile` class derived from `MemoryFile` has been added. It is
created from a sequence of bytes representing a zip file. Its `open()` method
can take a path, allowing access to files within the zip file.
- For convenience of users, Rasterio's `open()` function now takes Python file
objects as well as filenames and URLs (964). These file objects have their
contents read into `MemoryFile` instances within the `open()` call, so this
feature is equivalent to calling `MemoryFile(fp).open()`.

Refactoring:

- GDAL/OGR function definitions consolidated in gdal.pxi (949).
- Shim APIs have been created to provide support for GDAL 2.0 and 2.1 features
that may be absent in a system's GDAL installation.

Packaging:

- Shim API C modules are included in source distributions.

------------------

- New binary wheels using version 1.2.0 of sgillies/frs-wheel-builds. See
https://github.com/sgillies/frs-wheel-builds/blob/master/CHANGES.txt.

------------------

New features:

- Geographic bounds designated by a `--geographic` option may now be
passed to `rio clip` (932).
- The `rio edit-info` command's new `--unset-nodata` option can be used to
delete a dataset's nodata value (510).

Bug fixes:

- Passing a `--co nodata=null` option to CLI commands will create
datasets without any nodata value set. The `--co` option name follows
GDAL's command line utilities. A `--profile` alias for this option has
also been added.

------------------

New features:

- The `DatasetBase` class is now surfaced in `rasterio.io`. This is not too
promote the practice of subclassing `DatasetBase` but to help users that
would like to use type annotations (925) or make `isinstance()` assertions.
- Ground control points may be read from and written to datasets (when the
format permits) and may be used when reprojecting arrays with `reproject()`
or on the command line with rio-warp (203, 916). A new rio-gcps command
has been added, it can print a dataset's GCPs as GeoJSON Feature sequences
or FeatureCollections.
- A `BytesIO` like `MemoryFile` class has been added for an improved Python
interface to GDAL's in-memory filesystem (899, 925).
- Error handlers are pushed to GDAL on entering a GDAL environment and popped
when exiting (876, 902, 907).
- The GIL is released when calling GDAL's ChunkAndWarp() (912).

Bug fixes:

- transform_geom() would fail when encountering geometries with z coordinate
values, but now handles them properly (922, 923).
- show_hist() failed when nans were present in data (913) and this has been
fixed with the use of nanmin() and nanmax() in 914.