Descarteslabs

- Workflows client no longer validates `processing_level` parameter values, as these have been enhanced to support new products
and can only be validated server side.
- Catalog V2 bands now support the `vendor_band_name` field (known as `name_vendor` in Metadata/Catalog V1).
- Scenes support for masking in version 1.8.1 had some regressions which have been fixed. For this reason, version
1.8.1 has been pulled from PyPI.
- New task groups now default to a `maximum_concurrency` value of 5, rather than the previous 500. This avoids the common
problem of deploying a task group with newly developed code, and having it scale up and turning small problems into
big problems! You may still set values as large as 500.
- The Tasks client now provides an `update_group()` method which can be used to update many properties of an existing
task group, including but not limited to `name`, `image`, `minimum_concurrency`, and `maximum_concurrency`.
- Improved testing across several sub-packages.
- Various documentation fixes.

General

- Added a new `common.dltile` library that performs geospatial transforms and tiling operations.
- Upgraded various dependencies: `requests[security]>=2.25.1,<3`,`six>=1.15.0`,`blosc==1.10.2`,` mercantile>=1.1.3`,`Pillow>=8.1.1`,`protobuf>=3.14.0,<4`,`shapely>=1.7.1,<2`,`tqdm>=4.32.1`,`traitlets>=4.3.3,<6;python_version<'3.7'`,`traitlets==5.0.5,<6;python_version>='3.7'`,`markdown2>=2.4.0,<3`,`responses==0.12.1`,`freezegun==0.3.12`,`imagecodecs>=2020.5.30;python_version<'3.7'`,`imagecodecs>=2021.5.20;python_version>='3.7'`,`tifffile==2020.9.3;python_version<'3.7'`,`tifffile==2021.4.8;python_version>='3.7'`

Discover (alpha) - Added

- Added an **alpha** Discover client. Discover allows users to organize and share assets with other users. As an alpha release, we reserve the right to modify the Discover client API without any guarantees about backwards compatibility. See the [Discover API documentation](https://docs.descarteslabs.com/descarteslabs/discover/readme.html) for more details.

Metadata/Catalog V1 - Changed

- **breaking** Image (Scene) metadata now accepts and returns the `bucket` and `directory` fields as lists of strings, of a length equal to that
of the `files` fields. This allows the file assets making up an image to live in different locations. When creating new images,
a simple string can still be provided for these fields. It will automatically be converted to a list of (duplicated) strings as
necessary. As most users will never interact with these fields, the change should not affect user code.

Metadata/Catalog V1/Catalog V2 - Changed

- `derived_params` field for Image (scene) metadata now supported for product-specific service-implemented "native derived bands" which may
only be created for core products.

Scenes - Changed

- Scenes now uses the client-side `dltile` library to make DLTiles. This improves performance when creating a large number of DLTile objects.
- Scenes DLTile `from_shape` now has a parameter to return tile keys only instead of full tile objects. Usage details can be found [in the docs](https://docs.descarteslabs.com/descarteslabs/scenes/docs/geocontext.html#descarteslabs.scenes.geocontext.DLTile.from_shape).
- Scenes DLTile now has new methods: `iter_from_shape` that takes the same arguments as `from_shape` but returns an iterator ([from_shape docs](https://docs.descarteslabs.com/descarteslabs/scenes/docs/geocontext.html#descarteslabs.scenes.geocontext.DLTile.iter_from_shape)), `subtile` that adds the ability to subdivide tiles ([subtile docs](https://docs.descarteslabs.com/descarteslabs/scenes/docs/geocontext.html#descarteslabs.scenes.geocontext.DLTile.subtile)), and `rowcol_to_latlon` and `latlon_to_rowcol` which converts pixel coordinates to spatial coordinates and vice versa ([rowcol_to_latlon docs](https://docs.descarteslabs.com/descarteslabs/scenes/docs/geocontext.html#descarteslabs.scenes.geocontext.DLTile.rowcol_to_latlon) and [latlon_to_rowcol docs](https://docs.descarteslabs.com/descarteslabs/scenes/docs/geocontext.html#descarteslabs.scenes.geocontext.DLTile.latlon_to_rowcol)).
- Scenes DLTile now has a new parameter `tile_extent` which is the total size of the tile in pixels including padding. Usage details can be found [in the docs](https://docs.descarteslabs.com/descarteslabs/scenes/docs/geocontext.html#descarteslabs.scenes.geocontext.DLTile.tile_extent).
- **breaking** Removed the dependence on `Raster` for tiling. The `raster_client` parameter has been removed from the `from_latlon`, `from_key`, `from_shape`, and `assign`DLTile methods.
- Tiling using `from_shape` may return a different number of tiles compared to previous versions under certain conditions. These tiles are usually found in overlapping areas between UTM zones and should not affect the overall coverage.
- DLTile geospatial transformations are guaranteed to be within eight decimal points of the past implementation.
- DLTile errors now come from the `dltile` library and error messages should now be more informative.
- When specifying output bounds in a spatial reference system different from the underlying raster, a densified representation of the bounding box is used internally to ensure that the returned image fully covers the bounds. For certain methods (like `mosaic`) this may change the returned image dimensions, depending on the SRSs involved.
- **breaking** As with the Metadata v1 client changes, the `bucket` and `directory` fields of the Scene properties are now multi-valued lists.
- Scenes does not support writing GeoTiffs to file-like objects. Non-JPEG GeoTiffs are always uncompressed.

Raster - Changed
- `dltiles_from_shape`, `dltiles_from_latlon`, and `dltile` have been removed. **It is
strongly recommended to test any existing code which uses the Raster API when upgrading to this
release.**
- Fully masked arrays are now supported and are the default. Usage details can be found [in the docs](https://docs.descarteslabs.com/descarteslabs/client/services/raster/readme.html#descarteslabs.client.services.raster.Raster.ndarray)
- Added support to draw progress bar. Usage details can be found [in the docs](https://docs.stage.descarteslabs.com/descarteslabs/client/services/raster/readme.html).
- The signature and return value of `Raster.raster()` have changed. The `save=` parameter has been removed as the resulting download is always saved
to disk, to a file named by the `outfile_basename=` parameter. The method returns a tuple containing the name of the resulting file and the metadata
for the retrieval, which is now an ordinary Python dictionary.
- As with Scenes, when specifying output bounds in a spatial reference system different from the underlying raster, a densified representation of the bounding box is used internally to ensure that the returned image fully covers the bounds. For certain methods (like `mosaic`) this may change the returned image dimensions, depending on the SRSs involved.

(Note: v1.8.0 was an internal-only release)

General
- Upgraded various dependencies: `blosc==1.10.2`, `cachetools>=3.1.1`, `grpcio>=1.35.0,<2`, `ipyleaflet>=0.13.3,<1`, `protobuf>=3.14.0,<4`, `pyarrow>=3.0.0`, `pytz>=2021.1`
- Upgraded from using Travis to GitHub Actions for CI.

Catalog
- Added support for the `physical_range` property on `SpectralBand` and `MicrowaveBand`.

Workflows (channel `v0-19`) - Added
- Workflows sharing. Support has been added to manage sharing of `Workflow` objects with other authorized users. The `public` option for publishing workflows
has been removed now that `Workflow.add_public_reader()` provides the equivalent capability. See the [Workflows Guide](https://docs.descarteslabs.com/guides/workflows/sharing.html#sharing-workflows).
- Lots of improvements to [API documentation](https://docs.descarteslabs.com/descarteslabs/workflows/readme.html) and the [Workflows Guide](https://docs.descarteslabs.com/guides/workflows.html).

Workflows - Fixed
- Allow constructing `Float` instances from literal python integers.

This release was withdrawn due to a backwards compatility issue which arose.

Fixes a few buglets which slipped through. This release continues to use the workflows channel `v0-18`.

Workflows - Fixed
- Fixed a problem with the defaulting of the visual options when generating tile URLs, making it possible
to toggle the checkerboard option on a layer and see the difference.
- Support `axis=list(...)` for `Image`.
- Corrected the results of doing arithmetic on two widgets (e.g. adding two `IntSlider`s together should yield` an `Int`).
- For single-band imagery `VizOption` will accept a single two-tuple for the `scales=` argument.

Python Version Support
- Python 3.6 is now deprecated, and support will be removed in the next version.

Catalog
- Added support to Bands for new processing levels and processing step specifications
to support Landsat Collection 2.

Workflows (channel `v0-18`) - Added
- **`wf.widgets` lets you quickly explore data interactively.** Add widgets anywhere in your code just like normal values, and the widgets will display automatically when you call `.visualize`.
- **View shared Workflows and XYZs in GIS applications using WMTS.** Get the URL with `wf.wmts_url()`, `XYZ.wmts_url()`, `Workflow.wmts_url()`.
- Create publicly-accessible tiles and WMTS endpoints with `wf.XYZ(..., public=True)`. Anyone with the URL (which is a cryptographically random ID) can view the data, no login required. Set `days_to_expiration` to control how long the URL lasts.
- `wf.XYZ.list()` to iterate through all XYZ objects you've created, and `XYZ.delete` to delete them.
- Set default vizualization options (scales, colormap, bands, etc.) in `.publish` or `wf.XYZ` with `wf.VizOption`. These `viz_options` are used when displaying the published object in a GIS application, or with `wf.flows`.
- `ImageCollection.visualize()`: **display ImageCollections on `wf.map`**, and select the reduction operation (mean, median, mosaic, etc.) interactively
- `Image.reduction()` and `ImageCollection.reduction()` (like `ic.reduction("median", axis="images")`) to reduce an Image/ImageCollection with an operation provided by name
- `wf.map.controls` is accessible (you had to do `wf.map.map.controls` before)
- Access the parameters used in a Job with `Job.arguments` and `Job.geoctx`.

Workflows - Fixed
- Errors like `In 'or': : operand type(s) all returned NotImplemented from __array_ufunc__` when using the bitwise-or operator `|` are resolved.
- Errors when using computed values in the `wf.Datetime` constructor (like `wf.Datetime(wf.Int(2019) + 1)`) are resolved.
- `wf.Timedelta` can be constructed from floats, and supports all binary operations that Python does (support for `/, //, %, *` added)
- In `.rename_bands`, prohibit renaming a band to a name that already exists in the Image/ImageCollection. Previously, this would succeed, but cause downstream errors.
- `.bandinfo.get("bandname", {})` now works---before, providing `{}` would fail with a TypeError
- Indexing an `Any` object (like `wf.Any({"foo": 1})["foo"]`) behaves correctly
- `wf.Datetime`s constructed from strings containing timezone information are handled correctly

Workflows - Changed
- `.mask(new_mask)` ignores masked pixels in `new_mask`. Previously, masked pixels in `new_mask` were considered True, not False. Note that this is opposite of NumPy's behavior.
- If you `.publish` an object that depends on `wf.parameter`s or `wf.widgets`, it's automatically converted into a `wf.Function`.
- **breaking** `.compute` and `.inspect` no longer accept extra arguments that aren't required for the computation. If the object doesn't depend on any `wf.parameter`s or `wf.widgets`, passing extra keyword arguments will raise an error. Similarly, *not* providing keyword arguments for all parameters the object depends on will raise an error.
- **breaking** The `wf.XYZ` interface has changed; construct an XYZ with `wf.XYZ(...)` instead of `wf.XYZ.build(...).save()`
- Set `days_to_expiration` on `XYZ` objects. After this many days, the object is deleted.
- `Job` metadata is deleted after 10 days; `wf.Job.get(...)` on a job ID more than 10 days old will fail. Note that Job results have always been deleted after 10 days; now the metadata expires as well.
- `wf.Function` has better support for named arguments. Now, `f = wf.Function[{'x': wf.Int, 'y': wf.Str}, wf.Int]` requires two arguments `x` and `y`, and they can be given positionally (`f(1, "hi")`), by name in any order(`f(x=1, y="hi")` or `f(y="hi", x=1)`), or both (`f(1, y="hi")`). `wf.Function.from_callable` will generate a Function with the same names as the Python function you decorate or pass in. Therefore, when using `wf.publish` as a decorator, the published Function will automatically have the same argument names as your Python function.