v0.13.2

Fixed

• Docker module incorrectly using the run_command function.
• Experiment passing in a cutoff run folder to the docker command.

v0.13.1

Fixed

• Reportables that implement render using the old name instead of qualified_name, causing
  unintended figure image overwrites.

v0.13.0

Added

• Check for get_params() functions that aren't returning lists.

Changed

• An aggregate stage that is not explicitly given a set of records now takes manager records minus
  the record containing the currently running aggregate stage.

Fixed

• Record make_copy adding the new record to the artifact manager twice.
• Reportables ToC in report not correctly using the qualified names when cached reportables found.
• LinePlotReporter not adding a legend when dictionaries provided for both x and y.
• Potential error when collecting metadata if manager run info doesn't have "status".

v0.12.0

Added

• Bash/zsh tab-completion via argcomplete. (This requires installing argcomplete outside of the
  environment and adding a line to your shell's rc file in order to use. You can run
  curifactory completion [--bash|--zsh] to add the line, or just run curifactory completion for
  instructions.)
• resolve option to Lazy outputs - this allows not automatically loading the object on an input
  to the stage, directly providing the lazy instance instead. This allows delaying the loading, or
  simply getting the path of the object to deal with in some other way (e.g. passing to an external
  command.)

Fixed

• experiment ls incorrectly handling curifactory configurations with experiment/param modules located
  in subdirectories.

v0.11.1

Fixed

• --names flag incorrectly checking existence of parameterset name.

v0.11.0

Added

• Curifactory submodules to top level import, so separately importing submodules is no longer necessary.

Changed

• Minimum python version to 3.9.
• Parameterset name to be ignored by hashing mechanism.

Fixed

• No longer using backported package importlib_resources that wasn't in the setup.

v0.10.1

Fixed

• Hash computation not correctly handling sub-dataclasses recursively.

v0.10.0

Added

• Metadata output for every cached artifact. Alongside every output cache file will be a [file]_metadata.json,
  containing information about the run that generated it, the parameters, and previous stages run in the same record.
• track parameter to cachers, indicating whether the output files should be copied into a full store run folder or not.
  (It is true by default.)
• Optional cacher prefixes, which replaces the first part of a cached filepath name (normally the experiment name) with the provided
  prefix. This allows cross-experiment caching (use with care!)
• Optional cacher subdir, which places output files into the specified subdirectory in the cache/run folder (allows better organization,
  e.g. Kedro's data engineering convention of 01_raw, 02_intermediate, etc.)
• Allowing exact path overrides to be used by a cacher, making it cleaner to use them on the fly/outside of stages.
• --version flag on the curifactory command.

Changed

• Full store cached files are now placed into an artifacts/ subdirectory of the run folder.
• PickleCacher's extension is now correctly set to .pkl (we aren't actually running gzip on it.)
• Full store runs no longer call a cacher's save function a second time with a new path, instead relying
  on Record's path tracking to simply copy the cached files into the full store folder at the end of a stage.
• Cachers' path mechanism - rather than expecting a cacher's set_path to be called beforehand, save and load
  should call the cacher's get_path().
• The default cachers' save() functions return the path that was saved to.
• --name flag to --prefix to make it more consistent to caching terminology.

Fixed

• Reportable names doubling when loading from cache.
• Silent execution when no parametersets provided or a requested parameterset name wasn't found, (now errors and exits.)

Breaking changes notes

• Cacheable.set_path no longer exists, any custom cachers that override it should remove this function.
• Any self.path references in custom cachers save and load should replace it with self.get_path([optional_suffix]), and do this for any paths it writes out (rather than calling get_path, modifying the resulting path, and writing to that modified path)
• Any previous cached files from the PickleCacher will no longer be used as the extension has changed

v0.9.3

Fixed

• Lack of proper html escaping of args dump in output reports.

v0.9.2

Fixed

• String hash representation not recursively getting a string hash
  representation from any parameter sub-dataclasses.