• !79 [conda] Standardize conda recipe: According to https://www.idiap.ch/software/bob/docs/bob/bob.devtools/master/templates.html#conda-recipe which was updated in bob/bob.devtools!185
• !80 Updates to Sphinx 3.2.x: I had to update doc/conf.py since an old flag we were using autodoc_default_flags is not supported anymore. The new one is autodoc_default_options (https://www.sphinx-doc.org/en/master/usage/extensions/autodoc.html#confval-autodoc_default_options). After activating it, we observed a few more issues related to double-documentation of attributes in some classes. @amohammadi: there are some take-aways for Bob projects here as well.
• !81 Workaround multiprocess changes in macOS Python 3.8: The default implementation for the multiprocess changed from fork to spawn on macOS with Python 3.8. This merge request forces the use of fork for the tests on macOS. The runner will anyway use Linux so taking more time to solve this particular issue would not make much sense.
• !82 Add pyproject.toml: This follows bob packages new standard

• !78 Make all test databases root folder point to a valid location: This will allow the docker tests to more easily setup and cleanup its content. While the folder is not used as these databases are generative only, docker will create it anyway as mounting a non-exisiting folder will trigger its creation with the daemon ownership applied.

• !77 Pre commit cleanup: This merge request updates the pre-commit setup to match the one from beat/beat.editor> in terms of list of hooks to apply and provide the configuration of the new hooks as well. Then all files have been passed through the full setup of hooks. Commits have been split to have a grouped approach of the files having been cleaned.
• !75 Handle error the same way for all algorithm types on execution: Currently sequential and autonomous algorithms are not treating errors happening in the process method the same way. This result in missing information coming back up through beat/beat.core> This patch fixes this and also ensure that the outputs are properly closed. Fixes beat.core#98

• !75 Handle error the same way for all algorithm types on execution: Currently sequential and autonomous algorithms are not treating errors happening in the process method the same way. This result in missing information coming back up through beat/beat.core> This patch fixes this and also ensure that the outputs are properly closed. Fixes beat.core#98

• !73 Fixes a few warnings and depracations
• !74 Add helper property to get the environment configured for the database. Part of beat.core#64

• !72 Update test execution environment to Python for tests: This patch moves all the execution environment definition to the new one provided by beat/beat.env.builder>

• !71 Implement multiprocessing support for CachedDataSource. Fixes #32

• !69 Logger cleanup: improvements on logging as well as code cleanup making use of python features rather than exceptions
• !70 Improve file name handling and removal in CachedDataSink for better diagnostic and error handling.

• !68 Fix wrong instance variable name in error message when not enough data is written with a CachedDataSink

• !66 Improve CachedDataSink error messages
• eat/beat.backend.python!67 Fix loop and database executor wait when an early exit happens

• !61 Improve synchronized output handling
• !62 Improve AstractStorage error message in constructor
• !63 Removed Biometrics evaluation from BEAT definition in documentation
• !64 Fix hash generated by storage classes to follow the same logic as used in beat/beat.web>
• !65 Fix license file handling for conda

• !59 Do not systematically decode incoming zmq parts as binary data can be expected.
• !60 Fix write end_data_index None handling in LoopMessageHandler
• !61 Improve synchronized output handling by adding the name of the processor output written as parameter. This allows to properly write the output of the evaluator.

• !59 Do not systematically decode incoming zmq parts as the binary content sent may need to stay as is for further processing

• !40 Add BEAT framework classifier: Fixes #21
• !42 Use format for string formatting. The use of the new "fast" string notation not yet possible as production servers are on 3.5 and some environments 2.7.
• !41 Template protocol: This merge request implements the support for database V2 with protocol details moved into the ProtocolTemplate object.
• !43 Cleanup handling of database schema version: This way beat/beat.core> doesn't depend on updating the is_v1 flag.
• !44 Fix protocol view definition generation: The parameter field was missing.
• !45 Clean up test metadata for schema updates: The parameter definitions are more strict as of beat.core!65.
• !46 Improve asset information: This merge request moves some information in the Storage class so that it's easier to re-use.
• !47 Create a base class for Storage and CodeStorage: They share some logic and use so avoid code duplication and hidden missing pieces.
• !48 Assert cleanup: This merge request cleans the code base from its assert usage.
• !49 Add dummy documentation to assets in prefix: This will allow to do more tests in for example beat/beat.editor>.
• !50 Add *.rst to list of files to include from test prefix.
• !51 Handle duplicate key: This merge request implements handling of JSON file with duplicated keys.
• !52 [doc] fixed external and internal links: This merge request is fixing the link issues in the documentation of beat.
• !54 Implement loop output.
• !55 New loop user types: implement the support for both autonomous and sequential loop user types.
• !56 Fix loop dataformat export and properly export all dataformats used be it in input, outputs or loop communication.
• !57 Implement loop reading input sequentially.
• !58 Change loop parts name to use xxx_loop_processor and xxx_loop_evaluator which are clearer for the role of each.

• !37 Allow package to be tested outside of source checkout
• !38 Re-enables nitpicky checks on the documentation. Fix errors caused by not having that flag ON.
• !39 Move license to BSD 3 Clause

• !36 Fix handling of string when packing

• !26 Refactor execution: makes the "core" parts and the "execution" parts more cleanly separated.
• !27 Soft loop: Implement soft loop
• !29 Give loop full access to input content in prepare
• !31 Add parameter to specify configuration section to load for databases_provider
• !32 Solve an issue relative to CI paths for artifacts and caches that must be relative
• !33 New documentation is now available

• !35 Temporarily remove framework trove classifier to unblock pypi upload

• !23 Fix an issue where numpy guessed type mismatched with the given dtype. For example, int64 VS uint64.
• !24 Fix handling of scalar value when it's considered zero by python
• !25 Fix lazy loading of the algorithm's runner. It should only be created once.

• !21 Prepare algorithm on execution: The execute script was not calling prepare.
• !22 Implement warning if prepare is present with api_version 1 property for algorithm

New V2 API for data loading. This new version will allow algorithm to more efficiently load data to GPU when using one. This release implements three type of algorithms

• legacy (which is temporary for old algorithms to be ported)
• sequential which works like legacy meaning that each input is clocked
• autonomous which means that the algorithm is responsible for the data loading and execution