v1.2.2

small patch fix

the serial server had a bug where the appropriate data was only returned on a subset of possible end conditions

v1.2.1

Fix small bugs:

1. more reliable log directory setter. We now no longer produce a daisy_logs directory as an artifact when running tests
2. serial server doesn't mark all blocks as success. If block marked as failed it is released as a failed block

Feature
Server.run_blockwise now returns the task state dict mapping task names to task states. This is useful if you ever want to know how many blocks failed, skipped, were orphaned etc.

v1.2

Features:

• add support for MacOS and Windows (via supporting multiprocessing start method "spawn")
• add a Serial Server. run_blockwise convenience function can be run with multiprocessing=False
• class methods can now be used as worker start functions and process block functions

Bug Fixes:

• no longer double counts orphans in some cases

v1.1.1

Small patch release.

Fixes a bug where daisy would hang indefinitely if the check_function skipped all blocks in all available tasks resulting in downstream tasks never having their workers recruited.

v1.1

changes:

• increased max message size from 2**16 to 2**64
• check every second for whether all blocks have finished in the server event loop
• use context manager within run_blockwise to ensure process terminates
• remove deep copy in the calls to shrink. This was significantly slowing down the scheduler with excessive copying

v1.0

v1.0

Release v0.2

Bump to v0.2