Changelog#

2.2.6 - March 26, 2024#

✨ Improved#

  • Removed use/setting of the event loop in BaseClient.__init__().

2.2.5 - March 19, 2024#

✨ Improved#

  • AMQP REPLY log messages now include the header of the message.

🏷️ Changed#

  • #120 Callbacks that are synchronous functions are now called immediately instead of being scheduled with call_soon().

2.2.4 - February 26, 2024#

✨ Improved#

  • Log the command string when a new command is received by an actor.

2.2.3 - January 20, 2024#

✨ Improved#

  • #119 If an exception object is passed to BaseActor.write(), the filename and line number where the exception where raised are included in the output. The user can choose what traceback frame to output by passing traceback_frame to write().

🔧 Fixed#

  • Handle cases when a message tries to be pushed to an non-existing exchange.

⚙️ Engineering#

  • Update workflows and docs building.

2.2.2 - November 10, 2023#

✨ Improved#

  • Prevent multiple reconnections when using the AMQPClient context manager.

2.2.1 - November 10, 2023#

🚀 New#

  • AMQPClient now supports an asynchronous context manager that starts the client and closes the connection on exit.

2.2.0 - August 25, 2023#

🚀 New#

  • Added support for tasks. Tasks are coroutines that can be commanded remotely and receive a simple dictionary payload. They don’t provide command completion tracking or replies to users. See more here.

🔧 Fixed#

  • In ActorHandler, error messages are output as "e" instead of as "f".

  • Start the file logging even if a custom log instance is provided to BaseClient.

⚙️ Engineering#

  • Use ruff for linting.

  • Use sphinx_autodoc_typehints and updated pygments in documentation.

2.1.3 - July 27, 2023#

✨ Improved#

  • Added write_to_log argument to BaseClient.write() to decide whether a reply should be written to the log. Changed several commands (get-command-model, help) to emit replies without logging them.

2.1.2 - July 20, 2023#

⚙️ Engineering#

  • Updated unclick to 0.1.0b5.

2.1.1 - July 13, 2023#

✨ Improved#

  • Allow to add a model to a ModelSet after initialisation.

  • Flush messages after printing them to the CLU CLI.

2.1.0 - June 21, 2023#

🚀 New#

  • #117 Added a WebsocketServer class to implement a pass-through websocket client-server connection to the AMQP exchange.

✨ Improved#

  • AMQPClient can be called without a name, in which case a unique random one will be generated.

2.0.2 - May 24, 2023#

🔧 Fixed#

  • Prevent commands sends to the CLU CLI to block until finished.

2.0.2b1 - May 23, 2023#

⚙️ Engineering#

  • Remove pin on setuptools version.

2.0.1 - March 12, 2023#

🚀 New#

  • Added AMQPClient.add_reply_callback() which allows to register callback functions that are called with an AMQPReply object every time a reply is received. These replies are not filtered and the callback is called for each valid reply.

2.0.0 - March 10, 2023#

💥 Breaking changes#

  • By default AMQPClient.send_command() will await the command itself when the method is awaited. This means that cmd = await client.send_command('my_actor', 'ping') will await until the ping finishes and the replies are received. The previous behaviour can be forced with await_command=False, e.g., cmd = await (await client.send_command('my_actor', 'ping', await_command=False)). In general this should not be a breaking change since awaiting a done command will return immediately, but it’s marked as breaking since in some corner cases there could be some unexpected behaviour.

✨ Improved#

  • Added an internal argument in CluCommand that marks the received command as internal.

  • Bump unclick to 0.1.0b4.

🏷️ Changed#

  • When a command status changes to running, the status is emitted as internal.

2.0.0b2 - March 10, 2023#

🔧 Fixed#

  • Fixed asserting of command_id and user_id in testing module.

2.0.0b1 - March 10, 2023#

💥 Breaking changes#

  • Removed support for Python 3.7.

🚀 New#

  • Support Python 3.11.

  • Use aio_pika>=9.0.0 and aiormq>=6.6.4.

  • Support internal replies that are not shown in the CLI. Modified some command to use internal replies.

  • Added get-command-model command that uses unclick to return a JSON representation of a command or the entire command parser. This can be used to create a programmatic API that interfaces with an actor using command strings.

  • Allow to filter by actor(s) in the CLI.

⚙️ Engineering#

  • Added MessageCode enumeration.

1.9.2 - January 24, 2022#

✨ Improved#

  • The click context object can now be access as Command.context if the click parser is being used.

1.9.1 - December 2, 2022#

🔧 Fixed#

  • Deal with replies without value in MockReplyList.

1.9.0 - October 19, 2022#

🚀 New#

  • #111 If a LegacyActor uses the Click parser, and a connection to Tron is defined, adds a tron-reconnect command that can be used to force recreating the connection from the actor to Tron.

🔧 Fixed#

  • Avoid setting the event loop on init in CallbackMixIn. This caused the event loop to not be running in some cases when notify() was invoked.

  • Reworked how TronConnection uses the reconnecting protocol so that it actually reconnects.

1.8.2 - September 15, 2022#

🔧 Fixed#

  • #110 If the content of a keyword in LegacyActor is empty, output the keyword without the equal sign (e.g., test instead of test=).

1.8.1 - September 1, 2022#

💥 Breaking changes#

  • Removed the create_setup.py file. Poetry should now support all its use-cases.

🚀 New#

  • Added a KeywordStore that stores each time a keyword was output. It can be enabled by passing store=True when instantiating an actor and accessed as actor.store. See the documentation for more details.

  • Added cancel_command(), get_current_command_name(), and get_current_command_name() to click parser.

1.7.0 - August 14, 2022#

💥 Breaking changes#

  • The client model for an actor is now updated if the actor outputs a property that is not in its schema. This is not necessarily a breaking change, but it modifies the expectation that the client will validate replies against the actor schema. With this change we indicate that validating and respecting its own schema is a task for the actor and not for the client. There are at least a couple cases for which having the client enforce the actor schema was problematic: if the actor has patternProperties those are legal properties but are not updated in the client model; and if the actor decides to output a non-validated message. With this change, if the client receives a property that is not in the actor schema, a new Property is created int eh actor Model and its value is updated. That Property has Property.in_schema=False to indicate that the property has not been validated. Note that this change does not affect the TronConnection models since those are defined by actorkeys.

🚀 New#

  • Command.write() can now be called with a logging level instead of a string message code. For example Command.write(logging.DEBUG, text="Hi") is equivalent to Command.write("d", text="Hi").

🔧 Fixed#

  • Fixed a ConnectionResetError when closing the connection to a TCPStreamServer.

1.6.2 - May 25, 2022#

✨ Improved#

  • The commander of a command sent to Tron is now actor.actor by default, instead of actor.actor.target_actor.

  • Trying to change the status of a done command will issue a CluWarning and return, instead of raising a RuntimeError.

🔧 Fixed#

  • Prompted by COS-74, it was possible for a reply from a command with the same MID but different commander to be processed and mark a running command as complete. Now we are checking now only the MID but the commander as well.

1.6.1 - May 11, 2022#

🚀 New#

  • New Command.child_command() method that allows to run a child command in the same actor already running Command. The practical effect is to run another of the same actor’s commands as if it were part of the current Command.

✨ Improved#

  • Allow AMQP clients to listen to their own replies. This allows an actor sending a command to itself to know when the command is done.

  • Command.send_command() now accepts a new_command argument. When new_command=True, the new command will receive a new command ID and the commander ID will be the actor running the command (to all effects this is equivalent to BaseClient.send_command()). If new_command=False, the command ID and commander of the current actor will be used. In this case, and if the target actor is the same actor running the current command, consider using Command.child_command() instead.

1.6.0 - April 25, 2022#

🚀 New#

  • Actors can use set_message_processor() to set a function that will receive the message dictionary before being output to the users and can make modification on before it’s emitted. This is useful, for example, to improve compatibility in actors that can be run as legacy or AMQP actors.

🧱 Support#

  • Relax sdsstools dependency to >=0.4.13. This is desirable since sdsstools is not yet at the 1.0 level and minor version changes block poetry.

1.5.8 - February 10, 2022#

🔧 Fixed#

  • #107 Add an alias Reply.body to Reply.message to allow backwards compatibility.

1.5.7 - February 5, 2022#

✨ Improved#

  • Command.replies is now a list of Reply instances with a get method. Replies are unified for all kinds of clients.

1.5.6 - January 6, 2022#

🚀 New#

  • #106 Command now accepts a time_limit argument that will mark it as TIMED_OUT and done after an interval. time_limit can also be passed to send_command().

✨ Improved#

  • Command.actor is always typed as an actor type.

1.5.5 - December 14, 2021#

🚀 New#

  • Added a FakeCommand that writes to the log and that can be used when a command may not be present.

✨ Improved#

  • #104 The existing @cancellable decorator did not work in subcommands. The decorator has been removed and now it’s possible to pass cancellable=True to the command decorator (e.g., @command.parser.command(cancellable=True)). This takes care of adding a --stop option to the command. The underlying behaviour of command cancellation has not changed.

1.5.4 - November 26, 2021#

🔧 Fixed#

  • Do not try to update model when a keyword from Tron cannot be parsed.

1.5.3 - November 26, 2021#

🔧 Fixed#

  • Do not use a task to update the model. This caused failures when messges where being sent from a function called with an executor.

1.5.2 - November 7, 2021#

🔧 Fixed#

  • Fixed a bug in which commanders with multiple dot-separated components would not be correctly interpreted by handle_reply() and the CLI.

  • Command code e does not fail the command. This prevents cases of finishing/failing the command after it has already been failed.

  • Improved the performance of the CLI by outputting all the parts of the message at once.

1.5.1 - November 6, 2021#

🚀 New#

  • Support Python 3.10.

✨ Improved#

  • #101 The legacy actor now accepts command strings with a commander id (e.g., APO.Jose 10 status). The commander id is stored in the Command object. Added a Command.send_command() that will call the remote command propagating the commander ID. For example, if a command has commander APO.Jose and sends a command to guider, the guider actor will receive a command with commander APO.Jose.guider.

  • #102 When calling a model or property callback, ensure that the arguments sent are a frozen copy of the state of the model/property at the time of the callback. This prevents that if a model is updated twice in quick succession, the callback may receive only the second value. A consequence of this change is that the model callback now receives a flattened dictionary with all the model keywords as the first argument.

  • If not specified, a message with code e or f, or an exception, will use the keyword error.

1.5.0 - October 12, 2021#

✨ Improved#

  • #99 Add exception_module value to the output of an exception.

  • sdss/sdsstools#29 Allow to pass a custom PyYAML loader in from_config() that will be forwarded to read_yaml_file().

  • Actors now accept <actor> --help with the same result as <actor> help.

  • Add additional_properties parameter to LegacyActor.

  • When a legacy actor starts, if there is a TronConnection available it will try to send a hub startNubs <actor> to initiate the connection.

🔧 Fixed#

  • Use clu.client for TronConnection.send_command().

  • If a command in a LegacySurvey actor is left running after the client closes the connection, it would still try to output messages to it, causing a socket.send() raised exception error. Now if the client exists the command continues running, but outputs to that client are ignored.

1.4.0 - September 27, 2021#

🚀 New#

  • #98 Add unique() and cancellable() decorators for Click command parsers.

✨ Improved#

  • #95 Cast all arguments to string in ProxyClient.send_command().

  • Add get_keys parameter to LegacyActor.start() that is passed to TronConnection.start()

  • Use .client as default commander for TronConnection and {actor}.{target} for actor.

🔧 Fixed#

  • Avoid and error in the callback when a connection to the TCP server is closed.

1.3.0 - September 17, 2021#

💥 Breaking changes#

  • #86 additionalProperties is set to false by default if not specified, including if schema=None when initialising an actor.

🚀 New#

  • #85 Added BaseClient.proxy() method.

✨ Improved#

  • #90 If an exception object is passed as a keyword in a command or actor message, it will be unpacked into the exception type and message.

  • Make the error output when a reply fails to validate more clear.

🔧 Fixed#

  • #91 Documentation example for testing with CLU.

🧹 Cleanup#

  • Add invoke_mock_command() stub method to BaseClient to simplify type checking.

1.2.1 - June 20, 2021#

🚀 New#

🔧 Fixed#

  • LegacyActor now accepts the config parameter sent by from_config().

✨ Improved#

  • Subcommands now won’t write to the users when they start running.

1.2.0 - June 3, 2021#

🚀 New#

  • Command and BaseActor.write() now accept a silent argument that if True will execute the command normally and update the status and internal model, but won’t write to the user. Timed command can be run in silent mode the first iteration by initialising them with first_silent=True.

✨ Improved#

  • #77 Child commands will never emit : or f messages that may be confused as the parent being done.

  • Timed commands are run immediately when started.

  • from_config() now passes the configuration to the client __init__() so that it is accessible during initialisation.

  • If a timed command takes longer to run than the interval at which the poller checks if new timed commands should be run, prevent it from being issued multiple times.

1.1.2 - May 31, 2021#

🔧 Fixed#

  • Revert previous changes to the typing of Command that were causing problems, but keep the generic for the command future.

1.1.1 - May 30, 2021#

🔧 Fixed#

  • Correctly assign the type of the actor in a Command.

  • Fix error when TopicListener.stop() is called and there is not an active connection.

✨ Improved#

  • Device.start() now returns self.

  • #84 send_command now accepts multiple arguments before the keyword arguments. If they are passed, they will be concatenated to create the full command string. For example: client.send_command('my_actor', 'sum', '-v', 2, 4, command_id=5) is equivalent to client.send_command('my_actor', 'sum -v 2 4', command_id=5)

1.1.0 - May 29, 2021#

🚀 New#

  • #82 send_command now accepts a callback argument. If set, the callback will be called each time the actor replies and will receive the reply itself (AMQPReply in case of AMQPClient/Actor and clu.legacy.types.messages.Reply for Tron/TronConnection). Thanks to Florian Briegel for the idea.

✨ Improved#

  • #81 Improve typing of BaseCommand and command replies.

  • In TronConnection, do not fail with a ParseError if one of the keywords cannot be parsed. Instead, issue a warning and move on to the next one.

  • The CLI now checks that the preferred style (solarized-dark) is available. Otherwise defaults to pygments default style.

  • Copy Property before notifying the callbacks. This prevents the value passed being updated in the time that it takes for the callback to go out.

  • Property, BaseModel, and TronModel now have a last_seen attribute that is updated with the Unix time when the model or property/key are updated.

  • When the AMQP client is handling replies for a command, it will update the status every time it changes, not only when it is done or failed.

  • StatusCommand callbacks now receive the status itself as an argument.

1.0.3 - May 20, 2021#

✨ Improved#

  • When tracking the status of a command sent to Tron, update the status with each received reply, and store all the replies.

  • When as_complete_failer cancels the tasks after an exception, suppress all possible exceptions, not only CancelledError, since the original exception will be raised again. Add tests for as_complete_failer.

1.0.2 - May 18, 2021#

🔧 Fixed#

  • #78 Fixes a bug in which an actor with a defined TronConnection that had failed to start would still try to send commands to Tron.

✨ Improved#

  • #79 TronConnection now uses a ReconnectingTCPClientProtocol that will try to keep the socket to Tron open, allowing Tron to restart without losing connection.

🧹 Cleanup#

  • releases was misbehaving once we reached 1.x, and its interpretation of semantic versioning was a bit too extreme. Instead, we are now using a Markdown file with myst-parser. The previous changelog is still available here.

1.0.1 - May 16, 2021#

✨ Improved#

  • BaseActor receives a validate parameter that can be used to globally define whether the actor should validate its own messages against the model.

1.0.0 - May 12, 2021#

🚀 New#

  • Transition CLU to stable!

🧹 Cleanup#

  • Upgrade click to ^8.0.0.

The Pre-history#

The changelog for versions previous to 1.0.0 can be found here.