* Pull has_emeter property up to tapodevice base class
This will also use the existence of energy_monitoring in the component_nego query to decide if the device has the service.
* Move emeter related functions to tapodevice
* Remove supported_modules override for now
This should be done in a separate PR, if we want to expose the available components to cli and downstreams
* Dedent extra reqs
* Move extra_reqs initialization
* Fix tests
* Do login entirely within AesTransport
* Remove login and handshake attributes from BaseTransport
* Add AesTransport tests
* Synchronise transport and protocol __init__ signatures and rename internal variables
* Update after review
* Add support for TAPO/SMART KLAP and seperate transports from protocols
* Add tests and some review changes
* Update following review
* Updates following review
This refactors `Discover.connect_single` by moving device instance construction into a separate device factory module.
New `SmartDevice.connect(host, *, port, timeout, credentials, device_type)` class method replaces the functionality of `connect_single`,
and also now allows constructing device instances without relying on UDP discovery for type discovery if `device_type` parameter is set.
---------
Co-authored-by: Teemu R. <tpr@iki.fi>
Several KASA devices seem to have pretty strict buffer size limitations on incoming/outgoing data transfers.
Testing on KL125-US and HL103 has shown that sending a request size larger than about ~768 bytes will immediately crash the device. Additionally, a query that generates a response larger than ~4096 bytes will crash the KL125-US. I was unable to generate such a large response to test the HL103.
The KL125-US will only return such large queries when its monthly usage stats have been populated. This means that a new bulb would work fine, but after a month of data collection the bulb would break the 4K limit and start to crash.
To work around this issue, an estimated worst-case response size is calculated before sending a request by summing up all modules estimated response size. If the estimated size is greater than the device's max_response_payload_size then the query will be split into multiple queries.
This PR implements splitting queries expected to have large responses and also removes the module 'skip list' which was a previous workaround to the crash (which worked by simply reducing the number of modules queried, which prevented the overflow) since it is no longer necessary.
This PR does not attempt to address the "input buffer size limit." Thus far this limit has not been an issue.
* Add plumbing for passing credentials as far as discovery
* Pass credentials to Smart devices
* Rename authentication exception
* Fix tests failure due to test_json_output leaving echo as nop
* Fix test_credentials test
* Do not print credentials, fix echo function bug and improve get type parameter
* Add device class constructor test
* Add comment for echo handling and move assignment
* test_cli: provide return values to patched objects to avoid warning about non-awaited calls
* test_cli: restore alias after testing
* smartstrip: remove internal update() calls for turn_{on,off}, set_led
* Make sure power is always a float
* Fix discovery tests
* Make tests runnable on real devices
* Add a note about running tests on a real device
* test_strip: run update against the parent device
Some devices are known to fail when trying to query non-supported modules like emeter information.
This commit makes the initial update() only request the sysinfo, followed up by a second query
if emeter is supported by the device.
* Add doctests to SmartBulb
* Add SmartDevice doctests, cleanup README.md
* add doctests for smartplug and smartstrip
* add discover doctests
* Fix bulb mock
* add smartdimmer doctests
* add sphinx-generated docs, cleanup readme a bit
* remove sphinx-click as it does not work with asyncclick
* in preparation for rtd hooking, move doc deps to be separate from dev deps
* pytestmark needs to be applied separately for each and every file, this fixes the tests
* use pathlib for resolving relative paths
* Skip discovery doctest on python3.7
The code is just fine, but some reason the mocking behaves differently between 3.7 and 3.8.
The latter seems to accept a discrete object for asyncio.run where the former expects a coroutine..
* Move tests to device-type specific test files to make improvements more approachable
* protocol: remove the port parameter from query, as there are no other known ports, fix docstrings
* Revise docstrings, remove superfluous information and remove unused methods ({set,get_icon} and set_time)
* cli: indent device output to make it more easily readable when having multiple devices
* remove adjust flake8 ignores (we have no setup.py anymore)
* pyproject: include cli tool to coverage, add config for interrogate (docstring coverage)
* bulb: raise exception on color_temp error cases instead of returning zero values
* improve bulb tests, simplify conftest
* strip: rename plugs property to children and move it to smartdevice
