The OpenTabletDriver 0.6.x plugin pipeline has edge cases that can cause filter plugins to incorrectly drop reports.
For plugins with filters targeting the 0.6.x plugin API the following tests should be done.
Plugins targeting earlier API's like 0.5.x should preferably not be covered if it significantly increases the amount of code to maintain.
It is unclear how to implement these tests properly. I currently have 2 ideas:
- Have a separate repository that can be targeted as a GitHub action, that tests the filter in a virtual pipeline, which could be used to test pull requests on this repository in some way.
- Driver gets a "Plugin Developer" mode that would actively check active filters for these issues - this would help plugin developers from making these mistakes unintentionally.
While outside of the scope of this issue, future driver plugin API's should preferably be designed in a way that removes these necessary tests.
Minimum Test Coverage
At the very minimum, the tests should check for the following
Properly passes through IDeviceReports
Some filters incorrectly only pass through ITabletReports. Normal filters should ensure Consume(IDeviceReport) runs Emit?.Invoke(). Async filters should ensure ConsumeState() runs OnEmit() (or Emit?.Invoke()) on unconsumed reports.
Sets a PipelinePosition that isn't PreTransform or PostTransform (or an alias)
Currently, the driver only links filters that have a PipelinePosition with one of the above values as seen here:
https://github.com/OpenTabletDriver/OpenTabletDriver/blob/aa345e5b7365647f25c8651f3e35fb7dd4c13676/OpenTabletDriver.Plugin/Output/OutputMode.cs#L62-L90
This means that a PipelinePosition of e.g. 0 would result in the filter never being inserted into the pipeline.
If you think of anything else that would be worth testing for, please let us know here.
The OpenTabletDriver 0.6.x plugin pipeline has edge cases that can cause filter plugins to incorrectly drop reports.
For plugins with filters targeting the 0.6.x plugin API the following tests should be done.
Plugins targeting earlier API's like 0.5.x should preferably not be covered if it significantly increases the amount of code to maintain.
It is unclear how to implement these tests properly. I currently have 2 ideas:
While outside of the scope of this issue, future driver plugin API's should preferably be designed in a way that removes these necessary tests.
Minimum Test Coverage
At the very minimum, the tests should check for the following
Properly passes through
IDeviceReportsSome filters incorrectly only pass through
ITabletReports. Normal filters should ensureConsume(IDeviceReport)runsEmit?.Invoke(). Async filters should ensureConsumeState()runsOnEmit()(orEmit?.Invoke()) on unconsumed reports.Sets a
PipelinePositionthat isn'tPreTransformorPostTransform(or an alias)Currently, the driver only links filters that have a
PipelinePositionwith one of the above values as seen here:https://github.com/OpenTabletDriver/OpenTabletDriver/blob/aa345e5b7365647f25c8651f3e35fb7dd4c13676/OpenTabletDriver.Plugin/Output/OutputMode.cs#L62-L90
This means that a
PipelinePositionof e.g.0would result in the filter never being inserted into the pipeline.If you think of anything else that would be worth testing for, please let us know here.