Updated the record command to allow revent recordings to be made.
They can be performed in 3 ways; the default is a standard recording
as in WA2, it can start a specified package before starting the
recording or a workload can be specified which launches the workload
and prompts the user to record each stage, naming and storing the recordings
appropriately.
Added a new ReventWorkload as a base for revent based workloads.
Updated ReventGui to use `revent_recorder` and now uses `target.model`
instead of `taget.name`.
Previously if a parameter was not specified via config it would not
be used during initialisation even if the parameter had a default
value. Now any parameters with default values are populated as necessary.
Revent files are automatically placed in the sub folder `revent_files` in the
workload directory when recording, therefore when trying to retrieve recordings
the getter now looks inside of the sub directory.
Previously trying to use the `__base_filepath` import from inside the `Package`
class, resulted in the error "`_Package__base_filepath` is not defined."
Added a workload type to handle workloads that have both an APK with an
application and associated automation JAR. Added benchmarkpi
implementation using using the new workload.
- get_by_extension was comparing the expected extension to the entire
tuple returned by os.path.splitext(), rather than just the extension
part.
- UserDirectory getter was looking in the root dependencies directory,
rather than the subdirectory for the owner.
- Filer getter was not handling non-existing paths properly.
- Workload's update_result stage has now been broken up into two parts:
extract_results and update_output. This is to allow the assistant to
pull output from the target in between the two stages.
- Updated assistant implementations for Linux and Android targets from
the exisiting code.
- Extended target descriptor code to handle assistants and their
parameters as well.
- Updated the target manager to actually make use of the assistants.
Previously a config point would try to validate the name attribute of the object
it was meant to be setting, rather than what it was meant to be setting it to.
- Implemented result processor infrastructured
- Corrected some status tracking issues (differed between states
and output).
- Added "csv" and "status" result processors (these will be the default
enabled).
- Fixed up some of the signal map for instrumentation
- Changed how priorites are specified -- no longer method name prefixes
but dedicated decorators, including an easy way of specifying a custom
priority level (no longer need to manually connect signals)
- Updated ExecutionTimeInstrument to work with the new system
- Also removed some dead code
"params" is interpreted differently in section vs workload entires in
the agenda; previously, this was handled in the generic entry
construciton function by examining the ID prefix of the entry to
distinguish between the two.
This is unreliable as the user may specify their own IDs that won't have
the expected prefixes. To handle this, the "params" alias resolution for
sections is now handled in section specific part of entry processing
(workloads are handled autmatically because that is the default for
the corresponding JobConfig config point).
