Add @AllChildren annotation #1367
Conversation
| children = activityTypeRecord.effectModel().get().children().get(); | ||
| } else { | ||
| //if children have not been declared with an annotation, assume the activity might generate all activity types | ||
| children = missionModel.activityTypes().stream().map(ActivityTypeRecord::name).toArray(String[]::new); |
There was a problem hiding this comment.
Memory-wise, rather than returning a list of every activity in the plan, what about wrapping the return in Optional and returning Optional.empty? It would communicate the same idea-- that getChildren is returning the known children and that we don't know what the children are-- without needing a map of 300+ elements with each containing a list of 300+ elements (ie, in the Clipper case).
| import java.util.Map; | ||
|
|
||
| public interface SchedulerModel { | ||
| Map<String, DurationType> getDurationTypes(); | ||
| SerializedValue serializeDuration(final Duration duration); | ||
| Duration deserializeDuration(final SerializedValue serializedValue); | ||
| Map<String, List<String>> getChildren(); |
There was a problem hiding this comment.
Minor: If the data is provided in an array, what is gained by converting it into a List? Are there some List-only methods you expect the scheduler to use? Was it simpler to return a List in the code generator?
| import java.util.Map; | ||
|
|
||
| public interface SchedulerModel { | ||
| Map<String, DurationType> getDurationTypes(); | ||
| SerializedValue serializeDuration(final Duration duration); | ||
| Duration deserializeDuration(final SerializedValue serializedValue); | ||
| Map<String, List<String>> getChildren(); |
There was a problem hiding this comment.
I believe this change makes this PR breaking for existing models.
| .addStatement("final var result = new $T()", ParameterizedTypeName.get(ClassName.get(HashMap.class), ClassName.get(String.class), ParameterizedTypeName.get(List.class, String.class))) | ||
| .addCode( | ||
| missionModel | ||
| .activityTypes() | ||
| .stream() | ||
| .map( | ||
| activityTypeRecord -> | ||
| getChildrenStatement(missionModel, activityTypeRecord)) | ||
| .reduce((x, y) -> x.add("$L", y.build())) | ||
| .orElse(CodeBlock.builder()).build()) | ||
| .addStatement("return result") |
There was a problem hiding this comment.
Minor, nonblocking: this method could return an immutable map if it uses the Map.ofEntries() method.
Then instead of each getChildrenStatement returning result.put("TYPE", LIST), it would return Map.entry("TYPE", LIST).
|
This is blocked because part of an ongoing prototyping for dependency modeling in #1364. |
|
@dandelany I think we can close this PR. |
Description
Add a
@AllChildrenannotation to mission model so mission modelers can report, for a given activity, what other activity is called/spawned. This information will be used by the scheduler to evaluate whether it is useful to resimulate or not before goal evaluation: let's have G goal being satisfied by an activity of type A, G is the next goal to be evaluated. If an activity of type B was inserted at the last iteration, has B any chance of generating A and thus of satisfying the goal G ? If yes, simulation is required, otherwise it's not (for that only purpose, there might be other things that warrant a simulation, e.g. the goal needs to also evaluate an expression...).Some edge cases:
This PR does not contain changed to the scheduler to use this information. These changes would overlap too much with changes under review in #1323
Verification
A test with the banananation model has been added to verify that the generated code is accurate.
Documentation
To be done.
Future work
Other things in #1364 + use the annotation in the scheduler.