Procedural constraints annotation processor + build logic #1611
Conversation
|
This PR is currently orphaned, pair programming hosted by @mattdailis prior to handoff. This PR has compiled a constraint w/o arguments, so it's potentially close. @JoelCourtney to add example constraints to this branch Decision to be made: support non-records or updating docs? @JoelCourtney proposing to just update docs and allow non-records later. Let's start with only records and double-back for non-record support. |
JoelCourtney
force-pushed
the
feature/procedural-constraints-processor
branch
from
db9d10f to
4dc61a6
Compare
JoelCourtney
had a problem deploying
to
e2e-test
GitHub Actions
Failure
|
State of PR: The annotation processor worked without any extra work. It was actually so good at combining the scheduling / constraints workflows that the gradle task can't tell them apart, and the original
I haven't made the equivalent of the scheduler's |
JoelCourtney
temporarily deployed
to
e2e-test
GitHub Actions
Inactive
...ling/src/main/kotlin/gov/nasa/ammos/aerie/procedural/scheduling/SchedulingProcedureMapper.kt
Outdated
Show resolved
Hide resolved
JoelCourtney
force-pushed
the
feature/procedural-constraints-processor
branch
from
cefb4ec to
81e7fd7
Compare
JoelCourtney
temporarily deployed
to
e2e-test
GitHub Actions
Inactive
JoelCourtney
temporarily deployed
to
e2e-test
GitHub Actions
Inactive
JoelCourtney
force-pushed
the
feature/procedural-constraints-processor
branch
from
5291317 to
29b501a
Compare
JoelCourtney
had a problem deploying
to
e2e-test
GitHub Actions
Failure
JoelCourtney
force-pushed
the
feature/procedural-constraints-processor
branch
from
29b501a to
60626ae
Compare
JoelCourtney
force-pushed
the
feature/procedural-constraints-processor
branch
from
60626ae to
aea67e7
Compare
JoelCourtney
force-pushed
the
feature/procedural-constraints-processor
branch
from
aea67e7 to
b2ee0d9
Compare
|
| while(currentAnchorId != null && anchoredToStart){ | ||
| currentActivityDirective = completeMapOfDirectives.get(currentAnchorId); | ||
| currentAnchorId = currentActivityDirective.anchorId(); | ||
|
|
||
| if (currentAnchorId.equals(firstId)) { |
There was a problem hiding this comment.
This line needs to be reversed to avoid a NPE: if(firstId.equals(currentAnchorId)
This is because we know for a fact that firstId is not null, as we otherwise wouldn't have entered the while loop, but we expect currentAnchorId to become null at some point, assuming there is no cycle.
There was a problem hiding this comment.
Additionally, I'm not sure if tracking a single id (as opposed to the set of all seen ids so far) is sufficient to detect a cycle, as the cycle may start further up than the initial directive id. If it is, could you please break down why it is?
For context, the database uses the "set of all seen ids" approach to detecting cycles.
Set proposal based on what I did to get tests working on my branch:
final var seenIds = new HashSet<ActivityDirectiveId>();
seenIds.add(currentAnchorId);
while(currentAnchorId != null && anchoredToStart){
currentActivityDirective = completeMapOfDirectives.get(currentAnchorId);
currentAnchorId = currentActivityDirective.anchorId();
if (seenIds.contains(currentAnchorId)) {
throw new IllegalStateException("Anchor ID cycle detected on " + currentAnchorId);
}
//...This fix belongs in #1611 but I want tests on my branch to pass
Description
This PR does a few things related to procedural constraints workflow:
ConstraintProcedureannotation and handles it together with scheduling procedures in the same processor.EditablePlan,Plan, andSImulationResultsinterfaces for stateless-aerie plan objects.ObeyConservationOfBananaand tests it using plan stubs and stateless simulation. This is less a test of the constraint itself, and more a test-that-testing-is-actually-possible kind of thing.Verification
Documentation
I still need to update the procedural constraints docs.
Future work
We now have nearly everything for a full-loop external scheduler run. Once we make a plan serializer, you could make a driver that loads a plan from file, runs a series of constraints and goals (with simulation), and outputs a result plan.