tkobayas commented on PR #6909:
URL: https://github.com/apache/incubator-kie/pull/6909#issuecomment-5616000126
Hi @Rikkola , Sorry for the late comments. It's an interesting feature!
The followings are findings from AI. They are ranked as important, but I
think you may deal with them in separate PRs.
---
### 1. Mixed alpha + beta expressions in one step → `ClassCastException`
When a single step has both an alpha expression and a beta expression
referencing the anchor:
```java
sequence(
pattern(toy)
.expr("alpha", t -> t.getName().startsWith("a"))
.expr("beta", person, (t, p) ->
t.getName().equals(p.getName()))
)
```
`PhreakNodeFactory` wraps the constraints in `CombinedAlphaConstraint`,
whose `getRequiredDeclarations()` always returns an empty array.
`DynamicFilter` then evaluates the whole thing as an alpha constraint and never
supplies the anchor tuple to the beta part, which causes a `ClassCastException`.
A single beta expression works fine (existing tests cover that). The issue
is only with mixed alpha+beta in the same step.
If supporting this combination is out of scope for now, it would be good to
reject it at build time with a descriptive error.
**Relevant code:**
- `CombinedAlphaConstraint.java:54`
- `PhreakNodeFactory.java:184`
- `DynamicFilter.java:59`
### 2. Named consequence before a later sequence → wrong anchor skip
When a named consequence sits between two sequences:
```java
pattern(person),
sequence(pattern(toy).expr("ball", ...)),
when("always", person, p -> true).then(
on(toy).breaking().execute(...)
),
sequence(pattern(relationship).expr("go", ...))
```
`buildSequenceVarIndexesForConsequence()` calculates `skip = totalSeqs - 1 -
sequenceIndex`, but `totalSeqs` counts every sequence in the rule, including
those after the named-consequence terminal. When the first sequence completes,
its consequence has only one sequence anchor in the tuple chain but the
compiler recorded `skip=1`, producing:
```
IllegalStateException: SequencerFactSupplier: no SequencerMemory anchor
found at skip=1
```
The metadata should be calculated relative to each consequence terminal
rather than the rule-wide total. Recording a sequence/node identity instead of
a positional skip count would be more robust.
**Relevant code:**
- `KiePackagesBuilder.java:434-441`
- `LambdaConsequence.java:396-406`
### 3. `.bind(...)` variables inside a sequence → `NullPointerException` in
RHS
```java
Variable<String> toyName = declarationOf(String.class);
sequence(
pattern(toy)
.expr("ball", t -> t.getName().equals("ball"))
.bind(toyName, Toy::getName)
),
on(toyName).execute(name -> results.add(name))
```
`buildStepGate()` records only `patternImpl.getPatternVariable()` in
`sequenceVarIndexes`; variables from `.bind(...)` are not recorded and are
absent from the activation tuple too, so the consequence fails with:
```
NullPointerException: Cannot invoke "Declaration.getTupleIndex()" because
"declaration" is null
```
Since the API accepts a full `PatternDef` as a `SequenceStep`, its bindings
should work. If that's intentionally unsupported, `.bind(...)` should be
rejected during model construction or compilation.
**Relevant code:**
- `KiePackagesBuilder.java:627-632`
- `LambdaConsequence.java:251-258`
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]