Accept a Behavior Annex loop that iterates an enclosing iterator 🤖

[lwrage]

Fixes #3226

Cause and correction

A for or forall whose element_values named an iterative variable threw ClassCastException out of DeclarativeToStrictTranslator.Builder.toElementValues, and the Xtext validator turned that into the opaque Error executing EValidator marker at line 0 with no strict model for the annex.

Two things were wrong on that one line.

The nesting is legal and had nothing to become. AS5506/3 Rev A D.6 admits an array data component reference as the iterated values, a data component reference starts with a local variable identifier, and D.6 gives the iterator of the construct the name scope of that construct, where it acts as a local variable. Nested iteration over an array of arrays is written that way. IterativeVariableHolder nevertheless did not implement ElementValues, unlike every other holder that can appear there, so aadlba.ecore now lists it, with the regenerated model code, AadlBaSwitch case, and child creation descriptor. The existing type check needed nothing further: the holder is a ValueVariable, so it resolves through the value branch of AadlBaUtils.getTypeHolder, which processes the Data Model array representation for a value inside a loop and gives the iterated values the element type and dimension of the classifier the iterator names.

The cast could fail for other shapes too. The iterated values share the D.7 integer value grammar with the rest of the annex, so an integer literal, a property reference, and a name that resolves to a data port or a port value reach the same cast, and none of them is an ElementValues either. toElementValues now tests the translated value instead of casting it, leaves what it cannot represent out of the strict model the way the rest of the translator does, and BehaviorAnnexValidator.checkIteratedValues reports it on the values as written. That check gates the strict checkers, like the neighboring checks for constructs the strict model cannot carry, since a loop that iterates nothing is not a loop the checkers can say anything useful about.

Regression

ba/org.osate.ba.tests/models/issue3226/Issue3226.aadl nests iteration over an array of arrays: the outer loop iterates a behavior variable of an array-of-arrays type and the inner loop iterates the outer iterator. Both loops write their classifier, so the model does not depend on #3173. Issue3226Test asserts that validation reports nothing, and separately that the translated inner loop carries the outer iterator as its iterated values — a loop with no iterated values at all would also validate and would say nothing about what the loop iterates.

models/issue3226/NonIterableValues.aadl is the other side of the cast: an integer literal and a data port, each of which must be reported on the values as written rather than throw. The corpus goldens record both models; the plain-AADL diagnostics goldens are empty, which pins that neither fixture has an unrelated core-AADL error.

All three assertions fail on master — two on the ClassCastException, one on the Error executing EValidator marker it produces — and BehaviorAnnexCharacterizationTest and BehaviorAnnexTranslationTest fail on the same exception with the fixtures present and the fix absent.

Validation

Focused regression, before the fix (regression commit only):

mvn -T5 -s releng/osate.releng/settings.xml -Plocal -Dtycho.localArtifacts=ignore -Dpr.build=true \
  -Dsign=false -Dspotbugs=false -Dcodecoverage=false -Djavadoc=false -DfailIfNoTests=false \
  -Dtest=Issue3226Test clean install

Tests run: 3, Failures: 2, Errors: 1 — the error is the ClassCastException at DeclarativeToStrictTranslator.toElementValues.

Same command with the fix: Tests run: 3, Failures: 0, Errors: 0.

Clean root reactor with the complete test suite:

mvn -T5 -s releng/osate.releng/settings.xml -Plocal -Dtycho.localArtifacts=ignore -Dpr.build=true \
  -Dsign=false -Dspotbugs=false -Dcodecoverage=false -Djavadoc=false -DfailIfNoTests=false clean install

BUILD SUCCESS, no failures or errors in any module, including the whole org.osate.ba.tests corpus. Only the four issue3226 goldens changed between the regression commit and the fix commit; no other corpus model's diagnostics or resolved model moved.

Dependencies and merge order

Based on master (0def9cf), independent of the open Behavior Annex PRs, and it can merge in any order relative to them. It does textually conflict with #3227 (#3173) in BehaviorAnnexValidator.java: both add a check constant and a line to the same representability chain, adjacent to each other. Whichever merges second needs that one trivial resolution, keeping both checks.

The two changes are complementary rather than competing: #3227 makes the iterator classifier optional and derives the iterator's type from the iterated values, and this PR makes an enclosing iterator one of the iterated values that can be written, so after both a nested loop can also omit the inner classifier.

Residual risk

Adding an interface supertype to an existing EClass shifts no classifier ID and adds no feature, and the regenerated diff is confined to the implements clause, one getESuperTypes().add, one switch case, and one child creation descriptor plus its icon. org.osate.ba.aadlba.ElementValues has no consumer outside the translator and AadlBaTypeChecker.checkElementValues, both exercised here.

The new diagnostic is the first report for iterated values the strict model cannot carry, so models that previously crashed the validator now get an error instead. No model in the corpus writes one — the existing fixtures iterate event data ports, behavior variables, data fields, and integer ranges — so nothing else in the test suite changes.

Type conformance between the iterated values and the iterator is not checked here, because AdaLikeDataTypeChecker.conformsTo is disabled repo-wide; only the arrayness of the values is. This PR does not change that.

🤖 Generated with Claude Code