Trim allocations on the per-constraint and per-violation paths

[jungm]

Summary

Trims allocations on the per-constraint and per-violation paths.

 Frame.validate(constraint)
   ConstraintValidatorContextImpl.violations
-    Lazy<HashSet>
+    ArrayList(2) with a containment check
   createViolation
-    context path materialized, then copied again
+    derived once from the parent context's path
-    Arrays.deepHashCode(new Object[] {...})
+    field-by-field hash
 composing constraints of a @ReportAsSingleViolation constraint
-  build violations (path, message, hash) and discard them
+  record only whether they failed
 BeanFrame property frames, cascading
-  traversableResolver.isReachable/isCascadable (+ path copies)
+  skipped when exactly DefaultTraversableResolver without JPA
 PathImpl
-  LinkedList<NodeImpl>
+  ArrayList<NodeImpl>

Evidence

• Before / After: benchmarks

Numbers are JMH throughput (-p provider=bval -f 2 -wi 2 -i 3, 1 s iterations, -prof gc, JDK 25), measured against the PR below this one. The machine was busy, so throughput carries wide error bars; the bytes allocated per operation (B/op) are deterministic.

Benchmark	Before	After	Change	B/op before → after
composedInvalid (ops/ms)	669 ± 5%	1,192 ± 9%	1.78×	4,712 → 2,532
wideBeanInvalid (ops/ms)	68.1 ± 44%	80.6 ± 150%	1.18×	28,157 → 20,600
deepGraph (ops/ms)	87.9 ± 68%	97.0 ± 46%	1.10×	31,289 → 23,154
mixedDrivers (ops/ms)	536 ± 27%	507 ± 62%	0.95×	5,422 → 4,332

All benchmarks

Benchmark	Before	After	Change	B/op before → after
cascadedCycles (ops/ms)	2.42 ± 14%	0.95 ± 66%	0.39×	1,202,427 → 1,098,127
cascadedList (ops/ms)	0.32 ± 33%	0.29 ± 82%	0.92×	5,778,430 → 5,269,994
classLevelInvalid (ops/ms)	516 ± 61%	729 ± 96%	1.41×	3,896 → 3,256
classLevelValid (ops/ms)	978 ± 68%	1,332 ± 11%	1.36×	3,040 → 2,592
composedInvalid (ops/ms)	669 ± 5%	1,192 ± 9%	1.78×	4,712 → 2,532
composedValid (ops/ms)	1,042 ± 5%	951 ± 38%	0.91×	2,976 → 2,152
containerElements (ops/ms)	1.84 ± 19%	1.92 ± 9%	1.04×	1,623,248 → 1,371,279
deepGraph (ops/ms)	87.9 ± 68%	97.0 ± 46%	1.10×	31,289 → 23,154
elMessage (ops/ms)	657 ± 99%	836 ± 45%	1.27×	3,840 → 3,412
executableParameters (ops/ms)	532 ± 10%	422 ± 87%	0.79×	5,761 → 5,423
executableReturnValue (ops/ms)	496 ± 12%	338 ± 46%	0.68×	6,255 → 5,460
explicitGroup (ops/ms)	734 ± 11%	703 ± 32%	0.96×	4,930 → 4,084
mixedDrivers (ops/ms)	536 ± 27%	507 ± 62%	0.95×	5,422 → 4,332
validateProperty (ops/ms)	2,026 ± 4%	1,615 ± 27%	0.80×	1,795 → 1,764
validateValue (ops/ms)	2,479 ± 45%	3,477 ± 10%	1.40×	1,070 → 1,088
wideBeanInvalid (ops/ms)	68.1 ± 44%	80.6 ± 150%	1.18×	28,157 → 20,600
wideBeanValid (ops/ms)	72.0 ± 92%	120 ± 68%	1.67×	17,308 → 13,737
Jsr303Benchmark.bvalBootstrapAndValidate (ops/s)	1,652 ± 61%	1,523 ± 43%	0.92×	250,395 → 251,542
Jsr303Benchmark.bvalBuildFactory (ops/s)	3,164 ± 43%	7,148 ± 34%	2.26×	89,026 → 88,164
Jsr303Benchmark.bvalConstraintsFailure (ops/s)	471,733 ± 46%	412,829 ± 89%	0.88×	5,000 → 4,236
Jsr303Benchmark.bvalConstraintsFailureReuseValidator (ops/s)	514,364 ± 60%	458,567 ± 38%	0.89×	5,048 → 4,248
Jsr303Benchmark.bvalConstraintsSuccess (ops/s)	826,895 ± 91%	930,692 ± 50%	1.13×	3,068 → 3,032
Jsr303Benchmark.bvalConstraintsSuccessReuseValidator (ops/s)	1,133,176 ± 70%	575,500 ± 101%	0.51×	3,080 → 2,900
Jsr303Benchmark.bvalNoConstraints (ops/s)	159,318,160 ± 10%	132,245,313 ± 93%	0.83×	0 → 0
Jsr303Benchmark.bvalNoConstraintsReuseValidator (ops/s)	147,841,169 ± 67%	160,105,673 ± 54%	1.08×	0 → 0

Several throughput rows (e.g. cascadedCycles ±66%) are within noise even though their allocation went down.

• Tests: bval-jsr unit tests: 255 run, 0 failures · TCK: 1070 run, 0 failures, 1 skipped

Merge Danger

Door: two-way

Blast Radius: violation paths

• Violation paths and hash codes are built differently; their equality is unchanged, and the TCK checks paths extensively.
• TraversableResolver: custom resolvers, subclasses of DefaultTraversableResolver and JPA environments are still consulted. Only the exact default class without JPA, which always answers yes, is skipped.