Speed up EL messages, @Email and message interpolation

[jungm]

Summary

Removes the cost of the built-in constraints and of message interpolation, which dominate any validation that produces violations.

 DefaultMessageInterpolator.interpolate(template, context)
+  if cached for (locale, template, attribute map identity)
+    return cached
   resolve bundle keys and {attribute} references
+  if no "${" remains
+    cache and return
   ELFacade.interpolate
-    look-behind regex (100+ alternations) to disarm unescaped "#{"
+    scan for "#{" and count the preceding backslashes
 ConstraintD.getMessageTemplate
-  read reflectively from the annotation for every violation
+  memoized

 EmailValidator.isValid
-  DEFAULT_EMAIL_PATTERN.matcher(value).matches()
+  hand-written matcher, same language as DEFAULT_EMAIL_PATTERN
 AbstractPatternValidator.isValid
-  pattern.matcher(value).matches()
+  regexp ".*" without flags: no line terminator in value

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
elMessage (ops/ms)	51.6 ± 9%	696 ± 15%	13.49×	4,403 → 4,196
wideBeanInvalid (ops/ms)	12.4 ± 1%	59.2 ± 30%	4.76×	34,109 → 31,759
composedInvalid (ops/ms)	341 ± 8%	502 ± 2%	1.47×	6,092 → 5,652

All benchmarks

Benchmark	Before	After	Change	B/op before → after
cascadedCycles (ops/ms)	2.29 ± 43%	2.30 ± 12%	1.00×	1,217,820 → 1,259,572
cascadedList (ops/ms)	0.48 ± 23%	0.47 ± 15%	1.00×	5,862,671 → 6,110,831
classLevelInvalid (ops/ms)	668 ± 16%	683 ± 21%	1.02×	4,264 → 4,416
classLevelValid (ops/ms)	1,004 ± 7%	1,011 ± 16%	1.01×	3,224 → 3,248
composedInvalid (ops/ms)	341 ± 8%	502 ± 2%	1.47×	6,092 → 5,652
composedValid (ops/ms)	838 ± 9%	811 ± 13%	0.97×	3,240 → 3,172
containerElements (ops/ms)	1.48 ± 15%	1.66 ± 7%	1.12×	1,831,091 → 1,681,221
deepGraph (ops/ms)	102 ± 8%	90.9 ± 10%	0.89×	31,984 → 32,275
elMessage (ops/ms)	51.6 ± 9%	696 ± 15%	13.49×	4,403 → 4,196
executableParameters (ops/ms)	422 ± 5%	424 ± 21%	1.00×	6,296 → 6,211
executableReturnValue (ops/ms)	386 ± 9%	388 ± 19%	1.00×	6,950 → 6,841
explicitGroup (ops/ms)	472 ± 10%	577 ± 6%	1.22×	6,589 → 5,757
mixedDrivers (ops/ms)	487 ± 9%	353 ± 70%	0.73×	5,833 → 5,848
validateProperty (ops/ms)	1,361 ± 8%	1,010 ± 64%	0.74×	2,076 → 2,027
validateValue (ops/ms)	2,224 ± 5%	2,109 ± 40%	0.95×	1,352 → 1,325
wideBeanInvalid (ops/ms)	12.4 ± 1%	59.2 ± 30%	4.76×	34,109 → 31,759
wideBeanValid (ops/ms)	146 ± 15%	134 ± 41%	0.92×	18,688 → 17,553
Jsr303Benchmark.bvalBootstrapAndValidate (ops/s)	2,845 ± 43%	2,844 ± 48%	1.00×	245,360 → 245,596
Jsr303Benchmark.bvalBuildFactory (ops/s)	11,297 ± 5%	11,275 ± 3%	1.00×	88,057 → 87,993
Jsr303Benchmark.bvalConstraintsFailure (ops/s)	500,080 ± 9%	590,674 ± 16%	1.18×	5,648 → 5,520
Jsr303Benchmark.bvalConstraintsFailureReuseValidator (ops/s)	499,671 ± 11%	607,050 ± 8%	1.21×	5,588 → 5,520
Jsr303Benchmark.bvalConstraintsSuccess (ops/s)	1,213,713 ± 9%	1,001,414 ± 37%	0.83×	3,224 → 3,212
Jsr303Benchmark.bvalConstraintsSuccessReuseValidator (ops/s)	1,120,020 ± 28%	936,769 ± 49%	0.84×	3,212 → 3,248
Jsr303Benchmark.bvalNoConstraints (ops/s)	100,611,363 ± 21%	71,836,297 ± 29%	0.71×	0 → 0
Jsr303Benchmark.bvalNoConstraintsReuseValidator (ops/s)	119,052,819 ± 24%	119,638,823 ± 13%	1.00×	0 → 0

• Tests:

  ELFacadeTest           interpolate("${'a'} #{x} b") == "a $0x} b"       # outputs captured from master's regex
                         interpolate("${'a'} \#{x} b") == "a #{x} b"
  EMailValidationUtilsTest isValid(s) == isValid(s, DEFAULT_EMAIL_PATTERN)  # 19 examples + 100,000 random strings
  

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

Merge Danger

Door: two-way

Blast Radius: messages

• EL messages: a mistake in the scan would change interpolated text containing #{. The test pins master's exact outputs, including escaped forms.
• @Email without a custom regexp: the matcher must accept exactly what DEFAULT_EMAIL_PATTERN does. It's fuzzed against the regex.
• The interpolated-message cache holds one entry per (locale, template, constraint) that needs no EL, so it is bounded by the number of constraints. clearCache() clears it.

[tandraschko]

What about #{ here?