perf: valTag dispatch for O(1) materializer type routing

[He-Pin]

Motivation

The materializeRecursiveChild method in Materializer.scala dispatches on Val subtype using a Scala pattern match, which compiles to sequential instanceof checks. For 8 concrete Val subtypes, this means up to 7 type tests before reaching the correct branch. This is the hot path for all JSON materialization.

Key Design Decision

Introduce a valTag: Byte field on each Val subclass with contiguous values 0-7, enabling @switch annotation to generate JVM tableswitch bytecode for O(1) dispatch. Verified with javap -c that the bytecode is:

tableswitch { // 0 to 7
  0: ...  // TAG_STR
  1: ...  // TAG_NUM
  ...
  7: ...  // TAG_FUNC
  default: ...  // Materializable, TailCall, unknown
}

Custom Val subclasses (e.g. Materializable) use valTag = -1 and fall through to a nested pattern match in the default branch.

Modification

• Val.scala: Added private[sjsonnet] def valTag: Byte abstract method to sealed abstract class Val. Added TAG constants (TAG_STR=0 through TAG_FUNC=7). Every concrete subclass overrides valTag.
• Materializer.scala: Replaced pattern-match dispatch in materializeRecursiveChild with @switch tableswitch dispatch. Added null guard before valTag access to preserve original diagnostic. Hoisted xs.length out of materializeRecursiveArr while-loop.
• CustomValTests.scala: Added valTag: Byte = -1 to test ImportantString class.

Benchmark Results

JMH (35 benchmarks, single-threaded)

Benchmark	Before (ms/op)	After (ms/op)	Change
reverse	10.705	10.410	-2.8%
base64DecodeBytes	9.423	8.990	-4.6%
comparison2	74.108	72.667	-1.9%
realistic2	70.491	69.201	-1.8%

Zero regressions across all 35 benchmarks.

Native (hyperfine, 10 runs)

Benchmark	Before	After	jrsonnet	vs jrsonnet
comparison2	170.8ms	171.2ms	237ms	1.39x faster ✅
realistic2	295.9ms	294.1ms	483ms	1.64x faster ✅
reverse	49.6ms	50.1ms	33.6ms	1.49x slower

Native impact is neutral (within noise) — the primary value is architectural: tableswitch enables future materializer optimizations to benefit from O(1) dispatch.

Analysis

• JVM sees modest improvement (2-5%) because JIT can leverage the tableswitch more efficiently than cascading instanceof
• Native binary sees neutral impact because Scala Native already optimizes pattern matching well
• The primary value is architectural: O(1) dispatch is a foundation for future materializer fast-paths

References

• Upstream: jit branch commits 30b7495b (valTag dispatch) and 9ddb1a57 (tableswitch dispatch)
• https://github.com/He-Pin/sjsonnet/tree/jit

Result

All 140 tests pass. @switch tableswitch bytecode verified via javap -c. No regressions.