Keep a float32 a float

[komamitsu]

A behaviour change: a float32 on the wire now reads back as a float.

Why

The parser read float32 and float64 alike into a double and never told Jackson which it was (no getNumberTypeFP(), and getNumberType() always DOUBLE). So untyped binding (Object, Map<String, Object>, JsonNode) turned a Float into a Double, and its text became that of the widened double:

mapper.readValue(mapper.writeValueAsBytes(Map.of("f", 0.1f)), new TypeReference<Map<String, Object>>() {});
// before: {f=0.10000000149011612}  (a Double)
// after:  {f=0.1}                  (a Float)

Jackson's CBOR parser reports float32 as FLOAT32 (CBORParser.getNumberTypeFP()), and as documented, MessagePack libraries in languages with a 32-bit float type (Go, Rust, C#) keep it as one. msgpack-java core does not (unpackValue() widens every float to a double), but compatibility with that stack is not a goal here.

The change

• A float32 is reported as NumberType.FLOAT and NumberTypeFP.FLOAT32; a float64 as DOUBLE and DOUBLE64.
• getNumberValue() returns a Float for a float32.
• Its text, a key named by it, and getDecimalValue() follow the float's own shortest text (0.1), not the widened double's.
• Both widths stay in the one double field, which holds every float exactly, so the numeric conversions are shared.

DESIGN.md describes it in the parser section.

Tests

Four new tests in MessagePackParserTest, each failing on main: the accessors for a float32 and a float64, untyped reads (Float, Double, Float.NaN), a Float round trip through Map<String, Object>, and a float32 key's name. Two existing tests pinned the old behaviour (a float32 read untyped as a Double) and now expect a Float. ./gradlew clean build, 227 tests.

Performance

MsgpackReadBenchmark, 5 forks x 10 iterations x 2 s, both orders:

Run	Order	readPojoMsgpack main -> this PR	readPojoJson (control) main -> this PR
1	this PR first	684,869 -> 681,426 (-0.5%)	685,905 -> 699,627
2	main first	691,698 -> 690,822 (-0.1%)	701,942 -> 686,074

The MessagePack difference is within error both times. The JSON control, whose code is identical, moved about 2% towards whichever jar ran first, so the MessagePack/JSON ratio flipped with the order. No measurable effect. The fixture has no floats, so this checks that the extra case in every accessor switch does not slow the rest of the parser.

[Copilot]

Copilot review overview

🟡 Changes recommended

Add focused float32 getDecimalValue() regression coverage.

Review effort: Lite
Findings: 1

Open (1)

• Add float32 getDecimalValue assertion · New

What changed in this PR

Preserves MessagePack float32 semantics during Jackson parsing while retaining float64 behavior.

Changes:

• Adds width-aware floating-point types and values.
• Preserves float32 text, decimal conversion, and map keys.
• Updates tests and design documentation.

File	Summary
src/​test/​java/​org/​komamitsu/​jackson/​dataformat/​msgpack/​MessagePackParserTest.java	Adds and updates floating-point behavior tests.
src/​main/​java/​org/​komamitsu/​jackson/​dataformat/​msgpack/​MessagePackParser.java	Implements float32-aware parsing and accessors.
docs/​DESIGN.md	Documents floating-point width preservation.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[komamitsu]

Added in 0fe509a: aFloat32IsReportedAsAFloat now asserts getDecimalValue() is new BigDecimal("0.1") for the float32 0.1f (and for the float64 0.1). With the old widened conversion (BigDecimal.valueOf(doubleValue)) put back temporarily, it fails with 0.10000000149011612.

[Copilot]

Copilot review overview

🟢 Approval recommended

All reviewed changes are covered by tests and no unresolved blocking issues were identified.

Review effort: Lite
Findings: 1

Open (1)

• Add float32 getDecimalValue assertion