Compiler does not support BigDecimal/BigInteger in compiled expression path

[lukaszlenart]

Problem

The expression compiler generates Java source code (via javassist) that uses raw Java operators (+, -, *, /, ~, <<, >>) for arithmetic and bitwise operations. This works for primitives but fails for BigDecimal and BigInteger objects, since Java does not support operator overloading.

Additionally, there are minor type-handling gaps for Float widening, instanceof compilation, and method calls on auto-boxed primitives.

Documented by

PR #543 added 11 @Disabled dual-mode tests that document these divergences between interpreted and compiled paths:

Category	Tests	Root cause
BigDecimal arithmetic (-1b, +1b, 2*2.0b, etc.)	8	Operators can't apply to BigDecimal objects
BigInteger arithmetic (-1h, +1H, 2h*2, etc.)	7	Operators can't apply to BigInteger objects
BigInteger bitwise (~1h, 5h<<2, 5h>>2)	3	Same fundamental issue
Float widening (5f-2F)	1	coerceToNumeric doesn't distinguish Float from Double
instanceof (5 instanceof java.lang.Integer)	2	Generated source has syntax errors
Method on auto-boxed primitive (index.toString())	1	Missing boxing in generated source

Suggested approach

BigDecimal/BigInteger (18 tests) — main effort

1. NumericExpression.toGetSourceString() should detect when getterClass is BigDecimal/BigInteger and generate ognl.OgnlOps.add(a, b) instead of a + b
2. NumericExpression.coerceToNumeric() line 85 strips type info from ASTConst children via value.toString() — needs to preserve BigDecimal/BigInteger construction
3. ASTNegate.toGetSourceString() should generate ognl.OgnlOps.negate(a) instead of -a
4. ASTBitNegate similarly needs ognl.OgnlOps.bitNegate(a)
5. ASTConst.toGetSourceString() needs to emit new java.math.BigDecimal("value") / new java.math.BigInteger("value")

The OgnlOps methods (add, subtract, multiply, divide, negate, bitNegate, remainder) already handle BigDecimal/BigInteger correctly — they're what the interpreted path uses.

Float widening (1 test)

OgnlRuntime.getNumericLiteral() likely returns the same suffix for Float and Double. Needs a Float-specific "f" literal path in coerceToNumeric.

instanceof (2 tests)

ASTInstanceof.toGetSourceString() generates source that javassist can't parse ("missing member name"). Investigate generated output and fix syntax, or delegate to a runtime helper.

Method on auto-boxed primitive (1 test)

ASTMethod compilation for index.toString() fails because index returns int (primitive) and the compiler calls .toString() without boxing. Needs Integer.valueOf(...) wrapping in generated source.

References

• Issue Unit tests don't work if expressions are not compiled  #18 (original report)
• PR fix: dual-mode test coverage and Long/Float literal compilation #543 (dual-mode test coverage with @Disabled tests)