l1a/l1b: PASS; L2: FAIL (LREM unimplemented)

[opencode-agent]

Spec Compliance Report for lrem

Compiler	Status	Issues
l1a	✅ PASS	None
l1b	✅ PASS	None
l2	❌ FAIL	LREM (and all long arithmetic ops) not implemented

---

l1a (X86BytecodeVisitor.java:3371)

32-bit mode: Delegates to MathSupport.lrem() which explicitly checks den == 0 and throws ArithmeticException. Handles Long.MIN_VALUE % -1 edge case correctly per spec identity (a/b)*b + (a%b) == a.

64-bit mode: Uses x86 IDIV directly. Hardware raises #DE on divide-by-zero; VM catches EX_DIV0 and throws ArithmeticException (VmThread.java:1391-1393). Operand stack effect correct (pops 2 category-2, pushes 1 category-2).

---

l1b (X86BytecodeVisitor.java:4212)

Identical implementation to l1a with same correctness. Only difference is counter instrumentation.

---

L2 (GenericX86CodeGenerator.java, L2ByteCodeSupportChecker.java)

Complete failure: All 16 LREM case labels (and all other long arithmetic ops: LDIV, LMUL, LADD, LSUB, LAND, LOR, LXOR, LSHL, LSHR, LUSHR, LNEG, LCMP) throw IllegalArgumentException("Unknown operation"). L2ByteCodeSupportChecker doesn't reject lrem, so methods containing it pass canCompile() but fail at code generation.

Systemic gap: The entire long/double arithmetic category is unimplemented in L2.

Closes #350

opencode session  |  github run