F2I/F2L fix: NaN→0, +Inf→MAX_VALUE

[opencode-agent]

The fix is complete and builds successfully. Here's a summary of all changes:

Changes Made

6 files modified (5 source + 1 test):

1. core/src/core/org/jnode/vm/x86/compiler/l1a/IntItem.java — popFromFPU() for F2I/D2I in L1A compiler. Emits assembly guards before FISTP32:

• Saves float copy to [reg+disp-4] before conversion
• Checks result for Integer.MIN_VALUE (indefinite integer)
• For NaN → returns 0, +Inf → returns MAX_VALUE, -Inf → keeps MIN_VALUE
• For positive overflow → returns MAX_VALUE

2. core/src/core/org/jnode/vm/x86/compiler/l1a/LongItem.java — Same pattern for F2L/D2L with FISTP64, checking both 32-bit halves of the 64-bit result against Long.MIN_VALUE.

3. core/src/core/org/jnode/vm/x86/compiler/l1b/IntItem.java — Same fix for L1B IntItem.

4. core/src/core/org/jnode/vm/x86/compiler/l1b/LongItem.java — Same fix for L1B LongItem.

5. core/src/core/org/jnode/vm/x86/compiler/l2/GenericX86CodeGenerator.java — Fix for all 4 F2I cases (reg→reg, reg→mem, mem→reg, mem→mem) in the L2 (IR-based) compiler.

6. core/src/test/org/jnode/test/bugs/ConversionTest.java — Test with 18 assertions covering NaN, ±Inf, normal values, zero, overflow, and boundary conditions.

Fix Strategy

Each FISTP instruction is preceded by saving a copy of the FPU value (using FSTP+FLD), followed by a post-instruction fixup that checks the result against the "indefinite integer" pattern and corrects it per JLS §5.1.3.

Closes #456

opencode session  |  github run

[LSantha]

JNode fails to boot.