Constructor and factory defaults behind a conditional are recorded as the else-branch constant

Crate: ephervex-options (generated data) and tools/minecraft-options

Commit: 2b2af1e4f8166d02e4732d19e040a53745d97a51

Behavior

field_initial_values takes the single instruction before each putfield as the field's initial value, and analyse_option_window takes the constant before the last Integer/Double.valueOf. Neither checks whether that instruction is the else-arm of a conditional, so cond ? A : B is recorded as B with default_confidence: high. A later assignment to the same field silently replaces an earlier one.

• syncWrites = Util.getPlatform() == Util.OS.WINDOWS compiles to if_acmpne; iconst_1; goto; iconst_0; putfield and is recorded as false. The catalog ships sync_writes with "default": false.
• renderDistance = is64Bit ? 12 : 8 is recorded as 8 for 1.12.2 through 1.20.4, and simulationDistance as 8 for 1.18 through 1.20.4. The 64-bit default is 12. The jump to 12 at 1.20.5 only reflects the code shape.

Expected behavior

A constant is accepted as a default only when its putfield or valueOf is not a branch target and no jump separates it from the start of the expression. Conditional or platform-dependent defaults are left unproven, and a field assigned more than once is not given a default from the last write.

Impact

The shipped catalog presents a platform-dependent setting with a fixed default, and older-release analyzer output claims a wrong default as proven. Catalog defaults for render and simulation distance are currently correct only because they come from the canonical release.

Locations

• tools/minecraft-options/analyze_options.py field_initial_values (1597-1630), resolve_default (1633-1668), analyze_jar (1928-1942)
• tools/minecraft-options/analyze_options.py analyse_option_window default tail scan (1336-1350)
• crates/options/data/minecraft-preferences.json descriptor sync_writes

Validation and limitations

Verified with javap on 1.12.2, 1.16.5, 1.20.1, and 26.1 Options. A scan of every cached JAR found that the putfield offsets that are branch targets with a constant predecessor are exactly these fields.