# conversion subject (Java -> SLeeLa) `conversion.model` is the declarative model of the laws under which Java source becomes functionally equivalent SLeeLa source: types, boxing, numeric promotion, operator lowering, narrowing casts, compound assignment, string conversion, library result types, and the constructs refused (fail closed). It is consumed by `sleelactl/convert` (C lexer + model loader, C++ translator), registered in sleelactl as `java-convert`, and by `lib/java/converter` (a SLeeLa program; `JConvModel.sleela` reads the `ret`, `field`, `binop int` shift, division and remainder rows, the `cast long int` row (the 32-bit wrap) and the `tostr` rows; its output stays at syntax 3.1). Unlike math/physics it is not compiled into the `sleela` runtime: it governs a tool, and its executable half is `lib/java/JavaOperands.sleela`. Rules: `.kiro/steering/java-to-sleela-conversion.md`. ## Revisions | Revision | Change | |---|---| | 1.0.0 | Types, boxing, promotion, operators, casts, compound assignment, string conversion, 45 hand `ret` / `field` rows, refusals | | 1.3.0 (2026-11-13) | Law rows: `tostr reference` (string conversion of an object through `JavaFormat.objectText`), section 9c `chartext` (library parameters that take a char as text: `StringBuilder`/`StringBuffer` `append`, `insert`). Refusal rows `boxed`, `object-text`. | | 1.2.0 (2026-11-13) | Refusal rows `init-block` (static initializer blocks), `member-annotation`, `object-method` (`equals`/`hashCode`/`getClass` on a user type that does not declare them). No law changed. | | 1.1.0 (2026-11-13) | Section **10b**: 8,922 `ret` rows reflected from the public API of OpenJDK 25.0.2 (`lib/java/jdk-25/api/jdk-signatures.jsonl`, `jdk-api.jsonl`) by `generate_ret_rows.py`, for the procedural core of `java.base` (`java.lang`, `java.util`, `java.io`, `java.math`, `java.time`, `java.text`, `java.net`, `java.nio`, ...; the list is `SCOPE` in the script). Hand rows win. | ## Section 10b: reflected result types A row is written only when it is unambiguous: every overload of the member, declared or inherited (most derived declaration first), on every JDK type with that simple name, agrees on a model type (a primitive, `String`, or a top-level type by simple name); overloads that always return the type of their first argument are written `@arg0`. Type variables, arrays and nested types are left out, so such a result stays unknown and both converters keep failing closed on it (`sleelactl`) or leave it as written (`lib/java/converter`). A `void` overload takes no part in the agreement, because a call whose value is used cannot be the `void` one. ```sh python3 impl/subjects/conversion/generate_ret_rows.py # rewrite section 10b python3 impl/subjects/conversion/generate_ret_rows.py --check # exit 1 when out of date ``` Measured with 1.3.0 (2026-11-13): `tests/javaconv_differential_test.sh` 32 of 32 (22 programs, 9 refusals, stale-output check); the JDK 25.0.2 `java.base` subset of 884 files converts 420 (see `JAVA.CONVERSION.STATUS.md`). Measured with 1.1.0 (2026-11-13): `tests/javaconv_differential_test.sh` 14 of 14 pass, unchanged from 1.0.0. On a 17-file JDK sample `java-convert` gets further in two files (their first refusal moves from `[untyped]` to `[native]`); it still converts 2 of 17, because it refuses a file at its first unsupported construct.