fixes
This commit is contained in:
@@ -11,107 +11,6 @@
|
||||
* A **single canonical layout/decoder/spec** is used across compiler/linker/verifier/VM.
|
||||
* Any invalid program fails with clear diagnostics, not panics.
|
||||
|
||||
**Estimation scale:**
|
||||
|
||||
* **1 point** = small / straightforward
|
||||
* **3 points** = medium
|
||||
* **5 points** = large
|
||||
* Any work that feels bigger must be broken down into multiple PRs (≤ 5 points each).
|
||||
|
||||
---
|
||||
|
||||
## Phase 1 — Single Source of Truth: Bytecode Spec + Decoder (Highest ROI)
|
||||
|
||||
### PR-01 (3 pts) — Move OpcodeSpec to `prometeu-bytecode` and make it authoritative
|
||||
|
||||
**Briefing**
|
||||
|
||||
Today opcode metadata (imm sizes, stack effects, branch-ness, terminators) is duplicated and/or inconsistent across crates. This creates a perpetual maintenance nightmare.
|
||||
|
||||
**Target**
|
||||
|
||||
Create one authoritative opcode spec in **`prometeu-bytecode`** and delete/replace all “local” opcode knowledge.
|
||||
|
||||
**Scope**
|
||||
|
||||
* Create `prometeu-bytecode::opcode_spec` containing:
|
||||
|
||||
* `imm_bytes`
|
||||
* `pops`, `pushes` (stack effect)
|
||||
* `is_branch`, `is_terminator`
|
||||
* optional: `name`, `category`
|
||||
* Update callers to import from `prometeu-bytecode`.
|
||||
|
||||
**Requirements Checklist**
|
||||
|
||||
* [ ] There is exactly one canonical `OpcodeSpec` source.
|
||||
* [ ] All crates compile against that source.
|
||||
* [ ] No hardcoded operand sizes remain outside the spec.
|
||||
|
||||
**Completion Tests**
|
||||
|
||||
* [ ] Unit test enumerating all opcodes validates:
|
||||
|
||||
* every opcode has a spec
|
||||
* `imm_bytes` is defined
|
||||
|
||||
---
|
||||
|
||||
### PR-02 (5 pts) — Introduce canonical decoder in `prometeu-bytecode` and migrate VM to it
|
||||
|
||||
**Briefing**
|
||||
|
||||
The VM currently has its own decoder. The linker and other tools decode manually. This must be centralized.
|
||||
|
||||
**Target**
|
||||
|
||||
Add a single canonical decoder in `prometeu-bytecode` that produces typed decoded instructions.
|
||||
|
||||
**Scope**
|
||||
|
||||
* Add `prometeu-bytecode::decoder`:
|
||||
|
||||
* `decode_next(pc, bytes) -> DecodedInstr`
|
||||
* includes: opcode, pc, next_pc, raw immediate bytes slice
|
||||
* helpers: `imm_u8/u16/u32/i32/i64/f64` with size validation
|
||||
* Migrate VM to use `prometeu-bytecode::decoder`.
|
||||
|
||||
**Requirements Checklist**
|
||||
|
||||
* [ ] VM no longer has a bespoke decoder.
|
||||
* [ ] No slicing-based immediate parsing in VM core paths.
|
||||
* [ ] Decoder validates immediate sizes and fails deterministically.
|
||||
|
||||
**Completion Tests**
|
||||
|
||||
* [ ] Decoder unit tests for representative opcodes with each immediate size.
|
||||
* [ ] Roundtrip test: encode→decode (table-driven; property test optional).
|
||||
|
||||
---
|
||||
|
||||
### PR-03 (3 pts) — Delete/neutralize `abi::operand_size` duplication
|
||||
|
||||
**Briefing**
|
||||
|
||||
`prometeu-bytecode/src/abi.rs` provides partial operand sizing that can drift from the canonical spec.
|
||||
|
||||
**Target**
|
||||
|
||||
Make all operand sizing derived from the opcode spec.
|
||||
|
||||
**Scope**
|
||||
|
||||
* Replace `operand_size()` with `OpcodeSpec::imm_bytes`.
|
||||
* Remove or restrict legacy APIs that leak duplication.
|
||||
|
||||
**Requirements Checklist**
|
||||
|
||||
* [ ] There is no second operand-size table.
|
||||
|
||||
**Completion Tests**
|
||||
|
||||
* [ ] Test ensuring `operand_size()` (if retained) matches spec for all opcodes.
|
||||
|
||||
---
|
||||
|
||||
## Phase 2 — Canonical Layout + Verifier Contract (JVM-like Control Flow)
|
||||
|
||||
Reference in New Issue
Block a user