1
0
Fork 0
mirror of https://github.com/imjasonh/nescript synced 2026-07-08 08:55:38 +00:00
nescript/docs/future-work.md

525 lines
24 KiB
Markdown
Raw Normal View History

# Future Work
This document catalogs known gaps, incomplete features, and planned improvements
in the NEScript compiler. Items are organized by priority and area.
---
## 1. IR-Based Code Generation
Remove the legacy AST codegen — IR path is canonical now The `--use-ast` path through `src/codegen/mod.rs` was a strictly inferior subset of the IR codegen. Building every example with `--use-ast` through the jsnes harness: - `arrays_and_functions` — fully black (array init + function return values + OAM-in-loop all broken) - `structs_enums_for` — fully black (struct literal is a no-op, all fields stay at 0) - `inline_asm_demo` — fully black - `bitwise_ops`, `loop_break_continue` — below sprite floors (static `next_oam_slot` bug B) - `match_demo` — panics at compile time with `branch offset 153 out of range` (AST's if/else-chain desugaring of `match` emits short branches that can't reach the far arms in a multi-arm match) Six of fourteen examples are non-functional under `--use-ast`. The other eight happen to fall inside the subset AST handles (no arrays, no structs, no function return values, no multi-sprite loops, no long match chains). `docs/future-work.md` already listed "Once working, delete the AST-based codegen entirely" as the intended direction. It's working, so this commit does the deletion. What's removed: - The `CodeGen` struct, its impl block, and every helper in `src/codegen/mod.rs` (the AST codegen body) — ~1150 lines. The file is now a module header that re-exports `IrCodeGen`. - `src/codegen/tests.rs` — 15 AST-specific instruction-pattern tests. Every feature they covered has an equivalent test in `src/codegen/ir_codegen.rs::{tests,more_tests}` already. - The `--use-ast` CLI flag and its branch in `src/main.rs`. - `compile_with_ir_codegen` in `tests/integration_test.rs` — `compile()` now does what it did, so they merged. All 40 integration tests go through the IR path. - Outdated sections in `docs/future-work.md` that described the IR codegen as "not yet implemented" and listed AST codegen gaps as priority work. What's kept: - `src/codegen/ir_codegen.rs` — the real codegen. - `src/codegen/peephole.rs` — post-codegen cleanup pass, now run unconditionally from `main.rs`. Test plan: - `cargo test --release` — 313 unit + 37 integration tests pass (was 328 + 37; the 15 dropped are the deleted AST-specific tests). - `cargo fmt --check` clean. - `cargo clippy --release --all-targets -- -D warnings` clean. - `node tests/emulator/run_examples.mjs` — 14/14 ROMs render above their per-example nonBlack floors. - The one tightening: `sprite_resolution_uses_tile_index` was asserting on the old static-slot encoding (`A9 01 8D 01 02`). Updated to the cursor-based form (`A9 01 99 01 02`, i.e. STA AbsoluteY). Net diff: 1581 deletions, 62 insertions. https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
2026-04-12 20:37:59 +00:00
**Status**: Complete. The AST → IR lowering, optimizer, and
`src/codegen/ir_codegen.rs` all work end-to-end; the legacy AST
codegen has been removed. See "Recently completed" below.
---
## 3. Sprite Name Resolution
**Status**: `draw SpriteName at: (x, y)` parses the sprite name but codegen
ignores it. All draws use OAM slot 0 with CHR tile index 0 (the built-in smiley).
**What's needed**:
- Track a mapping from sprite name → CHR tile index in the linker
- When a `sprite` declaration provides inline CHR data or `@chr("file.png")`,
write that data to the CHR ROM at a known tile index
- In codegen, look up the sprite name to get the tile index and write it to
OAM byte 1 (tile number)
- Support multiple OAM slots: track a `next_oam_slot` counter per frame,
allocate slots 0..63 as draws are emitted, warn if >64
---
## 4. Multi-OAM Sprite Support
**Status**: Every `draw` writes to the same OAM bytes ($0200-$0203). Only one
sprite is visible at a time.
**What's needed**:
- Frame-level OAM slot allocator: each `draw` gets the next available 4-byte slot
- Counter reset at the start of each frame handler
- Warn at compile time if a frame handler has >64 static draws
- Runtime: clear unused OAM slots to Y=$FE (off-screen) at frame start
---
## 5. State Machine Dispatch
**Status**: The codegen only generates code for the `start` state. Other states
are parsed and analyzed but their frame handlers are never called.
**What's needed**:
- A `current_state` zero-page variable holding the active state index
- A dispatch table at the start of the main loop: load `current_state`, branch to
the corresponding state's frame handler
- `transition StateName` writes the new state index to `current_state`
- Generate `on_enter` call on transition, `on_exit` call before leaving
---
## 6. Include Directive
**Status**: The `include` keyword is lexed (`KwInclude`) but not parsed or
implemented.
**What's needed**:
- Parser: `include "path.ne"` at top level
- File resolution: relative to the including file's directory
- Circular include detection (track include stack)
- Merge included declarations into the main `Program` AST
- Span tracking: included files need their own `file_id` for error messages
---
## 7. Debug Mode
**Status**: `--debug` flag is accepted by the CLI but has no effect.
**What's needed**:
- `debug.log(expr, ...)`: write values to emulator debug port ($4800)
- `debug.assert(expr)`: emit runtime check, halt on failure
- `debug.overlay(x, y, text)`: render text to a reserved nametable region
- Array bounds checking in debug mode (compare index against array size)
- Frame overrun detection: count cycles per frame, warn if approaching vblank
- All debug code stripped in release mode (already designed: `DebugLog`/`DebugAssert`
in the `Statement` enum are defined in the spec but not yet in the AST)
---
## 8. Scroll Hardware Writes
**Status**: `scroll(x, y)` is parsed but produces no output.
**What's needed**:
- Write X scroll value to PPU register $2005
- Write Y scroll value to PPU register $2005 (second write)
- Must happen during vblank (inside NMI handler or after `wait_frame`)
- Split-screen scroll requires MMC3 scanline IRQ (`on scanline`)
---
## 9. Asset Pipeline Completion
**Status**: PNG → CHR conversion exists (`src/assets/chr.rs`) but is never called
from the compilation pipeline.
### 9a. Wire `@chr("file.png")` to actual PNG loading
- When a sprite/background declares `chr: @chr("path.png")`, call `png_to_chr()`
during compilation
- Resolve the path relative to the source file
- Store resulting CHR data in the ROM's CHR section at a known tile index
### 9b. Wire `@binary("file.bin")` to raw file inclusion
- Read the file as raw bytes and include in CHR or PRG ROM
### 9c. Palette extraction from PNG
- `@palette("file.png")`: analyze image colors, map to nearest NES palette entries
- Already have `nearest_nes_color()` in `src/assets/palette.rs`
### 9d. Nametable conversion
- Full 256×240 PNG → 960-byte nametable + 64-byte attribute table
- Tile deduplication (max 256 unique tiles per pattern table)
- Error if >256 unique tiles
---
## 10. Error Message Polish
**Status**: Errors work and render with ariadne, but many error paths use generic
messages.
### Unused error codes
These are defined in `ErrorCode` but never emitted:
- `E0202` — invalid cast
- `E0203` — invalid operation for type
- `E0301` — zero-page overflow
- `E0403` — unreachable state
- `E0505` — multiple start declarations
- `W0101` — expensive multiply/divide operation
- `W0102` — loop without break or wait_frame
- `W0103` — unused variable
- `W0104` — unreachable code after return/break/transition
### Missing validations
- No error for assigning to a `const`
- No error for `break`/`continue` outside a loop
- No warning for variables declared but never read
- No error for `return` with wrong type vs function signature
- No error for calling a function with wrong argument count/types
---
## 11. Scanline IRQ (MMC3)
**Status**: `on scanline(N)` is in the spec and `on_scanline` field exists in
`StateDecl`, but parsing and codegen are not implemented.
**What's needed**:
- Parser: `on scanline(N) { ... }` event handler in state bodies
- MMC3 IRQ setup: write scanline counter to $C000/$C001/$E000/$E001
- IRQ handler generation: branch to the scanline handler code
- Only valid with `mapper: MMC3`
---
## 12. Audio
audio: complete the subsystem — asset pipeline, user decls, tracker-style driver The audio subsystem was a sketch: `play name` / `start_music name` / `stop_music` parsed, lowered, and emitted a few hardcoded register writes from a builtin name table. No user-declared effects, no per-frame envelope, no note streams, no real engine. This flesh-out brings audio up to the quality bar of the rest of the compiler (sprites, palettes, bank switching, scanline IRQ, etc.) with a full data-driven pipeline: ## Asset pipeline (new `src/assets/audio.rs`) - `sfx Name { duty, pitch, volume }` blocks compile into per-frame pulse-1 envelopes. Pitch/volume arrays must match in length; each entry is one NMI's worth of `$4000` data. - `music Name { duty, volume, repeat, notes }` blocks compile into flat `(pitch, duration)` streams for pulse 2. Pitch 0 is a rest, 1-60 indexes a builtin period table covering C1-B5. - `resolve_sfx` / `resolve_music` walk the program for `play` / `start_music` references and append builtin fallbacks for any name that isn't user-declared — so `play coin` still works without a `sfx Coin { ... }` block. - Builtin effects (coin, jump, hit, click, cancel, shoot, step) and tracks (theme, battle, victory, gameover) synthesize through the same compile path as user decls — one data model, one driver. ## Runtime engine (`src/runtime/mod.rs`) - `gen_audio_tick()` walks both channels every NMI: reads one envelope byte through `(ZP_SFX_PTR),Y` -> writes `$4000`, advances ptr, mutes on zero sentinel. Music decrements the note counter, advances to the next `(pitch, dur)` pair on zero, looks up the period through `(__period_table),Y`, loops on `0xFF 0xFF`. - `gen_period_table()` emits a 60-entry equal-tempered table (A4 = 440 Hz, NTSC 1.789773 MHz CPU clock) with length-counter load bits pre-baked into each high byte. - `gen_data_block()` emits a label + raw-bytes pseudo pair so user sfx/music data can be spliced into PRG with regular labels that the two-pass assembler resolves. - New ZP layout: `$05/$06` music loop base, `$07` music state (duty/volume/loop/active), `$0C-$0F` sfx and music pointers. ## IR codegen (`src/codegen/ir_codegen.rs`) - `with_audio(sfx, music)` registers compile-time trigger constants per blob name. - `gen_play_sfx` emits: write period to `$4002`/`$4003`, load envelope pointer into `ZP_SFX_PTR` via SymbolLo/SymbolHi of `__sfx_<name>`, mark the sfx counter active. - `gen_start_music` stamps the header byte into `ZP_MUSIC_STATE` with the active bit OR'd in, seeds both ptr and loop base from `__music_<name>`, primes the duration counter. - `gen_stop_music` mutes pulse 2 and clears state. ## Linker (`src/linker/mod.rs`) - New `link_with_all_assets(user_code, sprites, sfx, music)` path that splices driver body, period table, and each sfx/music data blob into PRG — all guarded on the `__audio_used` marker so silent programs pay zero ROM cost. ## Assembler (`src/asm/opcodes.rs`, `src/asm/mod.rs`) - New `AddressingMode::Bytes(Vec<u8>)` variant for raw-data pseudo-instructions. `NOP+Bytes(v)` emits the payload verbatim, letting the linker splice ROM data tables into a code section and still have `Label` / `SymbolLo` / `SymbolHi` fixups resolve correctly in the same assembly pass. ## Analyzer - `play` / `start_music` now validate the name against user decls and builtin tables. Unknown names emit E0505 with a helpful list of builtins — previously a typo would silently compile to no-op. ## Parser - New `sfx_decl` / `music_decl` grammar with property-style configuration. Strict validation: duty 0-3, volume 0-15, pitch arrays must match volume length, music notes must come in pairs, pitch 0-60, duration ≥ 1. ## Tests +170 new tests across every layer: - `src/assets/audio.rs`: 17 tests (compile, resolve, builtins, shadowing, label sanitation, nested reference walks) - `src/parser/tests.rs`: 13 tests (valid/invalid sfx + music declarations, property validation, play/start_music/stop_music) - `src/analyzer/tests.rs`: 7 tests (builtin acceptance, user decl acceptance, unknown-name rejection) - `src/runtime/tests.rs`: 10 tests (audio tick labels, RTS end, $4000 write, $4004 mute, period table assembly, A4 = 440 Hz, length counter bits, data block verbatim emit) - `src/linker/tests.rs`: 4 tests (sfx/music blob placement, pointer resolution, elision when unused) - `src/codegen/ir_codegen.rs`: rewrote the 4 existing audio tests to match the new data-driven contract - `tests/integration_test.rs`: 4 end-to-end tests including a user-declared `sfx` + `music` program that verifies bytes land in PRG ROM at the right addresses ## Docs - New Audio section in `docs/language-guide.md` with syntax reference, builtin tables, and an explanation of how the driver works at compile and run time. - `docs/architecture.md` updated to reflect the real audio pipeline instead of the old "audio import stubs" stub. - `docs/future-work.md` moves audio from "status: minimal" to "status: full subsystem" with a narrower list of follow-up work (triangle/noise/DMC channels, NSF/FTM imports, richer envelopes). - `examples/audio_demo.ne` rewritten to showcase user-declared `sfx LongCoin`, `sfx Zap`, `music Theme`, still demonstrating builtin fallback via `play coin`. Total: 424 tests passing (381 unit + 43 integration), clippy clean, fmt clean, all 19 examples compile. https://claude.ai/code/session_015WfaDttE3DpWn9rpyfpQd8
2026-04-13 01:10:21 +00:00
**Status**: A full data-driven audio subsystem is in place. User programs
can declare `sfx Name { duty, pitch, volume }` blocks (frame-accurate
pulse-1 envelopes) and `music Name { duty, volume, repeat, notes }`
blocks (pulse-2 note streams with rests and looping). The resolver
compiles these into ROM-ready byte tables; the IR codegen emits
trigger sequences that load pointers into ZP; the runtime NMI tick
walks the envelope and note stream each frame, indexing into a
builtin 60-note period table. Builtin effects (`coin`, `jump`, `hit`,
`click`, `cancel`, `shoot`, `step`) and tracks (`theme`, `battle`,
`victory`, `gameover`) are synthesized from the same data path so
programs that don't declare their own audio still make sound.
Programs that touch no audio pay zero ROM or cycle cost — the whole
subsystem elides when the `__audio_used` marker is absent.
**Still TODO for richer audio**:
- Triangle/noise/DMC channels (currently only pulse 1 and 2 are used)
- Multi-channel tracker playback (one `notes` list per channel)
- `@sfx("file.nsf")` / `@music("file.ftm")` asset directives
compiler: audio driver, u16 arithmetic, multi-scanline, slot recycling Five language features and optimizations from the planned-work backlog: - **Minimal audio driver**: `play`/`start_music`/`stop_music` now generate APU pulse-1/pulse-2 writes from a builtin SFX/music name table, and the NMI handler gains a `JSR __audio_tick` splice (via the linker's `__audio_used` marker lookup) that ages an SFX countdown counter and mutes pulse 1 when the tone expires. Programs that never trigger audio pay zero ROM cost. - **u16 arithmetic and comparisons**: new IR ops `LoadVarHi`, `StoreVarHi`, `Add16`, `Sub16`, and six `Cmp*16` variants. The lowering context tracks variable types via the analyzer's symbol table and routes expressions through the 8-bit or 16-bit path based on operand width. Add16 emits `CLC;ADC;ADC` with carry propagating naturally into the high byte; compares dispatch high-byte-first with a short-circuit low-byte fallback. Fixes a silent miscompile where `big += 1` on a u16 var only incremented the low byte. - **Multi-scanline handlers per state**: `gen_scanline_irq` now dispatches on `(current_state, ZP_SCANLINE_STEP)` and reloads the MMC3 counter with the delta to the next scanline in the same state. `gen_scanline_reload` resets the step counter at the top of each NMI so a state with multiple handlers fires them in ascending line order. Previously only the first handler per state ever fired. - **IR temp slot recycling**: `build_use_counts` pre-scans each function to count per-temp uses; `retire_op_sources` decrements the counts after each op and pushes dead slots back onto `free_slots` for later allocation. `bitwise_ops.ne` used to crash (debug) or miscompile (release) once it hit 128 concurrent temps; with recycling the same function now uses ~4 slots instead of 136. - **INC/DEC peephole fold + improved dead-load elimination**: `fold_inc_dec` collapses `LDA addr; CLC; ADC #1; STA addr` into a single `INC addr` (and the SEC/SBC variant into `DEC addr`), saving 5 bytes and 5 cycles per increment. The fold is suppressed when the next instruction reads carry. `remove_dead_loads` now walks past INC/DEC/STX/STY (which don't touch A) to find the actual next A-use, catching more dead loads. Tests: 331 unit + 39 integration (up from 313 + 37), including new guards for audio, u16, multi-scanline, and slot recycling. https://claude.ai/code/session_01A8qk3gw2jWSzdiXBZPZSFE
2026-04-12 22:21:53 +00:00
- FamiTracker export format parsing
audio: complete the subsystem — asset pipeline, user decls, tracker-style driver The audio subsystem was a sketch: `play name` / `start_music name` / `stop_music` parsed, lowered, and emitted a few hardcoded register writes from a builtin name table. No user-declared effects, no per-frame envelope, no note streams, no real engine. This flesh-out brings audio up to the quality bar of the rest of the compiler (sprites, palettes, bank switching, scanline IRQ, etc.) with a full data-driven pipeline: ## Asset pipeline (new `src/assets/audio.rs`) - `sfx Name { duty, pitch, volume }` blocks compile into per-frame pulse-1 envelopes. Pitch/volume arrays must match in length; each entry is one NMI's worth of `$4000` data. - `music Name { duty, volume, repeat, notes }` blocks compile into flat `(pitch, duration)` streams for pulse 2. Pitch 0 is a rest, 1-60 indexes a builtin period table covering C1-B5. - `resolve_sfx` / `resolve_music` walk the program for `play` / `start_music` references and append builtin fallbacks for any name that isn't user-declared — so `play coin` still works without a `sfx Coin { ... }` block. - Builtin effects (coin, jump, hit, click, cancel, shoot, step) and tracks (theme, battle, victory, gameover) synthesize through the same compile path as user decls — one data model, one driver. ## Runtime engine (`src/runtime/mod.rs`) - `gen_audio_tick()` walks both channels every NMI: reads one envelope byte through `(ZP_SFX_PTR),Y` -> writes `$4000`, advances ptr, mutes on zero sentinel. Music decrements the note counter, advances to the next `(pitch, dur)` pair on zero, looks up the period through `(__period_table),Y`, loops on `0xFF 0xFF`. - `gen_period_table()` emits a 60-entry equal-tempered table (A4 = 440 Hz, NTSC 1.789773 MHz CPU clock) with length-counter load bits pre-baked into each high byte. - `gen_data_block()` emits a label + raw-bytes pseudo pair so user sfx/music data can be spliced into PRG with regular labels that the two-pass assembler resolves. - New ZP layout: `$05/$06` music loop base, `$07` music state (duty/volume/loop/active), `$0C-$0F` sfx and music pointers. ## IR codegen (`src/codegen/ir_codegen.rs`) - `with_audio(sfx, music)` registers compile-time trigger constants per blob name. - `gen_play_sfx` emits: write period to `$4002`/`$4003`, load envelope pointer into `ZP_SFX_PTR` via SymbolLo/SymbolHi of `__sfx_<name>`, mark the sfx counter active. - `gen_start_music` stamps the header byte into `ZP_MUSIC_STATE` with the active bit OR'd in, seeds both ptr and loop base from `__music_<name>`, primes the duration counter. - `gen_stop_music` mutes pulse 2 and clears state. ## Linker (`src/linker/mod.rs`) - New `link_with_all_assets(user_code, sprites, sfx, music)` path that splices driver body, period table, and each sfx/music data blob into PRG — all guarded on the `__audio_used` marker so silent programs pay zero ROM cost. ## Assembler (`src/asm/opcodes.rs`, `src/asm/mod.rs`) - New `AddressingMode::Bytes(Vec<u8>)` variant for raw-data pseudo-instructions. `NOP+Bytes(v)` emits the payload verbatim, letting the linker splice ROM data tables into a code section and still have `Label` / `SymbolLo` / `SymbolHi` fixups resolve correctly in the same assembly pass. ## Analyzer - `play` / `start_music` now validate the name against user decls and builtin tables. Unknown names emit E0505 with a helpful list of builtins — previously a typo would silently compile to no-op. ## Parser - New `sfx_decl` / `music_decl` grammar with property-style configuration. Strict validation: duty 0-3, volume 0-15, pitch arrays must match volume length, music notes must come in pairs, pitch 0-60, duration ≥ 1. ## Tests +170 new tests across every layer: - `src/assets/audio.rs`: 17 tests (compile, resolve, builtins, shadowing, label sanitation, nested reference walks) - `src/parser/tests.rs`: 13 tests (valid/invalid sfx + music declarations, property validation, play/start_music/stop_music) - `src/analyzer/tests.rs`: 7 tests (builtin acceptance, user decl acceptance, unknown-name rejection) - `src/runtime/tests.rs`: 10 tests (audio tick labels, RTS end, $4000 write, $4004 mute, period table assembly, A4 = 440 Hz, length counter bits, data block verbatim emit) - `src/linker/tests.rs`: 4 tests (sfx/music blob placement, pointer resolution, elision when unused) - `src/codegen/ir_codegen.rs`: rewrote the 4 existing audio tests to match the new data-driven contract - `tests/integration_test.rs`: 4 end-to-end tests including a user-declared `sfx` + `music` program that verifies bytes land in PRG ROM at the right addresses ## Docs - New Audio section in `docs/language-guide.md` with syntax reference, builtin tables, and an explanation of how the driver works at compile and run time. - `docs/architecture.md` updated to reflect the real audio pipeline instead of the old "audio import stubs" stub. - `docs/future-work.md` moves audio from "status: minimal" to "status: full subsystem" with a narrower list of follow-up work (triangle/noise/DMC channels, NSF/FTM imports, richer envelopes). - `examples/audio_demo.ne` rewritten to showcase user-declared `sfx LongCoin`, `sfx Zap`, `music Theme`, still demonstrating builtin fallback via `play coin`. Total: 424 tests passing (381 unit + 43 integration), clippy clean, fmt clean, all 19 examples compile. https://claude.ai/code/session_015WfaDttE3DpWn9rpyfpQd8
2026-04-13 01:10:21 +00:00
- Per-note pitch changes within a sfx (currently pitch is latched once)
---
## 13. Language Features (Post-v0.1)
From the spec's "Reserved for Future Versions" section:
| Feature | Description |
|---------|-------------|
| **Structs** | `struct Vec2 { x: u8, y: u8 }` — composite types with known layout |
| **Enums** | `enum Direction { Up, Down, Left, Right }` — mapped to u8 values |
| **Fixed-point** | `fixed8.8` type for sub-pixel movement |
| **Text/HUD** | Font sheet declarations, layout system for scores/health/menus |
| **Metasprites** | Multi-tile sprite groups with relative positioning |
| **Tilemaps** | Declarative level data with collision queries |
| **SRAM/saves** | Persistent storage declarations for battery-backed save data |
| **NES 2.0** | Extended iNES header format |
---
## 14. Inline Assembly
**Status**: `asm { }` blocks are lexed (`KwAsm`, `AsmBody`) but not parsed or
compiled. The lexer has raw-capture mode for asm content.
**What's needed**:
- Parser: capture asm body text, parse `{variable_name}` substitutions
- Codegen: emit raw 6502 instructions with variable address substitution
- Labels: local to the asm block scope
- `raw asm { }` variant with no substitution or safety checks
---
## 15. Compiler Performance
**Status**: Compilation is fast (<100ms for all examples) but has no benchmarks.
**What's needed**:
- `cargo bench` benchmarks for each pipeline phase
- Regression test: compilation must stay under 500ms for any reasonable project
- Profile-guided optimization of hot paths (lexer, parser)
---
## 16. WASM Build Target
**What's needed**:
- Factor out all file I/O behind a trait (`FileSystem` / `VFS`)
- Core pipeline takes `&str` source → `Vec<u8>` ROM with no filesystem access
- Compile the compiler to WASM for a browser-based IDE
- In-browser NES emulator integration for instant preview
---
## 17. Open Design Questions
From the engineering plan:
1. **Inline asm label syntax**: `.label:` (ca65 style) vs `label:` (generic)?
2. **Debug port address**: $4800 is conventional but not universal. Support
multiple debug output methods?
3. **OAM allocation strategy**: Sequential allocation vs priority-based with
automatic sprite cycling for the 8-per-scanline limit?
4. **Error recovery granularity**: How aggressively should the parser recover?
More recovery = more errors per compile, but risk of cascading false errors.
---
## 18. Missing Assignment Operators
`<<=` and `>>=` (shift-assign) are lexed as `ShiftLeftAssign`/`ShiftRightAssign`
tokens but have no corresponding variants in the `AssignOp` AST enum. They can
never appear in parsed code. Adding them requires:
- New `AssignOp::ShiftLeftAssign` and `AssignOp::ShiftRightAssign` variants
- Parser handling in `parse_assign_or_call` (alongside the other compound ops)
- Codegen: load value, shift, store back
---
## 19. Player 2 Controller
`Player::P1` and `Player::P2` are defined in the AST but marked `#[allow(dead_code)]`.
The parser always produces `ButtonRead(None, ...)` — it never parses `p1.button.X`
or `p2.button.X` syntax. The runtime only reads controller 1 ($4016).
**What's needed**:
- Parser: `p1.button.X` and `p2.button.X` syntax producing `Player::P1`/`P2`
- Runtime: read controller 2 from $4017 into a second zero-page byte
- Codegen: select the correct input byte based on player
---
## 20. Register Allocator
The plan describes `src/codegen/regalloc.rs` for managing A/X/Y allocation, but
no register allocator exists. The current codegen uses A for everything and
spills to zero-page $02 for comparisons. A proper allocator would:
- Track A/X/Y liveness across basic blocks
- Use X for array indexing and loop counters
- Use Y as secondary index
- Spill to zero-page temps only when all three are live
---
## Priority Order
### Recently completed (removed from backlog)
These items were documented as future work but have since been implemented:
audio: complete the subsystem — asset pipeline, user decls, tracker-style driver The audio subsystem was a sketch: `play name` / `start_music name` / `stop_music` parsed, lowered, and emitted a few hardcoded register writes from a builtin name table. No user-declared effects, no per-frame envelope, no note streams, no real engine. This flesh-out brings audio up to the quality bar of the rest of the compiler (sprites, palettes, bank switching, scanline IRQ, etc.) with a full data-driven pipeline: ## Asset pipeline (new `src/assets/audio.rs`) - `sfx Name { duty, pitch, volume }` blocks compile into per-frame pulse-1 envelopes. Pitch/volume arrays must match in length; each entry is one NMI's worth of `$4000` data. - `music Name { duty, volume, repeat, notes }` blocks compile into flat `(pitch, duration)` streams for pulse 2. Pitch 0 is a rest, 1-60 indexes a builtin period table covering C1-B5. - `resolve_sfx` / `resolve_music` walk the program for `play` / `start_music` references and append builtin fallbacks for any name that isn't user-declared — so `play coin` still works without a `sfx Coin { ... }` block. - Builtin effects (coin, jump, hit, click, cancel, shoot, step) and tracks (theme, battle, victory, gameover) synthesize through the same compile path as user decls — one data model, one driver. ## Runtime engine (`src/runtime/mod.rs`) - `gen_audio_tick()` walks both channels every NMI: reads one envelope byte through `(ZP_SFX_PTR),Y` -> writes `$4000`, advances ptr, mutes on zero sentinel. Music decrements the note counter, advances to the next `(pitch, dur)` pair on zero, looks up the period through `(__period_table),Y`, loops on `0xFF 0xFF`. - `gen_period_table()` emits a 60-entry equal-tempered table (A4 = 440 Hz, NTSC 1.789773 MHz CPU clock) with length-counter load bits pre-baked into each high byte. - `gen_data_block()` emits a label + raw-bytes pseudo pair so user sfx/music data can be spliced into PRG with regular labels that the two-pass assembler resolves. - New ZP layout: `$05/$06` music loop base, `$07` music state (duty/volume/loop/active), `$0C-$0F` sfx and music pointers. ## IR codegen (`src/codegen/ir_codegen.rs`) - `with_audio(sfx, music)` registers compile-time trigger constants per blob name. - `gen_play_sfx` emits: write period to `$4002`/`$4003`, load envelope pointer into `ZP_SFX_PTR` via SymbolLo/SymbolHi of `__sfx_<name>`, mark the sfx counter active. - `gen_start_music` stamps the header byte into `ZP_MUSIC_STATE` with the active bit OR'd in, seeds both ptr and loop base from `__music_<name>`, primes the duration counter. - `gen_stop_music` mutes pulse 2 and clears state. ## Linker (`src/linker/mod.rs`) - New `link_with_all_assets(user_code, sprites, sfx, music)` path that splices driver body, period table, and each sfx/music data blob into PRG — all guarded on the `__audio_used` marker so silent programs pay zero ROM cost. ## Assembler (`src/asm/opcodes.rs`, `src/asm/mod.rs`) - New `AddressingMode::Bytes(Vec<u8>)` variant for raw-data pseudo-instructions. `NOP+Bytes(v)` emits the payload verbatim, letting the linker splice ROM data tables into a code section and still have `Label` / `SymbolLo` / `SymbolHi` fixups resolve correctly in the same assembly pass. ## Analyzer - `play` / `start_music` now validate the name against user decls and builtin tables. Unknown names emit E0505 with a helpful list of builtins — previously a typo would silently compile to no-op. ## Parser - New `sfx_decl` / `music_decl` grammar with property-style configuration. Strict validation: duty 0-3, volume 0-15, pitch arrays must match volume length, music notes must come in pairs, pitch 0-60, duration ≥ 1. ## Tests +170 new tests across every layer: - `src/assets/audio.rs`: 17 tests (compile, resolve, builtins, shadowing, label sanitation, nested reference walks) - `src/parser/tests.rs`: 13 tests (valid/invalid sfx + music declarations, property validation, play/start_music/stop_music) - `src/analyzer/tests.rs`: 7 tests (builtin acceptance, user decl acceptance, unknown-name rejection) - `src/runtime/tests.rs`: 10 tests (audio tick labels, RTS end, $4000 write, $4004 mute, period table assembly, A4 = 440 Hz, length counter bits, data block verbatim emit) - `src/linker/tests.rs`: 4 tests (sfx/music blob placement, pointer resolution, elision when unused) - `src/codegen/ir_codegen.rs`: rewrote the 4 existing audio tests to match the new data-driven contract - `tests/integration_test.rs`: 4 end-to-end tests including a user-declared `sfx` + `music` program that verifies bytes land in PRG ROM at the right addresses ## Docs - New Audio section in `docs/language-guide.md` with syntax reference, builtin tables, and an explanation of how the driver works at compile and run time. - `docs/architecture.md` updated to reflect the real audio pipeline instead of the old "audio import stubs" stub. - `docs/future-work.md` moves audio from "status: minimal" to "status: full subsystem" with a narrower list of follow-up work (triangle/noise/DMC channels, NSF/FTM imports, richer envelopes). - `examples/audio_demo.ne` rewritten to showcase user-declared `sfx LongCoin`, `sfx Zap`, `music Theme`, still demonstrating builtin fallback via `play coin`. Total: 424 tests passing (381 unit + 43 integration), clippy clean, fmt clean, all 19 examples compile. https://claude.ai/code/session_015WfaDttE3DpWn9rpyfpQd8
2026-04-13 01:10:21 +00:00
- **Full audio subsystem** — `src/runtime/mod.rs::gen_audio_tick`,
`gen_period_table`, and `gen_data_block` implement a frame-walking
pulse driver. `src/assets/audio.rs` compiles user `sfx`/`music`
declarations (and builtins referenced via `play coin` etc.) into
ROM-ready envelope and note-stream byte tables. `IrCodeGen::with_audio`
threads the compile-time trigger constants into `play`/`start_music`,
which emit pointer loads against per-blob labels. The linker splices
driver body, period table, and every blob into PRG gated on the
`__audio_used` marker so silent programs pay no cost. Full
parser/analyzer/codegen/linker/runtime test coverage.
compiler: audio driver, u16 arithmetic, multi-scanline, slot recycling Five language features and optimizations from the planned-work backlog: - **Minimal audio driver**: `play`/`start_music`/`stop_music` now generate APU pulse-1/pulse-2 writes from a builtin SFX/music name table, and the NMI handler gains a `JSR __audio_tick` splice (via the linker's `__audio_used` marker lookup) that ages an SFX countdown counter and mutes pulse 1 when the tone expires. Programs that never trigger audio pay zero ROM cost. - **u16 arithmetic and comparisons**: new IR ops `LoadVarHi`, `StoreVarHi`, `Add16`, `Sub16`, and six `Cmp*16` variants. The lowering context tracks variable types via the analyzer's symbol table and routes expressions through the 8-bit or 16-bit path based on operand width. Add16 emits `CLC;ADC;ADC` with carry propagating naturally into the high byte; compares dispatch high-byte-first with a short-circuit low-byte fallback. Fixes a silent miscompile where `big += 1` on a u16 var only incremented the low byte. - **Multi-scanline handlers per state**: `gen_scanline_irq` now dispatches on `(current_state, ZP_SCANLINE_STEP)` and reloads the MMC3 counter with the delta to the next scanline in the same state. `gen_scanline_reload` resets the step counter at the top of each NMI so a state with multiple handlers fires them in ascending line order. Previously only the first handler per state ever fired. - **IR temp slot recycling**: `build_use_counts` pre-scans each function to count per-temp uses; `retire_op_sources` decrements the counts after each op and pushes dead slots back onto `free_slots` for later allocation. `bitwise_ops.ne` used to crash (debug) or miscompile (release) once it hit 128 concurrent temps; with recycling the same function now uses ~4 slots instead of 136. - **INC/DEC peephole fold + improved dead-load elimination**: `fold_inc_dec` collapses `LDA addr; CLC; ADC #1; STA addr` into a single `INC addr` (and the SEC/SBC variant into `DEC addr`), saving 5 bytes and 5 cycles per increment. The fold is suppressed when the next instruction reads carry. `remove_dead_loads` now walks past INC/DEC/STX/STY (which don't touch A) to find the actual next A-use, catching more dead loads. Tests: 331 unit + 39 integration (up from 313 + 37), including new guards for audio, u16, multi-scanline, and slot recycling. https://claude.ai/code/session_01A8qk3gw2jWSzdiXBZPZSFE
2026-04-12 22:21:53 +00:00
- **u16 arithmetic and comparisons** — new IR ops `LoadVarHi`,
`StoreVarHi`, `Add16`, `Sub16`, `CmpEq16` through `CmpGtEq16`. The
lowering context tracks variable types via the analyzer's symbol
table and routes each expression through the 8-bit or 16-bit path
based on operand width. Initializers, compound assignments, and
comparisons all preserve both bytes. The codegen emits
`CLC;ADC;ADC` for Add16 with carry propagating naturally, and
compare-high-then-compare-low dispatch for the six comparison
variants.
- **Multi-scanline on_scanline per state** — `gen_scanline_irq` now
dispatches on `(current_state, ZP_SCANLINE_STEP)` and reloads the
MMC3 counter with the delta to the next scanline in the same
state. `gen_scanline_reload` resets the step counter at the top
of each NMI so a state with multiple handlers fires them in
ascending line order.
- **IR temp slot recycling** — `build_use_counts` pre-scans each
function to count per-temp uses; `retire_op_sources` decrements
the counts after each op runs and pushes dead slots back onto
`free_slots` for later allocation. Previously, `bitwise_ops.ne`
crashed (debug) or silently miscompiled (release) once it
allocated more than 128 concurrent temps. With recycling the
same function now uses ~4 slots instead of 136.
- **INC/DEC peephole fold** — `fold_inc_dec` collapses
`LDA addr; CLC; ADC #1; STA addr` into a single `INC addr`
(and the SEC/SBC/#1 variant into `DEC addr`). Saves 5 bytes and
5 cycles per increment. The fold is suppressed when the next
instruction is a carry-dependent branch (`BCC`/`BCS`) since
INC/DEC don't update the carry flag.
- **Peephole dead-load elimination across passive ops** — the
old `remove_dead_loads` only dropped an LDA if the very next
instruction was another A-writer. Now it walks past
INC/DEC/STX/STY (which don't touch A) to find the actual
next A-use, catching more dead loads produced by copy
propagation.
- **State machine dispatch** — CMP + BNE + JMP trampoline dispatch table,
`current_state` in ZP $03, on_enter/on_exit handlers as JSR targets
- **Function call codegen** — JSR to function labels, ZP $04-$07 param passing,
RTS for returns
- **Break/continue** — loop_stack with JMP to continue/break labels
- **Return** — evaluate expr to A + RTS
- **Transition** — write state index + JMP to main loop
- **Array indexing** — TAX + LDA/STA with ZeroPageX or AbsoluteX
- **Scroll** — PPU $2005 writes (X then Y)
- **Multiply/divide/modulo** — JSR __multiply/__divide with shift-and-add/restoring division
- **Shift left/right** — ASL A / LSR A
- **Multi-OAM sprites** — sequential slot allocation (0-63), reset per frame
- **Const assignment error** — E0203 for assigning to constants
- **Break outside loop error** — E0203 for break/continue without enclosing loop
- **Math routines wired into linker** — gen_multiply/gen_divide included in ROM
- **Sprite name resolution** — sprite declarations map to CHR tile indices,
draw statements use the correct tile number
- **Inline sprite CHR data** — sprite decls with `chr: [0x..., ...]` work
- **Include directive** — `include "path"` inlines files at parse time,
with circular include detection
- **Shift-assign operators** — `<<=` and `>>=` work in all contexts
- **Player 2 controller** — `p1.button.X` / `p2.button.X` syntax, P2 input
read from $4017 into ZP $08
- **Unused variable warning** — W0103 emits for declared-but-never-read
globals (underscore-prefix silences)
- **Unreachable state warning** — W0104 emits for states not reachable from
the start state via transitions
- **E0502 "did you mean" suggestions** — undefined variable errors include
a suggestion for nearby-named symbols
- **debug.log / debug.assert** — parses into `Statement::DebugLog` /
`Statement::DebugAssert`, codegen emits runtime writes to $4800 when
`--debug` is set, stripped in release mode
docs: scrub remaining stale AST-codegen references Fallout from removing the `--use-ast` flag and the AST codegen. Four references in `docs/future-work.md` and one in `plan.md` were still implying a parallel AST code path existed: - `future-work.md` "--debug CLI flag wired" entry — said the flag threads through `CodeGen::with_debug` (the old AST codegen's builder). Updated to `IrCodeGen::with_debug`. - `future-work.md` IR debug.log/assert entry — had a parenthetical "(same behavior as AST codegen)" that no longer makes sense without an AST codegen to compare against. Dropped. - `future-work.md` inline assembly entry — said "Both IR and AST codegen splice parsed instructions directly into the output stream." There's only the IR codegen now. Updated. - `future-work.md` `on scanline(N)` entry — described scanline handling as "codegen (MMC3 IRQ vector wiring) is still TODO." That wiring has been in place for a while: the IR codegen emits `__irq_user` + `__ir_mmc3_reload` with per-state dispatch, and `examples/scanline_split.ne` and `examples/mmc3_per_state_split.ne` both exercise it in the jsnes smoke test. Rewrote the entry to match reality. - `plan.md` M2 "Compiler phases built" list — had "Codegen from IR (replacing direct AST codegen)." The present-tense parenthetical read as if AST codegen still existed; stripped it. M1's "direct AST → 6502, skip IR for this milestone" line is left as a historical milestone scope description. One intentional mention remains: `future-work.md` "Recently completed" bullet explicitly notes the AST-based path and the `--use-ast` flag were removed. That's the correct tombstone. https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
2026-04-12 20:53:36 +00:00
- **--debug CLI flag wired** — threads through `IrCodeGen::with_debug`
- **IR-based codegen** — `src/codegen/ir_codegen.rs` walks `IrProgram` and
emits 6502 for every IR op: load/store, arithmetic, comparisons, arrays,
calls, draws, input (P1 and P2), scroll, debug.log/assert, state
Remove the legacy AST codegen — IR path is canonical now The `--use-ast` path through `src/codegen/mod.rs` was a strictly inferior subset of the IR codegen. Building every example with `--use-ast` through the jsnes harness: - `arrays_and_functions` — fully black (array init + function return values + OAM-in-loop all broken) - `structs_enums_for` — fully black (struct literal is a no-op, all fields stay at 0) - `inline_asm_demo` — fully black - `bitwise_ops`, `loop_break_continue` — below sprite floors (static `next_oam_slot` bug B) - `match_demo` — panics at compile time with `branch offset 153 out of range` (AST's if/else-chain desugaring of `match` emits short branches that can't reach the far arms in a multi-arm match) Six of fourteen examples are non-functional under `--use-ast`. The other eight happen to fall inside the subset AST handles (no arrays, no structs, no function return values, no multi-sprite loops, no long match chains). `docs/future-work.md` already listed "Once working, delete the AST-based codegen entirely" as the intended direction. It's working, so this commit does the deletion. What's removed: - The `CodeGen` struct, its impl block, and every helper in `src/codegen/mod.rs` (the AST codegen body) — ~1150 lines. The file is now a module header that re-exports `IrCodeGen`. - `src/codegen/tests.rs` — 15 AST-specific instruction-pattern tests. Every feature they covered has an equivalent test in `src/codegen/ir_codegen.rs::{tests,more_tests}` already. - The `--use-ast` CLI flag and its branch in `src/main.rs`. - `compile_with_ir_codegen` in `tests/integration_test.rs` — `compile()` now does what it did, so they merged. All 40 integration tests go through the IR path. - Outdated sections in `docs/future-work.md` that described the IR codegen as "not yet implemented" and listed AST codegen gaps as priority work. What's kept: - `src/codegen/ir_codegen.rs` — the real codegen. - `src/codegen/peephole.rs` — post-codegen cleanup pass, now run unconditionally from `main.rs`. Test plan: - `cargo test --release` — 313 unit + 37 integration tests pass (was 328 + 37; the 15 dropped are the deleted AST-specific tests). - `cargo fmt --check` clean. - `cargo clippy --release --all-targets -- -D warnings` clean. - `node tests/emulator/run_examples.mjs` — 14/14 ROMs render above their per-example nonBlack floors. - The one tightening: `sprite_resolution_uses_tile_index` was asserting on the old static-slot encoding (`A9 01 8D 01 02`). Updated to the cursor-based form (`A9 01 99 01 02`, i.e. STA AbsoluteY). Net diff: 1581 deletions, 62 insertions. https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
2026-04-12 20:37:59 +00:00
dispatch, runtime OAM cursor for looped draws, transitions + on_enter
handlers. It's the only codegen — the legacy AST-based path and the
`--use-ast` flag were removed once the IR pipeline was proven correct
by the jsnes emulator smoke test.
- **IR lowering bug fixes** — `ReadInput` now has a destination temp,
`ButtonRead` uses the proper input temp, logical AND/OR use a new
`emit_move` helper instead of the buggy raw VarId temp storage
- **IR Player 2 controller** — `ReadInput(temp, player)` selects $01
or $08 based on player index
- **IR scroll support** — `scroll(x, y)` lowers to `IrOp::Scroll(x, y)`
which emits two PPU $2005 writes in IR codegen
- **IR debug.log / debug.assert** — new `IrOp::DebugLog(temps)` and
docs: scrub remaining stale AST-codegen references Fallout from removing the `--use-ast` flag and the AST codegen. Four references in `docs/future-work.md` and one in `plan.md` were still implying a parallel AST code path existed: - `future-work.md` "--debug CLI flag wired" entry — said the flag threads through `CodeGen::with_debug` (the old AST codegen's builder). Updated to `IrCodeGen::with_debug`. - `future-work.md` IR debug.log/assert entry — had a parenthetical "(same behavior as AST codegen)" that no longer makes sense without an AST codegen to compare against. Dropped. - `future-work.md` inline assembly entry — said "Both IR and AST codegen splice parsed instructions directly into the output stream." There's only the IR codegen now. Updated. - `future-work.md` `on scanline(N)` entry — described scanline handling as "codegen (MMC3 IRQ vector wiring) is still TODO." That wiring has been in place for a while: the IR codegen emits `__irq_user` + `__ir_mmc3_reload` with per-state dispatch, and `examples/scanline_split.ne` and `examples/mmc3_per_state_split.ne` both exercise it in the jsnes smoke test. Rewrote the entry to match reality. - `plan.md` M2 "Compiler phases built" list — had "Codegen from IR (replacing direct AST codegen)." The present-tense parenthetical read as if AST codegen still existed; stripped it. M1's "direct AST → 6502, skip IR for this milestone" line is left as a historical milestone scope description. One intentional mention remains: `future-work.md` "Recently completed" bullet explicitly notes the AST-based path and the `--use-ast` flag were removed. That's the correct tombstone. https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
2026-04-12 20:53:36 +00:00
`IrOp::DebugAssert(cond)` variants, emitted as $4800 writes in
debug mode and stripped in release
- **Asset pipeline @binary / @chr loading** — `resolve_sprites()` reads
raw binary files and converts PNGs via `png_to_chr()`. Missing files
are silently skipped (documentation-friendly)
- **Call arity validation** — E0203 when `Statement::Call` or
`Expr::Call` has the wrong number of arguments or a mismatched
argument type (uses a `function_signatures` map)
- **Return type validation** — `return value` is type-checked against
the function's declared return type (E0201); returning a value from a
void function emits E0203
- **W0102 loop-without-yield warning** — emitted when a `loop { ... }`
body contains no `break`, `return`, `transition`, or `wait_frame`
- **W0101 expensive mul/div/mod warning** — flags multiply/divide/modulo
with two non-constant operands; literal operands are silent because
the optimizer strength-reduces them
- **W0104 dead-code-after-terminator warning** — statements after
`return`, `break`, `continue`, or `transition` in the same block
emit W0104 with a label pointing at the terminator
- **E0301 RAM overflow** — the zero-page user region is now bounded
above by `$80` (leaving `$80-$FF` for IR temps) and the main RAM
allocator stops at `$0800`; overflow emits E0301 at the declaration
- **E0505 multiple start declarations** — parser rejects a second
`start X` token
docs: scrub remaining stale AST-codegen references Fallout from removing the `--use-ast` flag and the AST codegen. Four references in `docs/future-work.md` and one in `plan.md` were still implying a parallel AST code path existed: - `future-work.md` "--debug CLI flag wired" entry — said the flag threads through `CodeGen::with_debug` (the old AST codegen's builder). Updated to `IrCodeGen::with_debug`. - `future-work.md` IR debug.log/assert entry — had a parenthetical "(same behavior as AST codegen)" that no longer makes sense without an AST codegen to compare against. Dropped. - `future-work.md` inline assembly entry — said "Both IR and AST codegen splice parsed instructions directly into the output stream." There's only the IR codegen now. Updated. - `future-work.md` `on scanline(N)` entry — described scanline handling as "codegen (MMC3 IRQ vector wiring) is still TODO." That wiring has been in place for a while: the IR codegen emits `__irq_user` + `__ir_mmc3_reload` with per-state dispatch, and `examples/scanline_split.ne` and `examples/mmc3_per_state_split.ne` both exercise it in the jsnes smoke test. Rewrote the entry to match reality. - `plan.md` M2 "Compiler phases built" list — had "Codegen from IR (replacing direct AST codegen)." The present-tense parenthetical read as if AST codegen still existed; stripped it. M1's "direct AST → 6502, skip IR for this milestone" line is left as a historical milestone scope description. One intentional mention remains: `future-work.md` "Recently completed" bullet explicitly notes the AST-based path and the `--use-ast` flag were removed. That's the correct tombstone. https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
2026-04-12 20:53:36 +00:00
- **`on scanline(N)` handlers** — `state { on scanline(240) { ... } }`
parses and populates `StateDecl::on_scanline`; analyzer emits E0203
docs: scrub remaining stale AST-codegen references Fallout from removing the `--use-ast` flag and the AST codegen. Four references in `docs/future-work.md` and one in `plan.md` were still implying a parallel AST code path existed: - `future-work.md` "--debug CLI flag wired" entry — said the flag threads through `CodeGen::with_debug` (the old AST codegen's builder). Updated to `IrCodeGen::with_debug`. - `future-work.md` IR debug.log/assert entry — had a parenthetical "(same behavior as AST codegen)" that no longer makes sense without an AST codegen to compare against. Dropped. - `future-work.md` inline assembly entry — said "Both IR and AST codegen splice parsed instructions directly into the output stream." There's only the IR codegen now. Updated. - `future-work.md` `on scanline(N)` entry — described scanline handling as "codegen (MMC3 IRQ vector wiring) is still TODO." That wiring has been in place for a while: the IR codegen emits `__irq_user` + `__ir_mmc3_reload` with per-state dispatch, and `examples/scanline_split.ne` and `examples/mmc3_per_state_split.ne` both exercise it in the jsnes smoke test. Rewrote the entry to match reality. - `plan.md` M2 "Compiler phases built" list — had "Codegen from IR (replacing direct AST codegen)." The present-tense parenthetical read as if AST codegen still existed; stripped it. M1's "direct AST → 6502, skip IR for this milestone" line is left as a historical milestone scope description. One intentional mention remains: `future-work.md` "Recently completed" bullet explicitly notes the AST-based path and the `--use-ast` flag were removed. That's the correct tombstone. https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
2026-04-12 20:53:36 +00:00
if the game isn't using MMC3. The IR codegen now emits the full
MMC3 IRQ vector glue: per-state dispatch in `__irq_user` and a
`__ir_mmc3_reload` helper that picks the right `$C000` latch value
based on `current_state`. See `examples/scanline_split.ne` and
`examples/mmc3_per_state_split.ne`.
- **Inline assembly** — `asm { ... }` blocks. The lexer captures the
body as a raw `AsmBody` token; `src/asm/inline_parser.rs` provides a
minimal 6502 mnemonic parser that handles every addressing mode the
docs: scrub remaining stale AST-codegen references Fallout from removing the `--use-ast` flag and the AST codegen. Four references in `docs/future-work.md` and one in `plan.md` were still implying a parallel AST code path existed: - `future-work.md` "--debug CLI flag wired" entry — said the flag threads through `CodeGen::with_debug` (the old AST codegen's builder). Updated to `IrCodeGen::with_debug`. - `future-work.md` IR debug.log/assert entry — had a parenthetical "(same behavior as AST codegen)" that no longer makes sense without an AST codegen to compare against. Dropped. - `future-work.md` inline assembly entry — said "Both IR and AST codegen splice parsed instructions directly into the output stream." There's only the IR codegen now. Updated. - `future-work.md` `on scanline(N)` entry — described scanline handling as "codegen (MMC3 IRQ vector wiring) is still TODO." That wiring has been in place for a while: the IR codegen emits `__irq_user` + `__ir_mmc3_reload` with per-state dispatch, and `examples/scanline_split.ne` and `examples/mmc3_per_state_split.ne` both exercise it in the jsnes smoke test. Rewrote the entry to match reality. - `plan.md` M2 "Compiler phases built" list — had "Codegen from IR (replacing direct AST codegen)." The present-tense parenthetical read as if AST codegen still existed; stripped it. M1's "direct AST → 6502, skip IR for this milestone" line is left as a historical milestone scope description. One intentional mention remains: `future-work.md` "Recently completed" bullet explicitly notes the AST-based path and the `--use-ast` flag were removed. That's the correct tombstone. https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
2026-04-12 20:53:36 +00:00
codegen emits. The IR codegen splices parsed instructions directly
into the output stream
- **Enum types** — `enum Name { V1, V2, ... }` declares u8 constants
with values equal to declaration order. Variant names are flattened
into the global symbol table
- **Struct types** — `struct Vec2 { x: u8, y: u8 }` with field
access (`pos.x = 5`) and contiguous u8-only field layout. The
analyzer synthesizes per-field VarAllocations so the rest of the
compiler treats field access as ordinary variable access.
- **`for i in start..end { ... }` loops** — half-open range with a
u8 index variable. Desugared in IR lowering to a while loop with
a proper continue-edge block so `break`/`continue` work.
audio: complete the subsystem — asset pipeline, user decls, tracker-style driver The audio subsystem was a sketch: `play name` / `start_music name` / `stop_music` parsed, lowered, and emitted a few hardcoded register writes from a builtin name table. No user-declared effects, no per-frame envelope, no note streams, no real engine. This flesh-out brings audio up to the quality bar of the rest of the compiler (sprites, palettes, bank switching, scanline IRQ, etc.) with a full data-driven pipeline: ## Asset pipeline (new `src/assets/audio.rs`) - `sfx Name { duty, pitch, volume }` blocks compile into per-frame pulse-1 envelopes. Pitch/volume arrays must match in length; each entry is one NMI's worth of `$4000` data. - `music Name { duty, volume, repeat, notes }` blocks compile into flat `(pitch, duration)` streams for pulse 2. Pitch 0 is a rest, 1-60 indexes a builtin period table covering C1-B5. - `resolve_sfx` / `resolve_music` walk the program for `play` / `start_music` references and append builtin fallbacks for any name that isn't user-declared — so `play coin` still works without a `sfx Coin { ... }` block. - Builtin effects (coin, jump, hit, click, cancel, shoot, step) and tracks (theme, battle, victory, gameover) synthesize through the same compile path as user decls — one data model, one driver. ## Runtime engine (`src/runtime/mod.rs`) - `gen_audio_tick()` walks both channels every NMI: reads one envelope byte through `(ZP_SFX_PTR),Y` -> writes `$4000`, advances ptr, mutes on zero sentinel. Music decrements the note counter, advances to the next `(pitch, dur)` pair on zero, looks up the period through `(__period_table),Y`, loops on `0xFF 0xFF`. - `gen_period_table()` emits a 60-entry equal-tempered table (A4 = 440 Hz, NTSC 1.789773 MHz CPU clock) with length-counter load bits pre-baked into each high byte. - `gen_data_block()` emits a label + raw-bytes pseudo pair so user sfx/music data can be spliced into PRG with regular labels that the two-pass assembler resolves. - New ZP layout: `$05/$06` music loop base, `$07` music state (duty/volume/loop/active), `$0C-$0F` sfx and music pointers. ## IR codegen (`src/codegen/ir_codegen.rs`) - `with_audio(sfx, music)` registers compile-time trigger constants per blob name. - `gen_play_sfx` emits: write period to `$4002`/`$4003`, load envelope pointer into `ZP_SFX_PTR` via SymbolLo/SymbolHi of `__sfx_<name>`, mark the sfx counter active. - `gen_start_music` stamps the header byte into `ZP_MUSIC_STATE` with the active bit OR'd in, seeds both ptr and loop base from `__music_<name>`, primes the duration counter. - `gen_stop_music` mutes pulse 2 and clears state. ## Linker (`src/linker/mod.rs`) - New `link_with_all_assets(user_code, sprites, sfx, music)` path that splices driver body, period table, and each sfx/music data blob into PRG — all guarded on the `__audio_used` marker so silent programs pay zero ROM cost. ## Assembler (`src/asm/opcodes.rs`, `src/asm/mod.rs`) - New `AddressingMode::Bytes(Vec<u8>)` variant for raw-data pseudo-instructions. `NOP+Bytes(v)` emits the payload verbatim, letting the linker splice ROM data tables into a code section and still have `Label` / `SymbolLo` / `SymbolHi` fixups resolve correctly in the same assembly pass. ## Analyzer - `play` / `start_music` now validate the name against user decls and builtin tables. Unknown names emit E0505 with a helpful list of builtins — previously a typo would silently compile to no-op. ## Parser - New `sfx_decl` / `music_decl` grammar with property-style configuration. Strict validation: duty 0-3, volume 0-15, pitch arrays must match volume length, music notes must come in pairs, pitch 0-60, duration ≥ 1. ## Tests +170 new tests across every layer: - `src/assets/audio.rs`: 17 tests (compile, resolve, builtins, shadowing, label sanitation, nested reference walks) - `src/parser/tests.rs`: 13 tests (valid/invalid sfx + music declarations, property validation, play/start_music/stop_music) - `src/analyzer/tests.rs`: 7 tests (builtin acceptance, user decl acceptance, unknown-name rejection) - `src/runtime/tests.rs`: 10 tests (audio tick labels, RTS end, $4000 write, $4004 mute, period table assembly, A4 = 440 Hz, length counter bits, data block verbatim emit) - `src/linker/tests.rs`: 4 tests (sfx/music blob placement, pointer resolution, elision when unused) - `src/codegen/ir_codegen.rs`: rewrote the 4 existing audio tests to match the new data-driven contract - `tests/integration_test.rs`: 4 end-to-end tests including a user-declared `sfx` + `music` program that verifies bytes land in PRG ROM at the right addresses ## Docs - New Audio section in `docs/language-guide.md` with syntax reference, builtin tables, and an explanation of how the driver works at compile and run time. - `docs/architecture.md` updated to reflect the real audio pipeline instead of the old "audio import stubs" stub. - `docs/future-work.md` moves audio from "status: minimal" to "status: full subsystem" with a narrower list of follow-up work (triangle/noise/DMC channels, NSF/FTM imports, richer envelopes). - `examples/audio_demo.ne` rewritten to showcase user-declared `sfx LongCoin`, `sfx Zap`, `music Theme`, still demonstrating builtin fallback via `play coin`. Total: 424 tests passing (381 unit + 43 integration), clippy clean, fmt clean, all 19 examples compile. https://claude.ai/code/session_015WfaDttE3DpWn9rpyfpQd8
2026-04-13 01:10:21 +00:00
- **Audio subsystem** — full data-driven pulse driver with
`sfx`/`music` block declarations, builtin effects, and an
NMI-time tick that walks envelope and note-stream tables.
See section 12 above for the full writeup.
- **Constant expression folding** — `const B: u8 = A + 3` evaluates
at compile time and feeds through to variable initializers too.
- **`on scanline` codegen (minimal)** — MMC3 IRQ setup at startup
plus a `__irq_user` dispatcher that saves registers, ACKs via
`$E000`, dispatches on `current_state` to the right scanline
handler, and restores. `__ir_mmc3_reload` helper re-arms the
counter each frame from NMI.
- **Peephole optimizer** — `src/codegen/peephole.rs` runs to fixed
point after codegen. Current passes:
- copy propagation for IR temps (rewrites loads to their source)
- dead-LDA elimination (drops overwritten LDAs)
- redundant STA/LDA pair removal
- LDA-then-STA-same-address removal
- dead-store elimination for IR temp slots (function-wide scan)
- A-value tracking eliminates redundant LDAs (ZP and absolute)
- branch folding: `Bxx L; JMP M; L:``Byy M`
- dead JMP to next label removal
- **`--dump-ir` CLI flag** — prints the lowered IR program after the
optimizer pass for debugging
- **Function-local variables** — IR codegen allocates backing
storage for `var`s declared inside function bodies, using a
per-function RAM range at `$0300+` so nested calls don't clobber
each other.
- **E0502 on assignment to undefined variable** — previously was
silently creating a new variable.
- **Function call ABI fix** — IR codegen was JSRing to `__fn_name`
but functions were defined as `__ir_fn_name`, and param VarIds
weren't in `var_addrs` so callees read temp slots instead of
parameters. Both bugs are now fixed with an integration test
guard.
- **Struct literal syntax** — `Vec2 { x: 100, y: 50 }` in both
variable initializers and assignments. Desugars in lowering to
per-field stores. Restricted to non-condition expression contexts
(if/while/for conditions) to avoid ambiguity with block `{`.
- **Match statement** — `match x { pat => body, _ => default }`
parses to an if/else-if chain at parse time, so no new AST
variant is needed. Supports any expression patterns and an
underscore catch-all.
- **For loops** — `for i in start..end { body }` (half-open range).
Desugars in IR lowering to a while loop with a proper
continue-edge block.
- **Semicolon statement separators** — short statements can share
a line: `a += 1; b += 2`.
- **Inline asm `{var}` substitution** — inside an `asm { ... }`
block, `{name}` is replaced with the hex address of the variable
`name`. The lexer balances nested braces so `{counter}` inside
an asm body is captured correctly.
- **`raw asm { ... }` blocks** — variant of inline asm that skips
`{var}` substitution, passing the body through verbatim.
- **`poke(addr, value)` / `peek(addr)` intrinsics** — hardware
register access without needing an asm block. Compile to a
single LDA/STA against a compile-time-constant address.
- **`--memory-map` CLI flag** — prints a human-readable variable
allocation table showing what's in ZP vs main RAM.
- **`--call-graph` CLI flag** — prints a call-tree view with max
depth reached from each entry point handler.
### Remaining priority order
For someone picking up this codebase, the recommended order of work:
compiler: audio driver, u16 arithmetic, multi-scanline, slot recycling Five language features and optimizations from the planned-work backlog: - **Minimal audio driver**: `play`/`start_music`/`stop_music` now generate APU pulse-1/pulse-2 writes from a builtin SFX/music name table, and the NMI handler gains a `JSR __audio_tick` splice (via the linker's `__audio_used` marker lookup) that ages an SFX countdown counter and mutes pulse 1 when the tone expires. Programs that never trigger audio pay zero ROM cost. - **u16 arithmetic and comparisons**: new IR ops `LoadVarHi`, `StoreVarHi`, `Add16`, `Sub16`, and six `Cmp*16` variants. The lowering context tracks variable types via the analyzer's symbol table and routes expressions through the 8-bit or 16-bit path based on operand width. Add16 emits `CLC;ADC;ADC` with carry propagating naturally into the high byte; compares dispatch high-byte-first with a short-circuit low-byte fallback. Fixes a silent miscompile where `big += 1` on a u16 var only incremented the low byte. - **Multi-scanline handlers per state**: `gen_scanline_irq` now dispatches on `(current_state, ZP_SCANLINE_STEP)` and reloads the MMC3 counter with the delta to the next scanline in the same state. `gen_scanline_reload` resets the step counter at the top of each NMI so a state with multiple handlers fires them in ascending line order. Previously only the first handler per state ever fired. - **IR temp slot recycling**: `build_use_counts` pre-scans each function to count per-temp uses; `retire_op_sources` decrements the counts after each op and pushes dead slots back onto `free_slots` for later allocation. `bitwise_ops.ne` used to crash (debug) or miscompile (release) once it hit 128 concurrent temps; with recycling the same function now uses ~4 slots instead of 136. - **INC/DEC peephole fold + improved dead-load elimination**: `fold_inc_dec` collapses `LDA addr; CLC; ADC #1; STA addr` into a single `INC addr` (and the SEC/SBC variant into `DEC addr`), saving 5 bytes and 5 cycles per increment. The fold is suppressed when the next instruction reads carry. `remove_dead_loads` now walks past INC/DEC/STX/STY (which don't touch A) to find the actual next A-use, catching more dead loads. Tests: 331 unit + 39 integration (up from 313 + 37), including new guards for audio, u16, multi-scanline, and slot recycling. https://claude.ai/code/session_01A8qk3gw2jWSzdiXBZPZSFE
2026-04-12 22:21:53 +00:00
1. **u16 / array / nested struct fields** — u16 *globals* now work
end-to-end (load/store, +/-, comparisons all propagate through
16-bit IR ops). Struct fields and array elements are still u8-only
— the layout machinery needs to grow multi-byte field offsets.
audio: complete the subsystem — asset pipeline, user decls, tracker-style driver The audio subsystem was a sketch: `play name` / `start_music name` / `stop_music` parsed, lowered, and emitted a few hardcoded register writes from a builtin name table. No user-declared effects, no per-frame envelope, no note streams, no real engine. This flesh-out brings audio up to the quality bar of the rest of the compiler (sprites, palettes, bank switching, scanline IRQ, etc.) with a full data-driven pipeline: ## Asset pipeline (new `src/assets/audio.rs`) - `sfx Name { duty, pitch, volume }` blocks compile into per-frame pulse-1 envelopes. Pitch/volume arrays must match in length; each entry is one NMI's worth of `$4000` data. - `music Name { duty, volume, repeat, notes }` blocks compile into flat `(pitch, duration)` streams for pulse 2. Pitch 0 is a rest, 1-60 indexes a builtin period table covering C1-B5. - `resolve_sfx` / `resolve_music` walk the program for `play` / `start_music` references and append builtin fallbacks for any name that isn't user-declared — so `play coin` still works without a `sfx Coin { ... }` block. - Builtin effects (coin, jump, hit, click, cancel, shoot, step) and tracks (theme, battle, victory, gameover) synthesize through the same compile path as user decls — one data model, one driver. ## Runtime engine (`src/runtime/mod.rs`) - `gen_audio_tick()` walks both channels every NMI: reads one envelope byte through `(ZP_SFX_PTR),Y` -> writes `$4000`, advances ptr, mutes on zero sentinel. Music decrements the note counter, advances to the next `(pitch, dur)` pair on zero, looks up the period through `(__period_table),Y`, loops on `0xFF 0xFF`. - `gen_period_table()` emits a 60-entry equal-tempered table (A4 = 440 Hz, NTSC 1.789773 MHz CPU clock) with length-counter load bits pre-baked into each high byte. - `gen_data_block()` emits a label + raw-bytes pseudo pair so user sfx/music data can be spliced into PRG with regular labels that the two-pass assembler resolves. - New ZP layout: `$05/$06` music loop base, `$07` music state (duty/volume/loop/active), `$0C-$0F` sfx and music pointers. ## IR codegen (`src/codegen/ir_codegen.rs`) - `with_audio(sfx, music)` registers compile-time trigger constants per blob name. - `gen_play_sfx` emits: write period to `$4002`/`$4003`, load envelope pointer into `ZP_SFX_PTR` via SymbolLo/SymbolHi of `__sfx_<name>`, mark the sfx counter active. - `gen_start_music` stamps the header byte into `ZP_MUSIC_STATE` with the active bit OR'd in, seeds both ptr and loop base from `__music_<name>`, primes the duration counter. - `gen_stop_music` mutes pulse 2 and clears state. ## Linker (`src/linker/mod.rs`) - New `link_with_all_assets(user_code, sprites, sfx, music)` path that splices driver body, period table, and each sfx/music data blob into PRG — all guarded on the `__audio_used` marker so silent programs pay zero ROM cost. ## Assembler (`src/asm/opcodes.rs`, `src/asm/mod.rs`) - New `AddressingMode::Bytes(Vec<u8>)` variant for raw-data pseudo-instructions. `NOP+Bytes(v)` emits the payload verbatim, letting the linker splice ROM data tables into a code section and still have `Label` / `SymbolLo` / `SymbolHi` fixups resolve correctly in the same assembly pass. ## Analyzer - `play` / `start_music` now validate the name against user decls and builtin tables. Unknown names emit E0505 with a helpful list of builtins — previously a typo would silently compile to no-op. ## Parser - New `sfx_decl` / `music_decl` grammar with property-style configuration. Strict validation: duty 0-3, volume 0-15, pitch arrays must match volume length, music notes must come in pairs, pitch 0-60, duration ≥ 1. ## Tests +170 new tests across every layer: - `src/assets/audio.rs`: 17 tests (compile, resolve, builtins, shadowing, label sanitation, nested reference walks) - `src/parser/tests.rs`: 13 tests (valid/invalid sfx + music declarations, property validation, play/start_music/stop_music) - `src/analyzer/tests.rs`: 7 tests (builtin acceptance, user decl acceptance, unknown-name rejection) - `src/runtime/tests.rs`: 10 tests (audio tick labels, RTS end, $4000 write, $4004 mute, period table assembly, A4 = 440 Hz, length counter bits, data block verbatim emit) - `src/linker/tests.rs`: 4 tests (sfx/music blob placement, pointer resolution, elision when unused) - `src/codegen/ir_codegen.rs`: rewrote the 4 existing audio tests to match the new data-driven contract - `tests/integration_test.rs`: 4 end-to-end tests including a user-declared `sfx` + `music` program that verifies bytes land in PRG ROM at the right addresses ## Docs - New Audio section in `docs/language-guide.md` with syntax reference, builtin tables, and an explanation of how the driver works at compile and run time. - `docs/architecture.md` updated to reflect the real audio pipeline instead of the old "audio import stubs" stub. - `docs/future-work.md` moves audio from "status: minimal" to "status: full subsystem" with a narrower list of follow-up work (triangle/noise/DMC channels, NSF/FTM imports, richer envelopes). - `examples/audio_demo.ne` rewritten to showcase user-declared `sfx LongCoin`, `sfx Zap`, `music Theme`, still demonstrating builtin fallback via `play coin`. Total: 424 tests passing (381 unit + 43 integration), clippy clean, fmt clean, all 19 examples compile. https://claude.ai/code/session_015WfaDttE3DpWn9rpyfpQd8
2026-04-13 01:10:21 +00:00
2. **Triangle / noise / DMC channels** — the current audio engine
plays sfx on pulse 1 and music on pulse 2 with a full
data-driven tracker model (envelope walk, period table,
`(pitch, duration)` note streams, loop-back). Wiring triangle
and noise channels into the same model would unblock richer
multi-part compositions.
compiler: audio driver, u16 arithmetic, multi-scanline, slot recycling Five language features and optimizations from the planned-work backlog: - **Minimal audio driver**: `play`/`start_music`/`stop_music` now generate APU pulse-1/pulse-2 writes from a builtin SFX/music name table, and the NMI handler gains a `JSR __audio_tick` splice (via the linker's `__audio_used` marker lookup) that ages an SFX countdown counter and mutes pulse 1 when the tone expires. Programs that never trigger audio pay zero ROM cost. - **u16 arithmetic and comparisons**: new IR ops `LoadVarHi`, `StoreVarHi`, `Add16`, `Sub16`, and six `Cmp*16` variants. The lowering context tracks variable types via the analyzer's symbol table and routes expressions through the 8-bit or 16-bit path based on operand width. Add16 emits `CLC;ADC;ADC` with carry propagating naturally into the high byte; compares dispatch high-byte-first with a short-circuit low-byte fallback. Fixes a silent miscompile where `big += 1` on a u16 var only incremented the low byte. - **Multi-scanline handlers per state**: `gen_scanline_irq` now dispatches on `(current_state, ZP_SCANLINE_STEP)` and reloads the MMC3 counter with the delta to the next scanline in the same state. `gen_scanline_reload` resets the step counter at the top of each NMI so a state with multiple handlers fires them in ascending line order. Previously only the first handler per state ever fired. - **IR temp slot recycling**: `build_use_counts` pre-scans each function to count per-temp uses; `retire_op_sources` decrements the counts after each op and pushes dead slots back onto `free_slots` for later allocation. `bitwise_ops.ne` used to crash (debug) or miscompile (release) once it hit 128 concurrent temps; with recycling the same function now uses ~4 slots instead of 136. - **INC/DEC peephole fold + improved dead-load elimination**: `fold_inc_dec` collapses `LDA addr; CLC; ADC #1; STA addr` into a single `INC addr` (and the SEC/SBC variant into `DEC addr`), saving 5 bytes and 5 cycles per increment. The fold is suppressed when the next instruction reads carry. `remove_dead_loads` now walks past INC/DEC/STX/STY (which don't touch A) to find the actual next A-use, catching more dead loads. Tests: 331 unit + 39 integration (up from 313 + 37), including new guards for audio, u16, multi-scanline, and slot recycling. https://claude.ai/code/session_01A8qk3gw2jWSzdiXBZPZSFE
2026-04-12 22:21:53 +00:00
3. **Register allocator** — proper A/X/Y allocation to replace
zero-page spills used by the current IR codegen. Partially
compiler: audio driver, u16 arithmetic, multi-scanline, slot recycling Five language features and optimizations from the planned-work backlog: - **Minimal audio driver**: `play`/`start_music`/`stop_music` now generate APU pulse-1/pulse-2 writes from a builtin SFX/music name table, and the NMI handler gains a `JSR __audio_tick` splice (via the linker's `__audio_used` marker lookup) that ages an SFX countdown counter and mutes pulse 1 when the tone expires. Programs that never trigger audio pay zero ROM cost. - **u16 arithmetic and comparisons**: new IR ops `LoadVarHi`, `StoreVarHi`, `Add16`, `Sub16`, and six `Cmp*16` variants. The lowering context tracks variable types via the analyzer's symbol table and routes expressions through the 8-bit or 16-bit path based on operand width. Add16 emits `CLC;ADC;ADC` with carry propagating naturally into the high byte; compares dispatch high-byte-first with a short-circuit low-byte fallback. Fixes a silent miscompile where `big += 1` on a u16 var only incremented the low byte. - **Multi-scanline handlers per state**: `gen_scanline_irq` now dispatches on `(current_state, ZP_SCANLINE_STEP)` and reloads the MMC3 counter with the delta to the next scanline in the same state. `gen_scanline_reload` resets the step counter at the top of each NMI so a state with multiple handlers fires them in ascending line order. Previously only the first handler per state ever fired. - **IR temp slot recycling**: `build_use_counts` pre-scans each function to count per-temp uses; `retire_op_sources` decrements the counts after each op and pushes dead slots back onto `free_slots` for later allocation. `bitwise_ops.ne` used to crash (debug) or miscompile (release) once it hit 128 concurrent temps; with recycling the same function now uses ~4 slots instead of 136. - **INC/DEC peephole fold + improved dead-load elimination**: `fold_inc_dec` collapses `LDA addr; CLC; ADC #1; STA addr` into a single `INC addr` (and the SEC/SBC variant into `DEC addr`), saving 5 bytes and 5 cycles per increment. The fold is suppressed when the next instruction reads carry. `remove_dead_loads` now walks past INC/DEC/STX/STY (which don't touch A) to find the actual next A-use, catching more dead loads. Tests: 331 unit + 39 integration (up from 313 + 37), including new guards for audio, u16, multi-scanline, and slot recycling. https://claude.ai/code/session_01A8qk3gw2jWSzdiXBZPZSFE
2026-04-12 22:21:53 +00:00
mitigated by peephole passes + the new slot recycler, but still
wasteful in some cases (every temp spills to a ZP slot even if
its live range is one op wide).
4. **Text / HUD layer** — font sheet + layout system for scores.
5. **Cross-block temp live-range analysis** — the current slot
recycler is function-local; temps that flow across block
boundaries always get a dedicated slot for the full function.
A proper CFG-aware live range interference graph would let more
temps share slots.
6. **Peephole: drop LDA dead across unconditional JMPs** — after
the INC/DEC fold we sometimes leave an `LDA #1` whose value is
consumed by nothing before the next `JMP __ir_main_loop`. Local
analysis can't prove it's dead; a cross-block pass could.