The compile bench had a hand-maintained parallel copy of
`src/main.rs::compile`, and that copy went out of sync after
bank switching landed — the bench kept handing the linker
`PrgBank::empty(...)` slots even though the CLI started
populating per-bank instruction streams + trampoline requests.
The assembler then panicked with `unresolved label:
'__tramp_step_animation'` on `uxrom_user_banked.ne` under
`cargo test --all-targets`, which is what CI runs. A plain
`cargo test --release` (what CLAUDE.md used to document) never
builds the bench so the bug slipped through local validation.
Fix:
- New `nescript::pipeline` module with `compile_source(source,
source_dir, &CompileOptions)` that owns the full
`parse → analyze → lower → optimize → codegen → peephole →
link` pipeline including the per-bank stream + trampoline
reconstruction. Returns a `CompileOutput` carrying the ROM,
the linker result, analysis, IR, assets, instructions, and
source-loc markers so downstream tools have one place to
pull metadata from.
- `src/main.rs::compile` reduces to file I/O + preprocessing +
a single `compile_source` call + CLI-only side effects
(`--dump-ir`, `--call-graph`, `--asm-dump`, `--memory-map`,
`--symbols`, `--source-map`).
- `benches/compile.rs::compile_pipeline` becomes a one-line
`compile_source` call. It is now structurally impossible for
the bench to drift from the CLI path.
- `tests/integration_test.rs::compile_with_debug_artifacts`
likewise delegates to `compile_source`. This also fixes a
latent bug in the helper where it used `Linker::with_mapper`
without `.with_header(...)` — programs opting into
`header: nes2` would have quietly got an iNES 1.0 header
through this path.
- `CLAUDE.md`: updated the "Running the basics" section to
specify `cargo test --all-targets` (plain `cargo test` skips
benches) and to point at `scripts/pre-commit` with the exact
install command. Also installed the hook in this worktree.
All 24 existing `examples/*.nes` rebuild byte-identical through
the new pipeline. 624 tests + all 25 emulator goldens pass.
https://claude.ai/code/session_01MaNVcDmK9gsspRkdxowQAM
- Analyzer: new `W0108` warning when an array's byte size exceeds
256. The codegen lowers `arr[i]` to `LDA base,X` and the 6502's
X register is 8 bits, so elements past byte 255 are unreachable.
The old debug bounds check silently skipped arrays in that range;
it now clamps the compare to 255 and the analyzer diagnoses the
declaration up front.
- UxROM `__bank_select`: the routine previously wrote the bank
number to a fixed `$FFF0`, which works on emulators that don't
simulate bus conflicts (jsnes, Mesen permissive) but is broken
on real hardware because a single ROM byte can't match every
possible bank number. Fixed by `TAX; STA __bank_select_table,X`
— the store lands at `table + bank_num`, whose ROM byte is
exactly `bank_num`, so CPU bus = A = ROM = no conflict. New
`LabelAbsoluteX` addressing-mode variant in the assembler
resolves the table's base address through the existing fixup
pass. The two existing UxROM example ROMs shift a few bytes
but their goldens still match (jsnes is bus-conflict-permissive).
- Source maps: new `source_map_survives_aggressive_peephole_folding`
regression test. The reviewer was worried peephole could drop
`__src_<N>` labels and silently leave stale source-map entries.
Peephole actually treats labels as block boundaries and never
deletes them — the test pins that down by compiling a program
tailored to trip every peephole fold and asserting every
codegen-recorded source marker survives into the final linker
label table.
- Frame-overrun counter: new `debug_frame_overrun_counter_reads_back_from_user_code`
end-to-end test that proves the contract works: NMI emits
`INC $07FF`, user `peek(0x07FF)` lowers to `LDA $07FF`, and the
RAM allocator doesn't hand out `$07FF` to a user variable.
https://claude.ai/code/session_01MaNVcDmK9gsspRkdxowQAM
Writing $07 in `emit_play_triangle` and $0B in `emit_play_noise`
meant that a noise play following an in-progress triangle note
would clear bit 2 of $4015 and cut the triangle off mid-envelope
(and vice versa). Write $0F from both paths so every trigger keeps
pulse1, pulse2, triangle, and noise enabled; channels with no
active envelope stay silent via the runtime's per-channel counter
gating. Also fixes the attribute-byte packing comment in
`png_to_nametable` — the code was correct, the doc string had the
quadrant order backwards.
The only observable ROM change is `examples/noise_triangle_sfx.nes`
(two immediate operands shift) and its audio hash golden; the
committed PNG golden is byte-identical. Found in independent code
review after the section landed.
https://claude.ai/code/session_01MaNVcDmK9gsspRkdxowQAM
Adds a `bank Foo { fun bar() { ... } }` parser form so user functions
can opt into living in a switchable PRG bank instead of the fixed
bank, plus the IR codegen, runtime, and linker work to make calls
across the bank boundary actually run. Programs that don't use the
new syntax produce byte-identical ROMs to before — verified by
rebuilding every existing example and diffing.
Pipeline shape:
* Parser accepts both `bank Foo: prg` (legacy reserved slot) and
`bank Foo { fun ... }` (functions land in the named bank). Nested
functions get tagged `bank: Some("Foo")` on the FunDecl + IrFunction.
* Analyzer bumps the user zero-page start past `$10` whenever the
program declares any banked function, so `__bank_select`'s STA into
ZP_BANK_CURRENT can't clobber a user variable. Programs without
banked functions keep the legacy `$10` start.
* IrCodeGen emits each banked function into its own per-bank
instruction stream (`banked_streams: HashMap<String, Vec<Instruction>>`)
while the fixed-bank stream gets the dispatcher loop + state
handlers + top-level functions, exactly like before. Cross-bank
calls from the fixed bank rewrite `JSR __ir_fn_<name>` to
`JSR __tramp_<name>`; in-bank calls stay direct. Banked → fixed
calls are direct (the fixed bank is always mapped at $C000-$FFFF).
Banked → other-banked calls aren't supported in this pass and
panic loudly during codegen.
* Runtime's `gen_bank_trampoline` takes the trampoline label and
entry label as parameters now (one trampoline per banked function,
not one per bank) so the linker can request any number of stubs.
* Linker assembles banked banks twice: a discovery pass to learn
each bank's labels, then a final pass that seeds the merged label
table so banked code can JSR into the fixed bank's runtime helpers
(math, audio, etc.). The fixed-bank assembler is also seeded with
the cross-bank labels so the trampolines' `JSR __ir_fn_<name>`
resolves into the bank's $8000 window. New `asm::assemble_with_labels`
/ `asm::assemble_discover_labels` helpers wire this up.
* PrgBank carries `Vec<Instruction>` + a list of `BankTrampoline`
requests now, replacing the old `data: Vec<u8>` + single
`entry_label: Option<String>` shape. The compiler populates both
from the codegen output; the linker's two-pass assembly handles
the rest.
New example: `examples/uxrom_user_banked.ne` puts a sprite-stepping
helper inside `bank Extras { fun step_animation() { ... } }`. The
fixed-bank state handler calls it via the generated trampoline, and
the harness golden locks in pixel + audio output at frame 180.
UxROM is the only mapper exercised by the new example. MMC1 and
MMC3 also work through the same path (the linker emits the right
mapper-specific bank-select code), but no example uses them yet —
the existing `mmc1_banked.ne` / `mmc3_per_state_split.ne` keep
their fixed-bank-only layout.
Limitations carried forward:
* No banked → banked cross-bank calls (panics in codegen).
* No greedy size-packing; placement is explicit-only.
* MMC3 state handlers don't get banked (the per-state split path
is untouched).
Adds `channel: triangle` / `channel: noise` to the `sfx` declaration
form. The existing pulse-1 / pulse-2 driver is unchanged (and is
still byte-identical for programs that don't use the new channels)
— when a program declares a triangle or noise sfx the runtime
splices in an additional per-channel slot that writes to $4008-
$400B (triangle) or $400C-$400F (noise) on play. Includes a new
`examples/noise_triangle_sfx.ne` demo with committed golden PNG +
audio hash.
https://claude.ai/code/session_01MaNVcDmK9gsspRkdxowQAM
Implements three items from docs/future-work.md's
"PNG-sourced palette and nametable assets" section:
- `palette Name @palette("file.png")` — the parser accepts a PNG
shortcut form; the asset resolver decodes the image via the
new `png_to_palette` helper, mapping each pixel's RGB to the
nearest NES master-palette index and building a 32-byte blob
that enforces the universal-first-byte convention (same as
the grouped-form parser). Errors cleanly on missing files or
more than 16 unique colours.
- `background Name @nametable("file.png")` — the parser accepts
a PNG shortcut form; the resolver decodes a 256×240 image into
a 960-byte tile-index table (deduplicating up to 256 unique
8×8 tiles) plus a 64-byte attribute table (bucketed by
average quadrant brightness). CHR data is not yet generated
automatically — callers still need to provide matching CHR
via the existing sprite / `@chr(...)` pipeline; the
limitation is documented on the `png_to_nametable` helper
and can be lifted in a follow-up.
- `--memory-map` now prints a "PRG ROM data blobs" section
listing each palette (32 B) and background (960 + 64 B)
under its linker-assigned label, plus a grand total. The
memory-map code is factored into `write_memory_map` which
takes a writer so unit tests can drive it against a
`Vec<u8>`. Memory-map printing moved to after the link step
so palette/background CPU addresses are available.
Call-site changes: `resolve_palettes` and `resolve_backgrounds`
now take a `source_dir` path and return `Result<_, String>`
because PNG decoding can fail. Updated the CLI driver,
benches/compile.rs, and every integration-test compile helper.
All 23 committed examples rebuild byte-identical; 525 lib
tests + 72 integration tests + 3 bin tests pass; clippy clean.
Implements four items from docs/future-work.md's "Debug instrumentation"
section so debugging on real ROMs is no longer a guessing game:
1. Mesen `.mlb` symbol export via `--symbols <path>`. The linker now
returns a `LinkedRom { rom, labels, fixed_bank_file_offset }` struct
from `link_banked_with_ppu_detailed`; `src/linker/debug_symbols.rs`
renders that plus the analyzer's var allocations into a Mesen-
compatible label listing (function entry points get `P:` entries
at PRG-relative offsets; user vars get `R:` entries).
2. Source maps via `--source-map <path>`. IR lowering now emits a
`SourceLoc(span)` op before every statement; the codegen turns each
one into a `__src_<N>` label-definition pseudo-op and records the
span in a side table. Source-marker emission is opt-in
(`with_source_map(true)`) because labels become peephole block
boundaries — leaving the markers off preserves byte-identical
release ROMs.
3. Array bounds checking under `--debug`. Every `ArrayLoad` /
`ArrayStore` now emits a `CMP #size; BCC ok; JMP __debug_halt; ok:`
guard, and the codegen emits one shared `__debug_halt` trap at the
end of the fixed bank (writes $BC to the debug port then wedges in
a tight `JMP $`). Release builds skip the whole thing.
4. Frame-overrun detection under `--debug`. `gen_nmi` now takes a
`debug_mode` flag; when on, it checks `ZP_FRAME_FLAG` at the top of
the handler and increments a counter at `$07FF`
(`DEBUG_FRAME_OVERRUN_ADDR`) if the flag was still set — meaning
the main loop didn't reach `wait_frame` before the next vblank.
User code can read the counter via `peek(0x07FF)`. This is the
abbreviated form the future-work doc suggested: a bump-a-counter
hook rather than a full cycle-budget tracker, which would need a
new builtin. The codegen emits a `__debug_mode` marker label in
debug mode so the linker can select the overrun-aware NMI variant.
Release ROMs for every committed example are byte-identical before
and after this change (verified with `git diff examples/` after a
full rebuild). All 512 lib tests and 71 integration tests pass;
`cargo fmt` clean; `cargo clippy --all-targets -- -D warnings` clean.
https://claude.ai/code/session_01MaNVcDmK9gsspRkdxowQAM
Implements two items from docs/future-work.md's language-feature gaps:
NES 2.0 header support: `RomBuilder` gains a `header_format` field
and a matching `enable_nes2()` method. When enabled, byte 7 bits 2-3
are set to `10` and bytes 8-15 are populated per the NES 2.0 spec
(submapper, PRG/CHR MSBs, PRG/CHR RAM, timing). The header stays
16 bytes. Programs opt in via `game Foo { header: nes2 }`; the
default remains iNES 1.0 so every committed example ROM is byte
identical. `validate_ines` now detects and reports which format it
parsed.
u16 struct fields: the analyzer's `register_struct` accepts `u16`
fields with a two-byte size and the struct-variable allocator tracks
per-field sizes so the synthesized `pos.x`/`pos.y` globals get the
right address span. IR lowering's `LValue::Field` and
`Expr::FieldAccess` follow the same wide path as u16 globals, and
struct-literal initialization writes both bytes for u16 fields.
Array and nested-struct fields stay rejected with a clearer
message. Existing u8/i8/bool struct programs are unaffected.
https://claude.ai/code/session_01MaNVcDmK9gsspRkdxowQAM
Adds two items from the "Code quality / tooling" section of
docs/future-work.md. Both make it easier to chase regressions
without touching codegen.
- `nescript build --no-opt` skips the IR optimizer pass so
optimizer-introduced miscompiles can be bisected against the
unoptimized output. Threaded through CompileOptions and gated
at the single optimizer call site in src/main.rs. Covered by a
new integration test that compiles the same program twice
(opt on / opt off) and asserts both outputs are valid iNES
ROMs with matching headers and reset vectors.
- A criterion-based `benches/compile.rs` harness that times the
full parse -> analyze -> lower -> optimize -> codegen -> link
pipeline on every examples/*.ne file. Sources are pre-read
into memory so file I/O stays off the hot loop, and each
example gets its own Criterion group for easy regression
spotting.
Committed ROM bytes under examples/*.nes are unchanged; the
emulator goldens under tests/emulator/goldens/ are untouched.
Extends the analyzer with the warnings listed under "Error message
polish" in docs/future-work.md:
- W0102 (existing): now also fires for `while true { ... }` and
`loop { if cond { continue } }`. `continue` is explicitly not
counted as an exit — the loop still spins forever.
- W0105 (new): palette declarations whose sub-palette first-bytes
disagree. The NES mirrors $3F10/$3F14/$3F18/$3F1C onto $3F00/
$3F04/$3F08/$3F0C, so a 32-byte sequential write silently
overwrites the background universal colour; the grouped
`universal:` form auto-fixes this so only flat `colors: [...]`
declarations can trip it.
- W0106 (new): a call at statement position whose callee has a
declared return type — the value is silently dropped.
- W0107 (new): `fast` variables with fewer than three observed
reads+writes, which don't justify holding a scarce zero-page
slot. Leading-underscore names are exempt.
All four are warnings only — no IR, codegen, or runtime changes,
so every committed example ROM rebuilds byte-identical and no
emulator goldens flip. Tests added in src/analyzer/tests.rs.
One legitimate W0106 surfaces on examples/platformer.ne (the
`resolve_enemy_hit` helper uses early-return for control flow
but its return value is discarded by the caller); fixing it
would shift the ROM and flip goldens, so it is left in place
as an informational hint rather than a hard fix.
https://claude.ai/code/session_01MaNVcDmK9gsspRkdxowQAM
Rewrites every example with non-trivial asset declarations to use
the pleasant QoL syntax introduced in the previous commit. Every
example still compiles to a byte-identical ROM (verified by a
temp-path diff before committing), so the committed `.nes` files
and the 23 emulator goldens are unchanged.
* platformer.ne — the centerpiece end-to-end demo:
- `palette Main` goes to grouped form with a shared
`universal: 0x22` (sky blue), one shared colour per
sub-palette, and named NES colours throughout; the
long-standing `$3F10` mirror-trap warning is now handled
by the parser and the manual pitfall comment is gone.
- `sprite Tileset` is 15 tiles of ASCII pixel art instead
of 240 bytes of inline hex.
- `background Level` uses a `legend { '.': 15, '#': 9, ... }`
block plus `map:` strings for the 32×30 nametable, and
`palette_map:` rows for the attribute table. The map
reads top-down like the rendered screen.
- SFX latch-once `pitch: 0x30` scalars + `envelope:` alias.
- `music Theme` uses note names + `tempo: 10` default.
* audio_demo.ne — scalar sfx pitches, `envelope:` alias, and a
note-name `C4, E4, G4, ...` music track.
* palette_and_background.ne — grouped CoolBlues / WarmReds
palettes with `universal: black` + named colours, plus
`legend` + `map:` tilemaps for the two backgrounds.
* sprites_and_palettes.ne — Arrow and Heart sprites rewritten
as `pixels:` ASCII art.
Along the way, two small parser extensions support the rewrites:
- `parse_pixel_art` now accepts `a/b/c` as aliases for `#/%/@`,
matching the vocabulary every NES editor (and our own
gen_platformer_tiles.rs generator) uses.
- `palette_map_to_attrs` allows up to 16 metatile rows (the
full attribute-table coverage, including the off-screen
bottom half) and auto-replicates row 14 → row 15 when only
15 rows are supplied so the visible bottom of the screen
gets consistent sub-palette assignments by default. The old
15-row cap couldn't match a hand-packed `0xAA` attribute
table for the last row; the platformer required this to
stay byte-identical.
`scripts/gen_platformer_tiles.rs` is updated to emit the new
syntax directly (pixel-art `pixels:` block + `legend`/`map:`/
`palette_map:` for the background), so regenerating the
platformer tiles stays a one-liner.
474 lib tests + 64 integration tests pass (3 new parser tests
for `palette_map:` 15/16/17 rows and the `abc` alias). All 23
emulator goldens still match pixel- and sample-for-sample.
https://claude.ai/code/session_01PzaSFj3VahDzxEYTKCESkz
`const_fold_block`'s per-block dead-code pass was collecting temp
usage from only the block it was folding, so a `LoadImm` whose
destination is consumed by a *sibling* block (for example via the
merge block's branch terminator) was incorrectly treated as dead
and dropped. The `and` / `or` short-circuit lowering emits exactly
that shape: the false path writes `LoadImm(result, 0)` and joins
with the right path at an `and_end` / `or_end` block whose branch
terminator reads `result`. After the DCE the false path's store
was gone, leaving the zero-page result slot to carry whatever value
the *previous* `and` / `or` evaluation had written there — stale
data that bled into subsequent conditional branches.
I found this while instrumenting `examples/platformer.ne` through a
puppeteer-driven jsnes harness, stepping one frame at a time and
snapshotting the full zero-page trace of each scenario (title-skip,
hold-right, hold-left, jump-spam, coin-drift, enemy-stomp, long-run).
In a clean idle run the enemy-1 stomp bounce (`rise_count = 6`,
`fall_vy = 0`) fired at emulator frames 83 and 96 with `camera_x`
= 61 and 74, i.e. with `e1_sx` = 39 and 26, nowhere near the
intended `[72, 96)` pickup window. The trigger turned out to be
the slot alias: every time `c2_sx` landed in its pickup window
(so the coin-2 `and` stored 1 into ZP(130)) and the player was
mid-fall at or past `player_y = 152`, the enemy-1 stomp `and`
short-circuited to its false path, left ZP(130) at 1, and the
stomp `if` fired on stale data.
The fix is to compute function-wide source-operand usage once before
folding each function's blocks and OR it into the per-block liveness
check, so a LoadImm is only dropped if nobody — neither its own
block nor any other block in the function — reads the temp. Added a
regression test (`const_fold_preserves_loadimm_used_by_sibling_branch`)
that builds the exact CFG shape the `and` lowering emits and
verifies the false-path `LoadImm(result, 0)` survives optimization.
Impact on the example ROMs:
- `examples/platformer.nes`: enemy-1 stomp now fires only when
`e1_sx ∈ [72, 96)`, as the source intends. The pixel golden is
unchanged (`player_y` converges back to the ground line before
frame 180), but the audio hash flips because the spurious
`play hit` sfx calls during coin-2 passage are gone. Committing
the new `tests/emulator/goldens/platformer.audio.hash`.
- `examples/logic_ops.nes`, `examples/bitwise_ops.nes`,
`examples/match_demo.nes`, `examples/mmc3_per_state_split.nes`,
`examples/two_player.nes`: byte-different but observably
unchanged — their pixel + audio goldens still match to the byte.
They exercise `and` / `or` in the source and now compile through
the corrected DCE.
All other example ROMs are byte-identical to pre-fix. `cargo fmt`,
`cargo clippy --all-targets`, `cargo test --release` (498 tests),
and `tests/emulator/run_examples.mjs` (22/22 goldens) are clean.
https://claude.ai/code/session_013Bi4H4YQ5or5HtMB4doUFi
Adds six NES-friendly authoring shortcuts so programs don't have to
hand-pack hex bytes for every kind of art asset. Every new syntax is
strictly additive — existing examples keep their byte-identical ROMs
and goldens.
* palette: ~50 named NES colours (`black`, `sky_blue`, `dk_red`, …)
usable anywhere a colour byte is expected, plus a grouped-form
`bg0..sp3` + `universal:` shape that auto-fills every sub-
palette's first byte (fixing the `$3F10` mirror trap).
* sprite: `pixels:` ASCII-art alternative to 16-byte CHR, supporting
multi-tile sprites split in row-major reading order.
* sfx: scalar `pitch:` matching the v1 driver's latch-once behaviour,
plus `envelope:` as a friendlier alias for `volume:`.
* music: `tempo:` default duration + note-name notes (`C4, Eb4,
rest 10`) alongside the existing `pitch, duration` pair form.
* background: `legend { '.': 0, '#': 1 }` + `map:` string rows,
plus `palette_map:` grids that auto-pack the 64-byte attribute
table from 16×15 sub-palette digits.
A new `examples/friendly_assets.ne` exercises every shortcut at once
with a matching pixel + audio golden; the other 22 golden tests still
match byte-for-byte.
https://claude.ai/code/session_01PzaSFj3VahDzxEYTKCESkz
Adds `examples/platformer.ne`, a full side-scrolling game that
exercises nearly every subsystem of the compiler in one program:
custom CHR tileset, 32×30 background nametable with per-region
attribute palettes, 2×2 metasprite hero with gravity/jump physics,
wrap-around horizontal scrolling, moving enemies, coin pickups,
user-declared SFX + music, and a Title → Playing state machine
with autopilot so the headless jsnes harness captures real
gameplay at frame 180. Tile art + nametable are generated by
`scripts/gen_platformer_tiles.rs` (`cargo run --bin gen_platformer_tiles`).
Building this out uncovered three independent runtime bugs that
together made the example render as black-on-black smileys. All
three are fixed in this commit:
1. **`gen_init` enabled sprite rendering before the linker's
initial palette/background load runs.** The PPU's v-register
auto-increments on every `$2007` write *during active
rendering*, so the palette load (32 B) and nametable load
(1024 B) were scrambled past the first ~72 bytes — every
existing program with a `background Level { ... }` block was
silently rendering zero-filled VRAM. Fix: leave `PPU_MASK = 0`
at the end of `gen_init` and emit a new `gen_enable_rendering`
call *after* all initial VRAM writes complete.
2. **Audio tick corrupted `ZP_CURRENT_STATE`.** The audio
driver's period-table lookup reused `$02/$03` as a temporary
indirect pointer with a comment claiming the slots were free
because the tick doesn't call mul/div. But `$03` is also
`ZP_CURRENT_STATE` used by the state dispatch loop, so every
music note silently overwrote the state index with the high
byte of `__period_table` (`0xC5` in the platformer ROM),
wedging the state machine forever. Fix: `gen_nmi` now PHAs
`$02/$03` on entry and PLA-restores them on exit, and the
audio tick JSR moves inside that save/restore window (it used
to be spliced by the linker *before* the register saves, so
even A/X/Y were technically being trashed pre-save). Only
`audio_demo`'s audio hash shifts (its note timings move a few
cycles); every other golden is unchanged.
3. **Sub-palette mirroring footgun.** Writing a 32-byte palette
blob sequentially causes the sprite sub-palettes' "index 0"
slots at `$3F10/$3F14/$3F18/$3F1C` to clobber the background
universal colour at `$3F00/$3F04/$3F08/$3F0C` via NES hardware
mirroring. The example's palette sets all eight first bytes
to `$22` (sky blue) for this reason; `docs/future-work.md`
picks up a TODO to warn on inconsistent first-byte values in
the analyzer.
Also:
- `docs/platformer.gif` — 6-second recording of the example
running in jsnes, generated by the new
`tests/emulator/record_gif.mjs` puppeteer helper (encodes via
`gifenc`, committed as a dev-dependency under
`tests/emulator/package.json`).
- README / examples/README tables and the 497-test count are
updated to cover the new example.
https://claude.ai/code/session_01BcCcHi6FUmTh8jC7UgkA3A
Re-adds `palette Name { colors: [...] }` and
`background Name { tiles: [...], attributes: [...] }` as first-class
declarations, plus `set_palette Name` and `load_background Name`
statements for runtime swaps. Unlike the previous iteration that
quietly no-op'd, this one is fully wired through the pipeline and
its behavior is pinned by both unit tests and an emulator golden.
Pipeline:
- Lexer: re-adds `palette`, `background`, `set_palette`,
`load_background` keywords and tokenizes them.
- AST: `PaletteDecl` (name + 1..=32 colour bytes) and `BackgroundDecl`
(name + 0..=960 tile bytes + 0..=64 attribute bytes) live in
`Program`. `Statement::SetPalette` and `Statement::LoadBackground`
name-reference these declarations.
- Parser: `palette Name { colors: [...] }` / `background Name
{ tiles: [...], attributes: [...] }` blocks and their statement
forms parse via the existing byte-array helper.
- Analyzer: validates colour indices ($00-$3F), palette length
(<=32), nametable length (<=960), attribute length (<=64), and
duplicate decl names. `set_palette` / `load_background` targets
must reference a declared name (E0502 otherwise). When a program
declares palette or background, the analyzer bumps the user
zero-page allocator's starting address from `$10` to `$18` to
reserve `$11-$17` for the runtime update handshake — programs
that don't use the feature keep the old layout so their emulator
goldens stay byte-exact.
- Assets: `PaletteData` and `BackgroundData` resolve declarations
into zero-padded fixed-size blobs (32 / 960 / 64 bytes) and
expose `label()` / `tiles_label()` / `attrs_label()` for codegen
to reference.
- IR: new `IrOp::SetPalette(String)` and
`IrOp::LoadBackground(String)`; lowering forwards the names
verbatim.
- Codegen: `gen_set_palette` writes the palette label pointer into
ZP `$12/$13` and ORs bit 0 into the update flags at `$11`;
`gen_load_background` does the same for tile/attribute pointers
at `$14/$15/$16/$17` with bit 1. Both emit a `__ppu_update_used`
marker so the linker splices in the NMI apply helper only when
the feature is actually used.
- Runtime: `gen_initial_palette_load` and
`gen_initial_background_load` write the first declared
palette/background at reset time (before rendering is enabled,
where PPU writes are safe). `gen_nmi(has_ppu_updates)` takes a
new flag; when true it splices `gen_ppu_update_apply` at the top
of the NMI body, which checks the `$11` flags byte and copies
pending palette / nametable data to `$3F00` / `$2000` inside
vblank. All helpers use only ZP $02/$03 as scratch at reset time
and never clobber ZP slots live across NMI.
- Linker: new `link_banked_with_ppu` takes slice of `PaletteData` /
`BackgroundData`; splices each blob as a labelled data block in
PRG ROM, picks the first-declared as the reset-time load target,
enables background rendering automatically when a background is
declared, and threads `has_ppu_updates` into `gen_nmi`. Old
`link_banked` remains as a thin wrapper for callers without
palette/background data so existing tests don't shift.
Tests:
- Lexer: tokenization of the 4 new keywords (single added test case).
- Parser: 5 new tests for `palette` / `background` decls with and
without attributes, plus `set_palette` / `load_background`
statements.
- Analyzer: 9 new tests covering acceptance of declared
palettes/backgrounds, E0502 for unknown names, E0201 for
out-of-range NES colors and oversized blobs, E0501 for duplicate
names, and the zero-page-layout guard (palette/bg decls bump ZP
start; no decls keeps it at $10).
- Resolver: 3 new tests for zero-padding, truncation of oversized
decls, and label derivation.
- IR: 2 new lowering tests for `set_palette` and `load_background`.
- Integration: 5 new tests — blob contents spliced verbatim into
PRG, `STA $12` / `STA $14` emitted by set_palette /
load_background codegen, and a regression guard that programs
without palette/background still land user vars at $10.
- Emulator: new `examples/palette_and_background.ne` driven by a
frame counter that toggles between `CoolBlues` / `WarmReds` and
`TitleScreen` / `StageOne` every 90 frames. Golden PNG and audio
hash checked in under `tests/emulator/goldens/` and verified via
`node run_examples.mjs` — rendered image shows the blue
`CoolBlues` palette with the nametable populated from
`TitleScreen`.
Docs:
- `README.md` adds the feature to the headline list and the example
table.
- `docs/language-guide.md` restores the palette/background sections
with the full 32-byte layout table and `set_palette` /
`load_background` statement references.
- `docs/future-work.md` replaces the "removed as dead code" entry
with the remaining gaps (PNG-sourced palette and nametable
assets, cross-vblank large background updates, memory-map
reporting).
- `spec.md` restores the grammar productions and usage examples.
- `examples/README.md` lists the new demo.
All 497 unit + integration tests pass. Clippy clean. All 21
emulator goldens match after the update pass.
https://claude.ai/code/session_012fKB251HvEUQwG3tizFyqt
Two correctness bugs were silently producing wrong ROMs:
- `x << n` / `x >> n` always shifted by 1, regardless of `n`, because
the IR lowering for `BinOp::ShiftLeft`/`ShiftRight` hardcoded the
count. Now eval_const the RHS into a compile-time count; fall back
to a new `IrOp::ShiftLeftVar` / `ShiftRightVar` (runtime loop) when
the amount isn't constant. Strength reduction folds the variable
form back to a fixed count once the optimizer knows the value.
- `x / n` / `x % n` always returned 0, because the lowering emitted
`LoadImm(t, 0)` for `BinOp::Div`/`Mod` with a comment saying the
runtime call was "TODO for now". Added real `IrOp::Div` and
`IrOp::Mod`, wired them through use-counting and DCE, gave codegen
`__divide`-based implementations, and taught strength reduction to
rewrite power-of-two divisors into shifts and modulo-by-2ⁿ into
AND masks. Constant folding now handles `Mul`/`Div`/`Mod`/shifts
too, which were previously left for the codegen to emit inefficient
software calls.
Dead code removed (no backward-compat shims kept):
- `src/debug/` entirely. `DebugSymbols`, `SourceMap`, and the
Mesen/.sym emitters had no callers outside their own tests;
`main.rs` never wrote a symbol file. Documented the intent in
`docs/future-work.md` so it comes back intentionally if needed.
- `ErrorCode::E0202` (invalid cast) and `E0403` (unreachable state):
defined, formatted, and marked `#[allow(dead_code)]` but never
emitted. W0104 now carries the unreachable-state semantics too.
- `Level::Info`: never constructed.
- `load_background` / `set_palette` statements and their
`BackgroundDecl` / `PaletteDecl` parser support: parsed and
silently dropped by IR lowering (`// TODO: implement in asset
pipeline`). Removed keywords, AST variants, parser paths, analyzer
arms, and tests. `docs/future-work.md` documents the runtime
palette/nametable design for when it comes back.
Doc cleanup:
- `docs/architecture.md` was describing files that don't exist
(`analyzer/types.rs`, `optimizer/const_fold.rs`, `codegen/regalloc.rs`,
`rom/header.rs`, `debug/symbols.rs`, …). Rewrote it to match the
real flat `mod.rs` + `tests.rs` layout and the real pipeline order.
- `docs/future-work.md` was a hybrid of open work and "recently
completed" entries that duplicated the active stubs at the top of
the file. Collapsed to just the gaps that are actually still open.
- `README.md` claimed Mesen symbol export and 210 tests; updated both.
- `docs/language-guide.md` and `spec.md` described `palette` decls,
`set_palette` / `load_background`, `debug.overlay`, and error codes
that were never emitted. Trimmed.
- Stale comments on `Statement::Play`/`StartMusic`/`StopMusic`
claimed the audio subsystem was "a no-op at codegen time".
Tests:
- Regression tests for every fix above (`lower_shift_left_with_literal
_count_uses_that_count`, `lower_shift_right_with_variable_count
_uses_runtime_variant`, `lower_divide_emits_div_op_not_load_imm
_zero`, `lower_modulo_emits_mod_op_not_load_imm_zero`,
`strength_reduce_div_by_power_of_two`, `strength_reduce_mod_by
_power_of_two`, `strength_reduce_shift_var_with_constant_amount`).
- Renamed the `program_with_sprites_and_palette` integration test
(which was exercising the now-removed `load_background`/`set_palette`)
to `program_with_inline_sprite_chr`.
`examples/sprites_and_palettes.ne` lost its `palette`/`set_palette`
usage. Nothing in the emulator test presses A, so the headless
jsnes render shouldn't move, but the golden may need regeneration
via `UPDATE_GOLDENS=1` if it does.
https://claude.ai/code/session_012fKB251HvEUQwG3tizFyqt
Prior to this commit the linker always shipped a single 16 KB PRG
bank regardless of the declared mapper, so the README's MMC1/UxROM/
MMC3 support was aspirational. This commit gives the three banked
mappers a real multi-bank ROM layout:
* RomBuilder.set_prg_banks() writes any number of 16 KB banks
back-to-back so the iNES header reflects the true PRG size.
* Linker.link_banked() places switchable banks first, fixed bank
last, so the fixed bank maps to $C000-$FFFF (the address window
where vectors and the runtime live).
* runtime::gen_mapper_init() emits reset-time mapper config:
MMC1 serial-writes a control-register value that pins the last
bank at $C000 with the correct mirroring, UxROM relies on the
power-on default, MMC3 writes the $8000/$8001/$A000/$E000
registers to get a known PRG and mirroring state.
* runtime::gen_bank_select() is a mapper-specific subroutine
(callable with the target bank in A) that maps any physical
bank to $8000-$BFFF.
* runtime::gen_bank_trampoline() generates a cross-bank call
stub in the fixed bank that saves the caller's bank, switches,
JSRs the target, and restores the fixed bank.
* The CLI and integration helper thread declared `bank X: prg`
declarations through to the linker so MMC1/UxROM/MMC3 programs
actually produce multi-bank ROMs.
Coverage:
* Runtime unit tests (18 new): mapper init patterns for every
supported mapper, bank-select signatures, trampoline dispatch
order, UxROM bus-conflict table contents.
* RomBuilder tests (6 new): multi-bank layout, padding,
byte-level fidelity, per-bank size validation, legacy
single-bank fallback.
* Linker tests (13 new): multi-bank ROM sizes across MMC1/
UxROM/MMC3, fixed-bank placement, switchable-bank payload
fidelity, bank-select subroutine detection, NROM rejection
of switchable banks.
* Integration e2e tests (16 new): compile real .ne sources
through the full pipeline and assert on iNES headers,
mapper init signatures in the fixed bank, vector locations,
and a regression check against `examples/mmc1_banked.ne`.
Total: 474 tests pass under `cargo test` with
`RUSTFLAGS="-D warnings"`.
https://claude.ai/code/session_01UCressA5e8k1XsuoJYLav2
The CLI's build path was calling `Linker::new(mirroring)`, which
hardcodes the mapper number to NROM (0) regardless of the source
file's `mapper:` declaration. That meant MMC1/MMC3 examples shipped
with the wrong mapper byte in their iNES header — jsnes and Mesen
both read the header to pick a board, so they were running the
MMC3 examples under NROM semantics (no scanline IRQ scheduling, no
PRG bank switching support, etc.). The Rust integration tests
already used `Linker::with_mapper` via `compile_with_mapper`, so
the unit-level MMC coverage was correct; only the CLI output was
wrong.
Swap to `Linker::with_mapper(program.game.mirroring, program.game.mapper)`
so the header matches the source. Confirmed by inspecting the
rebuilt example ROMs:
mmc3_per_state_split.nes: flags6=40 (mapper=4) ← was 00
scanline_split.nes: flags6=40 (mapper=4) ← was 00
mmc1_banked.nes: flags6=11 (mapper=1) ← was 01
hello_sprite.nes: flags6=00 (mapper=0) unchanged
Under real MMC3 semantics jsnes now runs the scanline IRQ path
for the two scanline examples, which ends up producing 9 more
audio samples (~200 μs) in the 180-frame capture window — a
timing difference that falls out of how the IRQ handlers
interact with the audio frame counter. Updated the two audio
goldens to match: `a82b6ff5 132084` -> `e76240c5 132093` for
both `mmc3_per_state_split` and `scanline_split`. PNG goldens
are unchanged — the visible output is the same.
All 19 emulator goldens now match. 381 unit tests + 43 integration
tests green. Clippy and fmt clean.
https://claude.ai/code/session_015WfaDttE3DpWn9rpyfpQd8
Two CI fixes for the audio subsystem PR:
1. Update `tests/emulator/goldens/audio_demo.audio.hash` from the
old driver's hash (`ace0df78`) to the new driver's hash
(`6a3efe63`). Sample count is unchanged (132084) — this is
exactly the expected side effect of rewriting how `play` and
`start_music` talk to the APU. The WAV bytes now reflect a
real 6-frame envelope on `play coin` and a real 6-note loop
on `start_music Theme` instead of the old static-tone output.
2. Revert the incidental `Linker::new` -> `Linker::with_mapper`
swap in `src/main.rs`. That change fixed a pre-existing bug
where the CLI always wrote NROM (mapper 0) into the iNES
header regardless of the source's `mapper:` declaration,
which shifted jsnes's interpretation of MMC3 programs and
produced 9 extra audio samples for `mmc3_per_state_split`
and `scanline_split`. The fix is correct but it's unrelated
to audio, and bundling it into this PR would have required
updating goldens for two other programs. I'll file that as
a separate PR with its own golden update. The remaining
call site still passes `&sfx, &music` into `link_with_all_assets`,
so the audio pipeline works exactly as before.
Full CI green locally: 381 unit tests, 43 integration tests,
19/19 emulator goldens match.
https://claude.ai/code/session_015WfaDttE3DpWn9rpyfpQd8
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
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
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
`IrCodeGen::next_oam_slot` incremented at *compile time*: one
`draw` statement = one fixed OAM slot, baked into absolute-mode
stores at codegen. A `draw` inside a `while`/`for`/`loop` body
was lowered once and then always wrote to the same four OAM
bytes every iteration, so only the last iteration was ever
visible. The writeup in the earlier PR called this "bug B".
Fix: reserve ZP `$09` as `ZP_OAM_CURSOR`, reset it to 0 at the
top of every frame handler (right after the existing OAM clear
loop), and lower each `DrawSprite` IR op to:
LDY $09 ; load cursor
LDA <y_temp>
STA $0200,Y ; sprite Y
LDA #tile
STA $0201,Y ; tile
LDA #0
STA $0202,Y ; attr
LDA <x_temp>
STA $0203,Y ; sprite X
INC $09 x4 ; bump cursor by 4
Cost is ~+6 bytes per `draw` over the old static form. At 64
slots the u8 cursor wraps naturally, giving classic NES
"too many sprites" flicker instead of a silent compile-time
drop. `next_oam_slot` and its resets are gone from the IR
codegen entirely.
Secondary fix: `for i in 0..N` counters are now registered as
handler locals. `lower_statement` created a `VarId` for the
counter via `get_or_create_var` but never pushed it onto
`current_locals`, so the IR codegen's `var_addrs` lookup
returned `None` for every `StoreVar(i)` / `LoadVar(i)` and
silently emitted nothing. The counter stayed at 0 forever,
the loop spun indefinitely, and every iteration wrote the
first array element into OAM — turning all 64 sprites into
the same smiley. Same class as the handler-local `var` decl
bug from the earlier PR, just for for-loop variables.
Smoke-test deltas (all 14/14 still pass):
- arrays_and_functions: 104 -> 260 (player + 4 enemies)
- bitwise_ops: 104 -> 416 (player + flag sprites + pips)
- loop_break_continue: 208 -> 208 (already fixed by the earlier pass)
- structs_enums_for: 104 -> 260 (player + 4 enemies)
Regression tests:
- `ir_codegen::more_tests::ir_codegen_draw_sprite` — checks a
single `draw` emits `LDY cursor`, four `STA $020N,Y`, and
four `INC cursor`.
- `ir_codegen::more_tests::ir_codegen_multi_oam_uses_sequential_slots`
— rewritten for the new form: each draw gets its own
`LDY cursor` + 4 `INC cursor`.
- `ir_codegen::more_tests::ir_codegen_draw_in_loop_...` —
proves a `draw` inside a `while` compiles to ONE cursor-based
draw (not N unrolled statics and not zero), and asserts no
stray `STA $0204`/`$0208`/... absolute stores — those would
indicate bug B has regressed.
- `ir::tests::for_loop_counter_is_registered_as_handler_local`
— verifies `for i in 0..N` pushes `i` onto `current_locals`
so the IR codegen allocates it.
Smoke-test tightening: `tests/emulator/run_examples.mjs` now
has per-example `minNonBlack` floors. `arrays_and_functions`,
`structs_enums_for`, `loop_break_continue`, and `bitwise_ops`
all require multi-sprite rendering — if the OAM cursor bug
comes back, the smoke test fails loudly instead of passing on
the default `nonBlack > 0` check.
The legacy AST codegen in `src/codegen/mod.rs` still uses the
compile-time `next_oam_slot` approach. It's only reachable via
`--use-ast`, none of the examples use it, and its integration
tests only check iNES structure — left alone on purpose.
https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
Landing bug A from the previous writeup plus two adjacent bugs
that the fix exposed. All three miscompile anything that uses a
u8[N] global with a literal initializer.
1. Array-literal globals are now actually initialized.
`lower_program` only expanded `Expr::StructLiteral` into per-
field synthetic globals — `Expr::ArrayLiteral` hit
`eval_const`, returned `None`, and the array boot-cleared to
zero. `IrGlobal` now carries an `init_array: Vec<u8>`
populated by lowering, and the IR codegen startup loop emits
one `LDA #byte; STA base+i` pair per element.
2. Local variables no longer overlap array globals.
`IrCodeGen::new` advanced `local_ram_next` past
`max_global_base + 1` — for an array at `$0300-$0303` it
placed the first handler-local at `$0301`, inside the array.
The frame handler's stores through the local then corrupted
the array mid-frame. The allocator now walks the analyzer's
`VarAllocation` list and advances past `address + size` for
every RAM global, not just the base.
3. Peephole `remove_redundant_loads` honors indexed LDAs.
The pass tracked `LDA Immediate/ZeroPage/Absolute` but let
`LDA AbsoluteX/AbsoluteY/ZeroPageX/IndirectX/IndirectY` fall
through the match, leaving the A-equivalence tracker
unchanged. A later `LDA #v` that happened to match a stale
entry from BEFORE the indexed load would then be dropped as
"already in A" — a silent miscompile that turned every
`draw Sprite at: (arr[i], arr[j])` pattern into garbage
(the second array index would be computed from `arr[i]`'s
value, reading way out of bounds). Indexed LDAs now clear
the tracker.
Regression tests:
- `src/codegen/peephole.rs`: a synthetic
`LDA #0; TAX; LDA AbsX(arr1); STA temp; LDA #0; TAX;
LDA AbsX(arr2); ...` sequence asserts both `LDA #0`s survive.
- `src/ir/tests.rs`: verifies `var xs: u8[4] = [1,2,3,4]`
populates `IrGlobal::init_array` with `[1,2,3,4]`.
- `tests/integration_test.rs`: two IR-codegen tests — one checks
the startup instructions contain `LDA #v; STA base+i` for
every element, the other compiles a handler-local var
alongside an array global and asserts no post-init stores
land inside the array.
Smoke test impact (14/14 still passing, now more visible):
- arrays_and_functions: 56 -> 104 nonBlack, now animated
- loop_break_continue: 52 -> 208 (player + 3 hazards visible)
- structs_enums_for: 52 -> 104 (player + enemy visible)
Existing examples unchanged; no remaining work for bug B
(static OAM slot allocation in loops) — that's the next PR.
https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
Running the compiled example ROMs through a headless puppeteer +
local jsnes harness exposed four latent bugs that the
header-structure-only integration tests couldn't catch:
- src/asm/mod.rs: the first pass treated ANY instruction with
`AddressingMode::Label` as a label definition, silently dropping
every `JMP`/`JSR` to a label. Now only `NOP + Label` is a label
def; other opcodes emit the opcode byte plus a 2-byte absolute
fixup resolved in pass two. Without this, every example crashed
with "invalid opcode at $1xxx" once the CPU fell through into
the math runtime and hit an unbalanced `RTS`.
- src/ir/lowering.rs (lower_handler): handler-local `VarDecl`s
(e.g. `var i: u8 = 0` inside a `while`) were pushed onto
`current_locals` but the handler built its own throwaway
`locals` list, so those var ids never got RAM addresses and
every `LoadVar`/`StoreVar` for them silently emitted nothing.
Seed `current_locals` with the state's declared locals and
reuse it so `lower_statement`'s appends flow through to the
`IrFunction`. Fixes the black screen in `arrays_and_functions`.
- src/ir/lowering.rs (global init): struct-literal initializers
on globals (`var player: Player = Player { x: 120, ... }`) fell
through to `eval_const`, which returned `None` for a
non-literal, so no init code was emitted. Now the per-field
synthetic globals each get their own `init_value`. Fixes the
black screen in `structs_enums_for`.
- src/codegen/mod.rs: the legacy AST codegen was emitting
`JSR __fn_poke` / treating `peek` as `LDA #0` for the hardware
intrinsics. It only "worked" before because the broken
assembler swallowed the bogus JSR. Handle `poke`/`peek` as
direct STA/LDA to a compile-time-constant absolute address,
matching the IR codegen's intrinsic path.
The harness lives in `tests/emulator/`: a tiny HTML page that
wraps the `jsnes` npm package, driven by a puppeteer script that
loads each ROM, runs ~180 frames, snapshots the canvas, and
records a smoke-test verdict (booted without a CPU crash, non-zero
pixels rendered, frames differ over time). `npm install && node
run_examples.mjs` from `tests/emulator/` runs the full sweep.
9/9 example ROMs now load, render, and animate where expected.
All 324 unit + 35 integration tests still pass.
https://claude.ai/code/session_014Z5y3Q9krLcAxYpZQJhZ5V
The OAM clear at frame-handler entry is only needed if the
handler actually modifies OAM — handlers that never call \`draw\`
don't need to reset the shadow buffer at all, since the previous
handler's contents are still correct (or the startup zero-fill
still applies). Adds a \`function_draws_sprites\` helper that
walks the IR ops once and emits the clear loop only when at least
one \`DrawSprite\` exists.
Handlers like a title screen that doesn't render sprites pay zero
cycles for the OAM clear now.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
A \`return\` with no value inside a function that has a declared
return type is an error — the caller expects a value. Previously
it compiled silently and the callee returned whatever was in A
from the last computation.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Each frame handler now begins with a short 64-slot loop that
writes \$FE to the Y-position byte of every OAM entry, hiding any
sprites the previous frame drew that the current frame doesn't.
Without this, stopping a \`draw\` call one frame leaves the sprite
lingering at its last position forever.
Cost: ~384 cycles per frame (2.3% of the ~16 666 cycles available
between vblanks). The per-frame \`draw\` calls overwrite the slots
they actually use, so the clear is free at runtime for sprites
that ARE drawn.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Previously \`remove_redundant_loads\` tracked only a single value
for A at a time, so \`STA addr\` would overwrite the prior
\`Imm(v)\` with \`Zp(addr)\`, losing the ability to eliminate a
following \`LDA #v\`. Now A's known value is an equivalence class
(\`Vec<AValue>\`): every \`STA addr\` *adds* the address to the
class instead of replacing it, so
LDA #5
STA \$10
LDA #5 ; eliminated — A still equals 5
collapses to just two instructions. Adding this also catches the
earlier \`STA \$10; LDA \$10\` idiom because \`Zp(\$10)\` is now part
of the class after the STA.
Any op that clobbers A clears the whole class; writes via STX/STY
or INC/DEC to an address in the class invalidate just that entry.
Instruction counts tighten on every example, most visibly on
bouncing_ball (123→120) and arrays_and_functions (247→245).
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Prints a tree view of every function/handler and its direct
callees, plus the max call depth reached from each state-handler
entry point. Useful for stack-budget investigation since the NES
has only 256 bytes of stack.
=== Call Graph (max depth: 2 / 8) ===
Main::frame (max depth 2)
├── clamp
├── clamp
└── check_collision
check_collision
├── abs_diff
└── abs_diff
abs_diff
└── (leaf)
clamp
└── (leaf)
Max-depth labels are only shown on entry points where the analyzer
has computed a depth; transitive callees print without a label so
the output isn't confusing.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Dumps a human-readable table of variable allocations sorted by
address, separated into zero-page and main RAM sections with a
final byte-usage summary. Struct fields show up as individual
entries under their synthetic \`var.field\` names.
Example output for examples/structs_enums_for.ne:
=== NEScript Memory Map ===
Zero Page (\$00-\$FF):
\$00-\$0F [SYSTEM] reserved (frame flag, input, state, params, scratch)
\$0010 [USER] enemy_y (u8)
\$0011 [USER] i (u8)
RAM (\$0200-\$07FF):
\$0200-\$02FF [SYSTEM] OAM shadow buffer
\$0300 [USER] player.x (u8)
\$0301 [USER] player.y (u8)
\$0302 [USER] player.vx (u8)
...
Zero Page: 2/128 bytes used
Main RAM: 11/1280 bytes used
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Common PPU/APU/mapper access previously required either variable
aliases or inline asm. Now two built-in intrinsics handle the
single-register case directly:
poke(0x2006, 0x3F) // STA \$3F, \$2006
poke(0x2006, 0x00)
poke(0x2007, 0x0F)
var status: u8 = peek(0x2002)
- Analyzer: \`poke\` / \`peek\` are recognized as built-in intrinsics
so they don't require a function declaration. Arity is still
checked (E0203 on mismatch).
- IR: new \`IrOp::Poke(u16, IrTemp)\` and \`IrOp::Peek(IrTemp, u16)\`
variants carrying the compile-time constant address.
- IR lowering: recognizes the \`poke\`/\`peek\` call names, evaluates
the address as a const expression, and emits the intrinsic op.
Falls back to a regular call if the address isn't a constant.
- IR codegen: emits a single LDA/STA in ZP or absolute mode based
on whether the address fits in a byte.
- Optimizer: Poke has a source temp (liveness), Peek has a dest
(new value); both pass through the existing passes.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
raw asm {
LDA #\$42
STA \$00
}
Unlike regular \`asm\`, \`raw asm\` does not perform \`{var}\`
substitution — the body is passed to the inline parser verbatim.
Useful for writing completely unmanaged bytes that don't rely on
the analyzer's variable allocations, e.g. mapper init snippets.
Implementation:
- Parser: \`KwRaw\` followed by \`KwAsm\` emits
\`Statement::RawAsm(body, span)\`.
- IR lowering: prepends a \`\\0RAW\\0\` marker to the body when
emitting \`IrOp::InlineAsm\` so the codegen can distinguish raw
from regular without adding a second op variant.
- IR codegen: strips the marker and skips substitution when present.
- AST codegen: same, handling \`Statement::RawAsm\` directly.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Within \`asm { ... }\` blocks, \`{name}\` is replaced with the
resolved hex address of the variable at codegen time. The lexer's
asm-body capture now balances nested braces so it doesn't cut off
at the first \`{x}\`. Both IR and AST codegen paths preprocess the
body before passing to the inline parser:
var counter: u8 = 0
on frame {
asm {
LDA {counter}
CLC
ADC #\$01
STA {counter}
}
}
Zero-page addresses become \`\$XX\`, absolute addresses become
\`\$XXXX\`. Unknown names pass through unchanged so the asm parser
can surface the "unknown mnemonic" / "unexpected token" error.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Newlines remain the primary separator, but \`;\` now lets short
statements share a line for readability:
on frame {
a += 1; b += 2
if button.a { a -= 1; b -= 1 }
}
Implementation: \`parse_block\` consumes any \`;\` tokens between
statements.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
match state {
Title => { if button.start { state = Playing } }
Playing => { /* ... */ }
GameOver => { if button.a { state = Title } }
_ => {}
}
- Lexer: \`match\` keyword and \`=>\` (FatArrow) token
- Parser: \`parse_match\` after the existing loop constructs. Each
arm is \`pattern => { body }\`, with \`_\` as the catch-all. The
match scrutinee is parsed with struct-literal restriction enabled
so the following \`{\` is unambiguously the match body, not a
struct literal.
- The parser desugars match directly into an if/else-if chain so
the analyzer, IR lowering, and codegen don't need new AST variants
— each arm becomes \`scrutinee == pattern\` as the condition, and
the default arm (if any) becomes the final \`else\` block.
Tests cover parse + full pipeline integration for state-style
dispatch using an enum.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
struct Vec2 { x: u8, y: u8 }
var pos: Vec2 = Vec2 { x: 100, y: 50 }
on frame {
pos = Vec2 { x: pos.x + 1, y: pos.y }
}
- AST: new \`Expr::StructLiteral(name, fields, span)\` variant
- Parser: in expression position, \`Ident {\` enters struct-literal
mode when the new \`restrict_struct_literals\` flag is off.
\`if\`/\`while\`/\`for\` conditions set the flag so the \`{\` keeps
going to the following block. Condition contexts can still use
struct literals by parenthesizing them.
- Analyzer: validates that the struct type exists, each named field
belongs to it, and each field value has a compatible type.
- IR lowering: desugars \`var = StructLiteral { ... }\` (both in
assignments and variable initializers) into per-field StoreVar
operations against the analyzer-synthesized \`var.field\`
variables. No IR type for struct values is needed.
- AST codegen: no-op (legacy path).
- examples/structs_enums_for.ne now uses a struct literal for the
initial \`player\` state instead of per-field assignments.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Both peephole passes now track \`Absolute\` addresses in addition to
zero-page. This lets function-local variables (which live in RAM at
\$0300+) benefit from the same redundant-load elimination and
operand-rewriting that globals at ZP addresses already enjoyed.
Before:
LDA \$04 ; load param
STA \$0300 ; store to t
LDA \$0300 ; reload t
STA \$81 ; store to temp
LDA \$0300 ; reload t again
STA \$82 ; store to temp
LDA \$81
CLC
ADC \$82
STA \$0300
LDA \$0300
RTS
After:
LDA \$04
STA \$0300
CLC
ADC \$0300
STA \$0300
RTS
Renames the A-tracking enum's \`Addr\` variant to \`Zp\` alongside a
new \`Abs\` variant, matching the \`Source\` type used by copy prop.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Function bodies can declare local variables with \`var NAME: u8 = …\`.
Previously the lowering created a VarId for them but didn't track it
on the \`IrFunction.locals\` list, so the IR codegen had no address
to map it to and \`LoadVar\` / \`StoreVar\` silently did nothing. The
generated function body read and wrote random temp slots.
Fixes:
- Lowering: replaced the per-function \`locals\` local with a
long-lived \`current_locals\` field; \`lower_function\` resets it
on entry and moves it into the \`IrFunction\` at exit. Each
\`Statement::VarDecl\` inside a function body appends to
\`current_locals\`.
- IR codegen: iterate every function's \`locals\` list. Params 0..4
still map to \$04-\$07, and the remaining locals get addresses in
main RAM starting at \$0300. Each function's locals are disjoint,
so nested calls don't corrupt each other's state.
- Integration test \`program_with_function_local_variables\`
exercises nested calls with function-local state to guard against
regression.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Adds a \`for NAME in START..END { BODY }\` half-open range loop:
for i in 0..8 {
total += arr[i]
}
- Lexer: \`for\`, \`in\` keywords and the \`..\` range operator
- AST: new \`Statement::For\` variant with var/start/end/body
- Parser: \`parse_for\` after \`while\` / \`loop\`
- Analyzer: registers the loop variable as a u8 symbol for the body
(restoring any shadowed outer symbol afterwards), allocates it via
the normal RAM allocator, and tracks it as "used"
- IR lowering: desugars to \`var = start; while var < end { body;
var = var + 1 }\` using a \`for_step\` continue-edge block so
\`continue\` properly increments the index
- AST codegen: no-op (legacy path doesn't need for loops)
- Tests: parse + full-pipeline integration
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Constants can now reference other constants and use arithmetic /
bitwise / comparison operators. The lowering runs \`eval_const\` on
each \`ConstDecl\` and \`VarDecl\` initializer, threading the existing
\`const_values\` map for identifier lookups.
Examples that now work:
const BASE: u8 = 10
const OFFSET: u8 = BASE + 5 // 15
const MASK: u8 = FLAG_A | FLAG_B // bitwise-or of two earlier consts
var counter: u8 = BASE // initializer is now 10
Enum variants are registered before constants so a const like
\`const FIRST: u8 = VariantA\` also resolves at compile time.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Two new passes that save 2 instructions per conditional:
- \`fold_branch_over_jmp\` rewrites \`BNE taken; JMP else_block;
taken:\` into \`BEQ else_block; taken:\`. Emitted by the IR codegen
for every \`if\` without a matching \`else\`. Bounded lookahead of
~60 instructions keeps the rewritten branch within the signed
8-bit reach (-128..+127).
- \`remove_jmp_to_next_label\` drops \`JMP foo; foo:\` since the JMP
is a no-op — execution falls through to the label anyway.
Instruction count impact (IR vs AST, lower is better):
before after
arrays_and_functions 262 vs 236 -> 243 vs 235
bouncing_ball 135 vs 144 -> 122 vs 143
coin_cavern 290 vs 291 -> 261 vs 289
hello_sprite 73 vs 64 -> 64 vs 63
mmc1_banked 261 vs 247 -> 235 vs 246
sprites_and_palettes 121 vs 113 -> 108 vs 112
state_machine 233 vs 229 -> 209 vs 228
IR codegen now beats the AST codegen on 5 of 7 examples and is
competitive on the other two.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
The \`play\`, \`start_music\`, and \`stop_music\` keywords were lexed
but not parsed. Programs using them failed with a generic "unexpected
token" error. Now:
- \`Statement::Play(sfx_name, span)\`
- \`Statement::StartMusic(track_name, span)\`
- \`Statement::StopMusic(span)\`
parse successfully and flow through analyzer / IR lowering / codegen
as no-ops. Semantics match the spec but produce no output — no audio
driver exists yet. Users who need sound can still wire in custom
code via \`asm { ... }\` blocks.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Previously an assignment to an undeclared name was silently accepted
because the IR lowering would synthesize a fresh VarId for it. Now
the analyzer emits E0502 with the same "did you mean" suggestion
path as read-side undefined variable errors.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Two bugs in the IR codegen's function call path:
1. \`IrOp::Call\` was emitting \`JSR __fn_name\` but functions are
defined at \`__ir_fn_name\`. The JSR went to an unresolved label
and the generated ROM would jump to garbage.
2. Caller stored arguments to ZP \$04-\$07 but the callee's
\`LoadVar\` for each parameter VarId resolved to a temp slot at
\$80+ because parameter VarIds weren't in \`var_addrs\`. The
callee never read the arguments.
Fixes:
- Register each function's first four parameter VarIds at
\$04-\$07 in \`var_addrs\` during codegen construction, so
\`LoadVar\` for a parameter emits \`LDA \$04\` / \`\$05\` / etc.
- Call emits \`JSR __ir_fn_{name}\` matching the function label.
Adds a unit test that exercises both sides of the contract:
the caller must STA to \$04/\$05 and JSR to \`__ir_fn_sum\`, and the
callee must LDA from \$04 to read the first parameter.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Extends the \`on_scanline\` codegen to support multiple scanline
handlers across states:
- \`__irq_user\` now dispatches by \`current_state\`: each state with a
scanline handler gets a CMP/BNE/JSR entry in the dispatch table.
States without a handler fall through to just acknowledge the IRQ.
- New \`__ir_mmc3_reload\` helper that (re)loads the MMC3 counter
latch based on \`current_state\`. States without a scanline handler
fall through to disable the IRQ (\$E000 write).
- Linker detects the \`__ir_mmc3_reload\` label in user code and
splices a JSR into it at the top of the NMI handler, so the
counter is reloaded once per frame with the current state's
target scanline.
- IRQ handler no longer re-enables IRQ on ACK (the NMI reload now
handles that) so it won't fire multiple times per frame.
- Program init chooses the start state's scanline count (if any) or
the first scanline handler found as a fallback.
Also fixes \`dump_asm\`: a \`NOP\` with a \`Label\` operand is a label
definition, but any other opcode with a \`Label\` operand is a real
instruction like \`JSR foo\`. The old dump was hiding JSR/JMP targets.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Wires \`on scanline(N)\` handlers through IR lowering and codegen:
- IR lowering: each scanline handler becomes a regular IR function
named \`{state}_scanline_{N}\`
- IR codegen: when any scanline handler exists, emits MMC3 IRQ setup
at program start (\$C000 latch, \$C001 reload, \$E001 enable, CLI)
and a \`__irq_user\` handler that saves registers, acknowledges via
\$E000, JSRs the scanline handler, restores registers, and RTIs
- Linker: vector table prefers \`__irq_user\` over the default \`__irq\`
stub when both exist
Scope of this first pass is intentionally minimal: supports ONE
scanline handler per program (the first one found in IR function
order). Per-state dispatch and multi-scanline reload will come later.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Adds composite \`struct\` types with field access:
struct Vec2 { x: u8, y: u8 }
var pos: Vec2
pos.x = 100
pos.y = pos.x + 5
- Lexer: \`struct\` keyword
- AST: \`StructDecl\` with \`StructField\` list; \`NesType::Struct(name)\`
for struct-typed variable declarations; \`Expr::FieldAccess\` and
\`LValue::Field\` for reads/writes
- Parser: top-level \`struct Name { field: type, ... }\` (optional
trailing comma) and \`ident.field\` syntax in both expression and
lvalue position
- Analyzer: \`register_struct\` computes contiguous field offsets
(no padding) and stores them in \`struct_layouts\`. Struct variables
synthesize a \`VarAllocation\` per field under the name
\`"struct_var.field"\`, and \`Expr::FieldAccess\` / \`LValue::Field\`
resolve against those. Unknown struct types and unknown fields
emit E0201.
- IR lowering + AST codegen: treat struct field access as ordinary
variable access against the synthetic per-field symbols. No new IR
ops are needed.
v1 structs only support primitive fields (u8/i8/bool). Nested structs,
u16 fields, and array fields are not yet allowed.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
Adds two new passes that dramatically close the IR-vs-AST output
size gap:
- **copy_propagate_temps**: tracks the source of each \`STA\` into an
IR temp slot (\$80-\$FF). Subsequent \`LDA\` of that slot is rewritten
to load from the original source directly. The same rewrite applies
to arithmetic/logical ops that read a temp (\`ADC\`/\`SBC\`/\`AND\`/
\`ORA\`/\`EOR\`/\`CMP\`). The now-unused \`STA\` is picked up by the
dead-store pass on the next iteration.
- **remove_dead_loads**: drops an \`LDA\` whose value is immediately
overwritten by another A-clobbering load (\`LDA\`/\`PLA\`/\`TXA\`/\`TYA\`).
The existing dead-store pass is also extended to scan through labels
and branches within a function body, since IR temp slots are
SSA-unique per function. Function boundaries (\`__ir_fn_\`/\`__fn_\` labels,
\`RTS\`/\`RTI\`) still stop the scan.
Instruction count comparison (IR vs AST codegen on examples/):
before after
arrays_and_functions.ne 359 vs 236 -> 264 vs 236 (-95 ir)
bouncing_ball.ne 192 vs 144 -> 135 vs 144 (-57 ir, IR now wins)
coin_cavern.ne 412 vs 291 -> 290 vs 291 (-122 ir, IR ties)
hello_sprite.ne 116 vs 64 -> 73 vs 64
mmc1_banked.ne 378 vs 247 -> 261 vs 247
sprites_and_palettes.ne 194 vs 113 -> 121 vs 113
state_machine.ne 323 vs 229 -> 233 vs 229
IR is now competitive with the AST codegen on every example and
beats it on bouncing_ball and coin_cavern.
https://claude.ai/code/session_01W6eQFStA66EuMKHUFo2rx3
