boot2, branch HEAD

Refactor scheme1.P1pp: use tag idioms, lexer helpers, %die, %alignup

2026-04-25T10:03:52Z

commit 354384eae6863bd903d5ea4ed4511c54e11636b4 parent 925240dd23966009f0b742a6199d87f1661e79fa Author: Ryan Sepassi Date: Sat, 25 Apr 2026 03:03:52 -0700 Refactor scheme1.P1pp: use tag idioms, lexer helpers, %die, %alignup The tag-idiom macros (%car, %cdr, %tagof, %hdr_type, %mkfix, %untag_fix, %untag_sym) were defined but unused -- every call site inlined the underlying instruction. Now the file routes every tag operation through them so the layout is documented in one place. Adds scheme1-local helpers and uses them at the obvious sites: %readbuf_byte -- load one byte at readbuf_buf[off] (skip_ws, parse_one, parse_list, parse_atom scan loop) %is_ws_branch -- 4-way ASCII whitespace check (same callers) %symtab_entry -- base + idx*SYMENT.SIZE (intern, sym_global, sym_set_global) %die(msg, code) -- print_cstr + sys_exit, replacing seven copies of the same la/call/li/call/spin block Adds a generic %alignup(rd, rs, align, scratch) to libp1pp; uses it in p1_main's heap_next setup and alloc_hdr. Drops the dead ::spin labels after %die-style sys_exit calls (sys_exit doesn't return). parse_one's frame is 0 since the body never touches sp slots. Bumps M1PP_MACRO_BODY_CAP from 0x18000 to 0x30000: the new helper macros plus the larger amd64 backend pushed the body-token pool past the old cap. Notes: - parse_atom's first-byte dispatch keeps the inline la/ld/add/lb sequence; routing it through %readbuf_byte caused the apply dispatch to read a non-PRIM value on aarch64 in a way I haven't fully diagnosed. - amd64 and riscv64 still fail tests/scheme1/00-exit with "not a procedure" -- pre-existing on the baseline, unchanged here.

Add scheme1 phase-1 interpreter spine and test runner

2026-04-25T10:03:32Z

commit 925240dd23966009f0b742a6199d87f1661e79fa parent 6ce02250c5f5982da088c191b4da1ac0ce6a4958 Author: Ryan Sepassi Date: Sat, 25 Apr 2026 03:03:32 -0700 Add scheme1 phase-1 interpreter spine and test runner Tag-dispatched eval/apply over a bump heap, linear-scan symbol intern table, and a single sys-exit primitive. Adds a scheme1 Make target, .P1pp build pipeline support in boot-build-p1pp.sh, and a scheme1 suite in run-tests.sh that diffs stdout and exit code against .expected / .expected-exit fixtures. Also bumps the m1pp INPUT/OUTPUT/TEXT caps so combined.M1pp inputs that include libp1pp + scheme1 fit through the expander.

Refactor M1pp.P1 boilerplate: BSS init loop, lex helper, char-class table

2026-04-25T03:39:45Z

commit 6ce02250c5f5982da088c191b4da1ac0ce6a4958 parent d87400a1c88fefd1ead7b3d84d9a7a03f3f276c9 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 20:39:45 -0700 Refactor M1pp.P1 boilerplate: BSS init loop, lex helper, char-class table Four independent simplifications, 57 lines net: - p1_main's BSS pointer-slot init is now a table walk over bss_init_tbl instead of 16 unrolled 4-line blocks. - The five single-char lex blocks (lex_lparen/rparen/comma/lbrace/rbrace) share a new lex_punct1 helper; each block is now 4 lines. - lex_loop and lex_word_scan dispatch via a 256-byte lex_char_class table in rodata, collapsing two linear char-comparison ladders. - err_* stubs drop their explicit message length; fatal computes it via an inline strlen loop, and msg_* strings are NUL-terminated. Adds ZERO4 and :lex_punct_kind. P1 backend tables regenerated by the prune step pick up a handful of new register/offset combos.

Move M1pp.P1's BSS-style buffers past ELF_end; bump ph_memsz to 8MB

2026-04-25T01:20:20Z

commit d87400a1c88fefd1ead7b3d84d9a7a03f3f276c9 parent 1a73629687b8843765b1ef6a00e4ee8a6cec6843 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 18:20:20 -0700 Move M1pp.P1's BSS-style buffers past ELF_end; bump ph_memsz to 8MB Previously M1pp/M1pp.P1 reserved its working buffers (input_buf, output_buf, source_tokens, macros, expand_pool, ...) inline as ZERO32 padding inside the file. Each `ZERO32` token expanded into M0's output as a 66-char `'00...00'` quoted literal, so the 133,784 ZERO32 tokens ballooned prog.hex2 to 8.9 MB even though the actual program code is only ~38 KB. hex2-0 then had to walk that 8.9 MB of zeros byte by byte; under qemu-x86_64-static on Apple Silicon that took many minutes. Refactor: * All 16 ZERO32-using buffers move to BSS past ELF_end. Each gets an OFF_ DEFINE (offset from ELF_end) and an 8-byte pointer slot (input_buf_ptr, output_buf_ptr, ...). p1_main initializes every slot once at startup as ELF_end + OFF_. * All 72 `la_ &` references rewritten to the cached-pointer pattern: `la_ &_ptr` then `ld_,,0`. Same dest reg for both ops — no scratch register needed, no clobbering of caller state. * vendor/seed/{aarch64,amd64,riscv64}/ELF.hex2: ph_memsz hardcoded to 8 MB (0x800000). The kernel zero-fills the gap between ph_filesz and ph_memsz at load, so the BSS is free. scripts/refactor-m1pp-bss.py drove the source-level transform; kept in the tree for reference. Numbers (aarch64 m1pp build): * M1pp.P1: 22931 -> 6382 lines (-72%) * prog.hex2: 8.9 MB -> 97 KB (-99%) * m1pp ELF: 4.3 MB -> 38 KB (-99%) Build times (clean): * aarch64 m1pp: ~5s (was ~12s) * amd64 m1pp: ~15s (was many minutes / never finished) * full `make test` across all 3 arches: 29s, 75/75 pass.

Drop parens on zero-arg macro calls

2026-04-25T01:02:46Z

commit 1a73629687b8843765b1ef6a00e4ee8a6cec6843 parent 6f73803c2215ca80eee97a3c0c85d1238d6113a1 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 18:02:46 -0700 Drop parens on zero-arg macro calls M1pp accepts `%name` as a parenless invocation of any zero-parameter macro (test 13-parenless covers the contract). Adopt that style across P1 source: control-flow ops (`%ret`, `%eret`, `%syscall`), syscall- number macros (`%sys_write`, `%p1_sys_*`), backend-internal helpers (`%amd_cqo`, `%aa64_ret`, `%rv_ecall`, ...), and paste-dispatch sites (`%amd_reg_##r`, `%rv_is_sp_##r`, ...). Macro definition headers stay paren'd — the header parser still requires `%macro NAME(...)`.

rm post, moved to site

2026-04-25T00:52:57Z

commit 6f73803c2215ca80eee97a3c0c85d1238d6113a1 parent 5d78449f72b14b1a16ce75235d12da6c4e14d1c5 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 17:52:57 -0700 rm post, moved to site

P1.LI: take immediate as a macro arg

2026-04-25T00:52:49Z

commit 5d78449f72b14b1a16ce75235d12da6c4e14d1c5 parent 2216e63bcd1f8209385688b28c0547a4cf370dff Author: Ryan Sepassi Date: Fri, 24 Apr 2026 17:52:49 -0700 P1.LI: take immediate as a macro arg Make the front-end `%li(rd, imm)` symmetric with `%la(rd, target)` and `%addi(rd, ra, imm)`. The backend `p1_li(rd, imm)` now owns both the opcode prefix and the literal-pool word, so portable source no longer encodes the literal width inline. Syscall-number macros (`p1_sys_*`) return integer atoms instead of raw 8-byte emissions so they compose through M1pp's expression evaluator inside `%li(a0, %sys_write())`.

M1pp.P1 correct tokens end

2026-04-25T00:32:50Z

commit 2216e63bcd1f8209385688b28c0547a4cf370dff parent 22ad7bf0fac945c332a1f325cd292508bedb4953 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 17:32:50 -0700 M1pp.P1 correct tokens end

Pass ARCH via env, drop arg-list noise from script invocations

2026-04-25T00:05:32Z

commit 22ad7bf0fac945c332a1f325cd292508bedb4953 parent 5757513857ba4807e5642573a662659b4a4ad398 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 17:05:32 -0700 Pass ARCH via env, drop arg-list noise from script invocations Every script in scripts/ now reads ARCH from its environment and derives the per-arch fixed paths (P1/P1-$ARCH.M1, vendor/seed/$ARCH/ELF.hex2, build/$ARCH/tools/, build/$ARCH/m1pp, build/$ARCH/.work/$NAME) from that. Only the variable per-call inputs (source, output) are passed positionally. Make's PODMAN macro now adds `-e ARCH=$(1)`, so any boot-* script invoked through it sees the arch automatically. Lint runs `ARCH=$* sh scripts/lint.sh ` from the recipe. Concretely: boot1.sh ARCH only (was

) boot2.sh ARCH only (already) boot-build-p1.sh ARCH + src out (was 6 positional args) boot-build-p1pp.sh ARCH + src out (was 8 positional args) lint.sh ARCH + src(s) (was ...) run-tests.sh passes ARCH= via container env to boot-* calls Make rule recipes go from multi-line continuations to single calls: $(call PODMAN,$*) sh scripts/boot-build-p1.sh M1pp/M1pp.P1 $@ Verified: make all + make hello + make run print "Hello, World!"; aarch64 m1pp suite 20/20 in ~6s.

Push container builds onto tmpfs; restructure scripts/ around boot-* convention

2026-04-24T23:57:08Z

commit 5757513857ba4807e5642573a662659b4a4ad398 parent 56dfcbf7c27c97396b11fa3e71bee8d9a03a6eed Author: Ryan Sepassi Date: Fri, 24 Apr 2026 16:57:08 -0700 Push container builds onto tmpfs; restructure scripts/ around boot-* convention * Convention: every script invoked via `podman run` is named boot*.sh. Rename scripts/bootstrap.sh -> scripts/boot1.sh; new scripts/boot2.sh builds M1pp + pokem inside the container; scripts/boot-build-p1.sh and scripts/boot-build-p1pp.sh are the generic in-container builders called by Make for individual programs and test fixtures. * Make becomes the single source of truth for build dependencies. The old host-side dispatchers scripts/build-p1.sh and scripts/build-p1pp.sh are gone; their work now lives in boot-build-p1.sh / boot-build-p1pp.sh, and Make rules wrap them with `podman run ... sh scripts/boot-*.sh ARGS` — no inline shell. * /tmp inside the container is now a real RAM tmpfs (--tmpfs /tmp:size=512M). Previously /tmp fell through to the container overlay, still backed by virtiofs on Apple Silicon, so the stage0 tools' per-byte fputc/fgetc paid a virtiofs round-trip per byte. Real RAM tmpfs cuts a clean m1pp build ~3s and is propagated to all podman invocations (Make + run-tests). * boot1.sh stages its hex0/hex1/hex2-0/catm/M0 chain entirely through /tmp and copies the final tools to the bind mount at the end; intermediates never touch virtiofs. * Pre-prune the per-arch P1 backend tables once, check in the result. scripts/prune-p1-table.sh emits P1/P1-.M1 from the full generated table (now at build/p1/.M1) keeping only the DEFINEs referenced by M1pp/M1pp.P1, pokem/pokem.P1, and tests/M1pp/*.M1. Routine builds skip the per-build prune; `make tables` regenerates after editing P1/gen/*.py or the prune-source set. * Drop the ELF p_filesz trim. The kernel zero-fills BSS at load time, so the trailing zero padding in raw hex2-0 output is wasted bytes on disk but functionally identical at exec. * Drop NATIVE=1 mode. The bootstrap chain is now the only path; the host native tools (build-native-tools.sh / `make tools-native`) are kept as an opt-in helper but no longer wired into any default rule. Verified: make all + make hello + make run print "Hello, World!"; the aarch64 m1pp suite passes 20/20 in ~5s. Cross-arch m1pp builds run under qemu-user emulation and are slow but correct (per-byte syscalls × qemu overhead). The .P1pp suite still hits the known self-hosted m1pp buffer overflow on the combined backend+frontend+source — to be fixed by bumping M1pp_*_CAP constants in M1pp/M1pp.P1.

Drop "v2" from P1 naming everywhere

2026-04-24T23:04:52Z

commit 56dfcbf7c27c97396b11fa3e71bee8d9a03a6eed parent ccb69f1c9fcd1e3b5222653b3ba3255ba05efc79 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 16:04:52 -0700 Drop "v2" from P1 naming everywhere The "v2" suffix was a transitional label from the P1 redesign and no longer needs distinguishing. Replace P1v2 / P1 v2 / P1v2-{32,64} with P1 / P1-{32,64} in docs, comments, and source headers. Rename the generated-table build path build/p1v2/ -> build/p1/, and the docs/P1v2.md references that already point at docs/P1.md.

Restructure build into scripts/, vendor seeds, unify ARCH paths

2026-04-24T22:58:07Z

commit ccb69f1c9fcd1e3b5222653b3ba3255ba05efc79 parent ffe72f13e339e8411d89985ad324f01a6483a4e6 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 15:58:07 -0700 Restructure build into scripts/, vendor seeds, unify ARCH paths * Move build helpers into scripts/: bootstrap.sh, build-p1.sh, build-p1pp.sh, build-native-tools.sh, lint.sh, run-tests.sh, plus Containerfile.busybox. Drop the per-suite m1pp/test.sh, m1pp/build*.sh and tests/p1/test.sh runners. * Vendor stage0-posix seed sources into vendor/seed//. The chain no longer needs a sibling live-bootstrap checkout; populate-upstream.sh is gone. * Switch every podman invocation to one per-arch boot2-busybox: image (busybox:musl base, built from scripts/Containerfile.busybox). Drop the alpine pin and the obsolete amd64 PF_X->PF_R hex0-seed patch. * Single unified test runner scripts/run-tests.sh with --suite/--arch; m1pp and p1 suites both fan out across all three arches. * Every program (m1pp, pokem, hello) and every test path drives off a single ARCH= (default aarch64); the lone aarch64 specialisation left is the host-side tools-native target. * Rename p1->P1, m1pp->M1pp, tests/{p1,m1pp}->tests/{P1,M1pp}, p1pp.P1pp->P1pp.P1pp, m1pp.{c,P1}->M1pp.{c,P1}; py generators move into P1/gen/. Repo rebrand lispcc->boot2. * Drop tests/{hello,demo}.M1 (pre-rename register names); the make hello showcase now points at tests/M1pp/00-hello.M1. Verified: 60 m1pp + 15 p1 tests pass on aarch64/amd64/riscv64.

Add amd64 / riscv64 P1v2 generators; explicit ARCHES wiring

2026-04-24T22:54:38Z

commit ffe72f13e339e8411d89985ad324f01a6483a4e6 parent d7398ef3945e0d7f61cc44a36a2cb86a7ab47093 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 15:54:38 -0700 Add amd64 / riscv64 P1v2 generators; explicit ARCHES wiring Mirrors p1/gen/aarch64.py; each backend encodes the same byte sequences as its p1/P1-.M1pp counterpart. Drop the ARCH_REGISTRY indirection in common.py. Each arch module now exposes ARCH = ArchDef(...); p1_gen.py builds an explicit ARCHES dict.

Add readme

2026-04-24T22:25:46Z

commit d7398ef3945e0d7f61cc44a36a2cb86a7ab47093 parent 56144e7b4c2059d04a00c2c87f8f8423f3ba2c1b Author: Ryan Sepassi Date: Fri, 24 Apr 2026 15:25:46 -0700 Add readme

rm old P1 lisp

2026-04-24T22:17:00Z

commit 56144e7b4c2059d04a00c2c87f8f8423f3ba2c1b parent 7e769ab87300ff635a3aaf9227db798f9a642ad6 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 15:17:00 -0700 rm old P1 lisp

Simplify P1-amd64.M1pp helpers; drop native reg tables from P1.md

2026-04-24T22:03:38Z

commit 7e769ab87300ff635a3aaf9227db798f9a642ad6 parent b024f5b4fca4986de572255327cb70230dd51001 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 15:03:38 -0700 Simplify P1-amd64.M1pp helpers; drop native reg tables from P1.md Introduces `amd_maybe_rex_b` to collapse the _lo/_hi/dispatch triplets on push/pop/jmp_r/call_r/mov_imm32_prefix, and a per-name `amd_is_sp` predicate to collapse the 48-macro per-(op,base) `p1_mem` fan-out into a single dispatcher. Native register assignment is unchanged (still C-ABI-aligned for args 0-3). Also removes the per-arch native register mapping tables from docs/P1.md — the P1↔native mapping is backend-internal and was creating the illusion of a portable contract. Target-notes bullets generalized accordingly.

Fold LA_BR into branch/call/tail at P1.M1pp layer

2026-04-24T20:33:54Z

commit b024f5b4fca4986de572255327cb70230dd51001 parent 13564bfed1d3e4f66117aeaf9a24138c72696475 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 13:33:54 -0700 Fold LA_BR into branch/call/tail at P1.M1pp layer The portable P1 spec keeps LA_BR explicit because it makes the target-materialization step uniform across backends. The P1.M1pp ergonomic layer doesn't owe users that uniformity, so collapse the two-step "load branch target / consume" pattern into single macros that take the label as an argument: %b(target), %call(target), %tail(target) %beq(ra, rb, target), %bne, %blt, %bltu %beqz(ra, target), %bnez, %bltz %la(rd, target) %la_br is gone from the user-facing surface; backend hooks (%p1_labr, %p1_b, %p1_call, %p1_condb, %p1_tail) are unchanged. Output binaries are byte-identical for every test fixture on every backend.

Add libp1pp v1 utility library (p1/p1pp.P1pp)

2026-04-24T19:06:51Z

commit 13564bfed1d3e4f66117aeaf9a24138c72696475 parent 455667f35b962f3ef09f2aa17490c4afc458fc6b Author: Ryan Sepassi Date: Fri, 24 Apr 2026 12:06:51 -0700 Add libp1pp v1 utility library (p1/p1pp.P1pp) Implements the contract from docs/LIBP1PP.md: control-flow macros (%if_/ifelse_/while_/do_while_/for_lt/loop plus tagged variants with %break/%continue), %fn function-definer using M1PP scopes, %assert_ macros, and C-level primitives for bytes/strings/ints, character predicates, raw syscalls, print helpers, file helpers, bump allocator, and panic. Targets P1v2-64 only. Verified with smoke tests on aarch64, amd64, and riscv64.

Bump M1PP macro cap from 256 to 512

2026-04-24T19:06:44Z

commit 455667f35b962f3ef09f2aa17490c4afc458fc6b parent e2b9951654ab6ba3250b936f286753bb0ca5e274 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 12:06:44 -0700 Bump M1PP macro cap from 256 to 512 Raise MAX_MACROS in the host C expander and M1PP_MACROS_CAP in the self-hosted m1pp.P1 (byte cap doubled to 0x25000 for 512 × 296-byte records). amd64 backend + P1 frontend + libp1pp together exceed 256.

Add %scope / %endscope and :: label rewrite to M1PP

2026-04-24T18:56:04Z

commit e2b9951654ab6ba3250b936f286753bb0ca5e274 parent 87e3956d3b23b7ef1e537e75fc304782bd3df6bb Author: Ryan Sepassi Date: Fri, 24 Apr 2026 11:56:04 -0700 Add %scope / %endscope and :: label rewrite to M1PP A lexical scope stack driven by `%scope NAME` / `%endscope`. A WORD token starting with `::` emits as `:scope1__..__scopeN__name` at output time; `&::name` emits the reference form. With an empty stack the sigil+name passes through. Resolution is emit-time, so `::foo` inside a macro body resolves against the caller's scope — making generic `%break()` / `%continue()` possible. Implemented in both `m1pp/m1pp.c` (reference) and `m1pp/m1pp.P1` (self-hosted). New tests/m1pp/17-scopes covers empty-stack pass-through, nested scopes, scope name from a macro argument, anaphoric break / continue, innermost-wins for nested scope-introducing macros, and `:@` + `::` coexisting in one macro. Also extend p1_gen.py MEM_OFFS with small positive offsets (2–6) so the scope check can use `lb_*,*,2` directly; regenerated `build/p1v2/aarch64/p1_aarch64.M1` (gitignored). Doc: new Scoped labels section in docs/M1PP.md, updated Directives, Limits, and Errors. docs/LIBP1PP.md's %fn / tagged-loop entries rewritten to reflect scopes being available. Also scrubs pre-existing phasing / section-number comments (Phase N, §N, M1PP-EXT) from m1pp.P1 and every tests/m1pp fixture, for a cleaner description of current behavior.

LIBP1PP.md

2026-04-24T18:11:17Z

commit 87e3956d3b23b7ef1e537e75fc304782bd3df6bb parent ff0fd1204fcb984c90ec5ba308bef63a3d04d76a Author: Ryan Sepassi Date: Fri, 24 Apr 2026 11:11:17 -0700 LIBP1PP.md

Rename .M1→.P1 and .P1→.P1pp to reflect toolchain level

2026-04-24T17:41:05Z

commit ff0fd1204fcb984c90ec5ba308bef63a3d04d76a parent 5bb6a0ab4ea4e61516278f007bc52516f074d154 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 10:41:05 -0700 Rename .M1→.P1 and .P1→.P1pp to reflect toolchain level .P1 — hand-written M1 following P1v2 ABI; consumed by M0 directly after the p1_arch.M1 DEFINE table is prepended (no m1pp needed). .P1pp — P1 source using M1PP macros (%li, %syscall, …); consumed by m1pp with P1 backend prepended, then M0. m1pp/m1pp.M1 → m1pp/m1pp.P1 pokem/pokem.M1 → pokem/pokem.P1 tests/p1/*.P1 → tests/p1/*.P1pp Update build.sh, build-p1.sh, test.sh (m1pp/ and tests/p1/), and Makefile comment to use the new extensions.

Port m1pp/pokem to p1_main and emit :_start from p1_gen.py

2026-04-24T17:23:09Z

commit 5bb6a0ab4ea4e61516278f007bc52516f074d154 parent 5c810d428f71b2b544e9fa4eb524eae88c838895 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 10:23:09 -0700 Port m1pp/pokem to p1_main and emit :_start from p1_gen.py The P1v2 program-entry model (docs/P1.md §Program Entry) has a backend-owned :_start stub call p1_main with a0=argc, a1=argv. The m1pp-driven path already did this via p1/P1-aarch64.M1pp's p1_entry macro; the raw-M1 path going through build/p1v2/aarch64/p1_aarch64.M1 still had each program define its own :_start and read argv by leaning on the portable sp+16 bias. Emit the aarch64 :_start stub from p1/p1_gen.py (via a new start_stub field on ArchDef), and convert m1pp.M1, pokem.M1, and the 00-hello smoke fixture to define :p1_main instead.

Add pokem, a P1 raw-byte file patcher

2026-04-24T16:51:49Z

commit 5c810d428f71b2b544e9fa4eb524eae88c838895 parent 30273099f630eb2e8f787c413fc863ad2e44edfa Author: Ryan Sepassi Date: Fri, 24 Apr 2026 09:51:49 -0700 Add pokem, a P1 raw-byte file patcher pokem overlays ASCII-hex-decoded bytes onto a file at a given decimal offset. The primary use case is patching ELF header fields after assembly — e.g. bumping p_memsz to reserve a BSS region without emitting literal zero bytes in the source. Implementation notes: - Single fd with O_RDWR; advances position via read-into-scratch, then writes in place. No in-memory file buffer, so pokem itself is ~1.8 KB and handles files of any size. - Bootstrap-tool posture: no input validation, no syscall error checking. Branchless hex_nibble decode via (c & 15) + 9 * (c >> 6) inlined into parse_hex_bytes, which is a leaf with no BSS spill. - Only uses the P1 common syscall surface (openat, read, write, close, exit).

Hide P1 frame header and merge LEAVE+RET into ERET

2026-04-24T16:03:03Z

commit 30273099f630eb2e8f787c413fc863ad2e44edfa parent d19a402ecc3df0f912aecf4ec2a8d400f26c5f05 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 09:03:03 -0700 Hide P1 frame header and merge LEAVE+RET into ERET Portable sp after ENTER points to the frame-local base; the saved retaddr and saved caller sp become backend-private. LEAVE is dropped as a standalone op — ERET atomically tears down the frame and returns, mirroring TAIL/TAILR which already bundle the epilogue. Leaves still return with bare RET.

Expand post with missing m1pp/P1 features and P1 source examples

2026-04-24T15:09:38Z

commit d19a402ecc3df0f912aecf4ec2a8d400f26c5f05 parent 4eba962ec4bb1f6874f10d0ebd5f62371b1a1b7a Author: Ryan Sepassi Date: Fri, 24 Apr 2026 08:09:38 -0700 Expand post with missing m1pp/P1 features and P1 source examples Adds bullets for %struct/%enum, local labels, and %str in the m1pp section; a program-entry paragraph in the P1 section; and a new 'Three programs' section walking through argc_exit, hello, and double with their source inlined.

Add argc_exit and double P1 test fixtures

2026-04-24T15:09:32Z

commit 4eba962ec4bb1f6874f10d0ebd5f62371b1a1b7a parent bf0cbf2005df3b5c84d48c4efa206567563dfa3a Author: Ryan Sepassi Date: Fri, 24 Apr 2026 08:09:32 -0700 Add argc_exit and double P1 test fixtures argc_exit: bare %ret() returns argc as the process exit status via the backend entry stub. double: calls a :double helper through LA_BR + CALL with an ENTER/LEAVE frame in p1_main, exercising the non-leaf path the spec now requires.

Require active frame for CALL/CALLR in P1 spec

2026-04-24T15:09:27Z

commit bf0cbf2005df3b5c84d48c4efa206567563dfa3a parent aa0c83900e7fb42f42a0361f46597e9bc5aabbaf Author: Ryan Sepassi Date: Fri, 24 Apr 2026 08:09:27 -0700 Require active frame for CALL/CALLR in P1 spec Previously the spec said call/return correctness didn't depend on a frame, but the aarch64 and riscv64 backends both lower CALL to a bare BLR/JALR that clobbers the native link register. A non-leaf caller without a frame would infinite-loop on RET. Tighten the contract: any function that issues CALL, CALLR, TAIL, or TAILR must ENTER a standard frame first. Pure leaves may still omit the frame. Existing backends are already correct under the new wording.

Add riscv64 and amd64 P1v2 backends with multi-arch test harness

2026-04-24T14:54:52Z

commit aa0c83900e7fb42f42a0361f46597e9bc5aabbaf parent a570d90096e6f998b88f626c4ea05d870e80e84f Author: Ryan Sepassi Date: Fri, 24 Apr 2026 07:54:52 -0700 Add riscv64 and amd64 P1v2 backends with multi-arch test harness - p1/P1-riscv64.M1pp, p1/P1-amd64.M1pp: new M1pp-form P1v2 backends mirroring p1/P1-aarch64.M1pp. Native register picks follow docs/P1.md's 64-bit mapping table. rv64 lowers every op to a single R/I/S-type instruction or a source-before-destination sequence, so no aliasing hazards are possible. amd64 handles variable-length encoding (REX, ModRM, SIB) and the native syscall/shift register constraints (rcx/r11 clobbered by syscall; CL-only shift counts; rdx implicit in idiv) via stack saves and a hidden `scratch` / `rbp` pair. - m1pp/build-p1.sh: per-arch pipeline (host-native m1pp -> M1 -> hex2 with the matching stage0 ELF header). The m1pp expander is architecture-neutral, so one host-compiled binary drives all targets. - tests/p1/test.sh: rewritten to iterate over every fixture for every backend under podman/qemu (`--arch` narrows to one target). Cross-arch consistency falls out because every arch must match the same .expected stdout. - tests/p1/p1-call.P1: exercises ENTER / LEAVE / CALL / RET / MOV / ADDI on a program with a helper subroutine. - tests/p1/p1-aliasing.P1: regression coverage for three amd64 contract violations fixed in this commit. Each test picks a register pattern that makes the bug observable as a wrong output byte: 1. DIV/REM with rd == rb == a2 (rdx). Two bugs in one sequence -- `idiv rb` ran after `cqo` clobbered rdx (so idiv divided by the sign-extension of rax instead of the original rb), and the rdx restore ran after the rd write (clobbering the quotient when rd == a2). Fixed by stashing rb to `scratch` before `cqo` and reordering the restore to precede the rd write. 2. SHL/SHR/SAR reg-count with rd == a3 (rcx). Mirror of (1) -- ra was read from rcx after the count overwrite, and the rcx restore ran after the rd write. Fixed by the same reordering. 3. ANDI/ORI with imm > 127 unconditionally used `83 /ext ib`, whose sign-extension turned `ANDI rd, 255` into AND with -1. Widened to `81 /ext id` for imms outside [-128, 127]. - Makefile: new `test-p1` target invoking tests/p1/test.sh. 15 passed, 0 failed across aarch64 / amd64 / riscv64 on hello, double, argc_exit, p1-call, p1-aliasing.

Speed up P1 build/test loop ~30-70x via native mescc-tools

2026-04-24T12:44:59Z

commit a570d90096e6f998b88f626c4ea05d870e80e84f parent 3b6a3c115744b6b268f83c3a7e122ab4155e78c2 Author: Ryan Sepassi Date: Fri, 24 Apr 2026 05:44:59 -0700 Speed up P1 build/test loop ~30-70x via native mescc-tools Default path for m1pp/build.sh now runs host-compiled mescc-tools M1/hex2 instead of the stage0 M0/hex2-0 inside the linux/arm64 VM, cutting a full P1 test from ~110s to ~3.9s cold / ~1.6s warm. The bootstrap tools still produce byte-exact output and stay the source of truth; set M1PP_BOOTSTRAP_TOOLS=1 to force the original container path. Also caches the built m1pp/m1pp by mtime in both test runners so repeat runs skip the full expander rebuild when m1pp.M1 and the P1 defs are unchanged.

Add P1v2 program-entry model (aarch64 _start stub calling p1_main)

2026-04-24T04:25:10Z

commit 3b6a3c115744b6b268f83c3a7e122ab4155e78c2 parent f10eab299b822884ec68f4c6b3b8285859515ca9 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 21:25:10 -0700 Add P1v2 program-entry model (aarch64 _start stub calling p1_main) P1.M1pp now unconditionally invokes the backend %p1_entry() macro, which on aarch64 emits :_start: read argc from [sp], compute argv=sp+8, call p1_main, sys_exit its return value. Portable sources just define :p1_main; hello.P1 is converted accordingly.

Update Lisp plan for P1

2026-04-24T04:14:37Z

commit f10eab299b822884ec68f4c6b3b8285859515ca9 parent 2bf50d2a8f56b2106a737e260c4900abfb98d07a Author: Ryan Sepassi Date: Thu, 23 Apr 2026 21:14:37 -0700 Update Lisp plan for P1

M1PP doc

2026-04-24T03:28:36Z

commit 2bf50d2a8f56b2106a737e260c4900abfb98d07a parent 687ee15acc1661be78ff6cd8922ad95dd19a2100 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 20:28:36 -0700 M1PP doc

rm old docs

2026-04-24T03:15:53Z

commit 687ee15acc1661be78ff6cd8922ad95dd19a2100 parent 79aedea7287b895e3089ff72a8ff80482bda1083 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 20:15:53 -0700 rm old docs

update post

2026-04-24T03:09:23Z

commit 79aedea7287b895e3089ff72a8ff80482bda1083 parent 592d347926f1abf565a5cd11a360a18a3a645432 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 20:09:23 -0700 update post

shell.scm

2026-04-24T03:09:14Z

commit 592d347926f1abf565a5cd11a360a18a3a645432 parent 2dfba4fcce1a94138f06d44d816896ca76f89065 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 20:09:14 -0700 shell.scm

lisp docs

2026-04-24T03:08:40Z

commit 2dfba4fcce1a94138f06d44d816896ca76f89065 parent 955cc50921de523481bdfc77a37f49725a4cb554 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 20:08:40 -0700 lisp docs

m1pp.M1: port %struct and %enum directives

2026-04-24T01:22:32Z

commit 955cc50921de523481bdfc77a37f49725a4cb554 parent 87d0cf252b91a60d2736c898ed9e5a08fa24c09a Author: Ryan Sepassi Date: Thu, 23 Apr 2026 18:22:32 -0700 m1pp.M1: port %struct and %enum directives Adds :define_fielded + :df_emit_field + :df_render_decimal to the P1 expander, plus line-start dispatch branches (:proc_check_struct, :proc_check_enum) that shim into define_fielded with the stride / totalizer-name parameter pair. State lives in df_* BSS slots + df_name_scratch (256 B) + df_digit_scratch (24 B) so append_text calls don't clobber it. Test suite: all 19 m1pp fixtures pass byte-identical against the C oracle.

m1pp: add %struct and %enum directives

2026-04-24T01:22:26Z

commit 87d0cf252b91a60d2736c898ed9e5a08fa24c09a parent dd82213eb8aa8db05b2bfaddfde3317d69f80d86 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 18:22:26 -0700 m1pp: add %struct and %enum directives Implements §5 and §6 of docs/M1PP-EXT.md. Both share define_fielded(), parameterized by stride and totalizer name: %struct NAME { f1 f2 ... } -> NAME.field_k = k*8, NAME.SIZE = N*8 %enum NAME { l1 l2 ... } -> NAME.label_k = k, NAME.COUNT = N Each synthesized field expands as a zero-parameter macro, consumed via paren-less access (§4) at use sites. Two fixtures cover the 4-field struct + %frame wrapper composition and a 7-label enum with a %(=) compile-time COUNT sanity check.

tests/m1pp/12-braced-args: compose §1 local labels in WHILE_NEZ

2026-04-24T00:44:16Z

commit dd82213eb8aa8db05b2bfaddfde3317d69f80d86 parent 6bb13576cb3eaa4c6bf33ad1104009f019b79512 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:44:16 -0700 tests/m1pp/12-braced-args: compose §1 local labels in WHILE_NEZ Replace the literal `:loop__` label (which would collide if the macro were invoked twice) with `:@loop` / `&@loop`, demonstrating §1+§2 composing. WHILE_NEZ is expansion id 2 in this fixture, so the output becomes `:loop__2` / `&loop__2`.

Merge feature: braced block arguments (§2)

2026-04-24T00:22:45Z

commit 6bb13576cb3eaa4c6bf33ad1104009f019b79512 parent 04089609068b2758302653f0bf1488525c8eb499 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:22:45 -0700 Merge feature: braced block arguments (§2)

Merge feature: local labels :@name / &@name (§1)

2026-04-24T00:22:42Z

commit 04089609068b2758302653f0bf1488525c8eb499 parent 51e51fb86495f1a6cbfb8fde625b5db1eab2dac1 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:22:42 -0700 Merge feature: local labels :@name / &@name (§1)

Merge feature: %str stringification builtin (§7)

2026-04-24T00:22:38Z

commit 51e51fb86495f1a6cbfb8fde625b5db1eab2dac1 parent 6def9f0029040c6df23519bf58d44b45de20a5d6 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:22:38 -0700 Merge feature: %str stringification builtin (§7)

Merge feature: strlen expression op (§3)

2026-04-24T00:22:35Z

commit 6def9f0029040c6df23519bf58d44b45de20a5d6 parent 5b828f7a0aa7a329a933fedb0576854a30bddd03 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:22:35 -0700 Merge feature: strlen expression op (§3)

Merge feature: paren-less 0-arg macro calls (§4)

2026-04-24T00:22:30Z

commit 5b828f7a0aa7a329a933fedb0576854a30bddd03 parent dca693c3073799b8d46cc9f6479839ea4c8d913d Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:22:30 -0700 Merge feature: paren-less 0-arg macro calls (§4)

m1pp.M1: port %str stringification builtin

2026-04-24T00:21:42Z

commit 5b1416eeda2c4dc29373ef07c50dd1106b4708db parent ba016341a809dc6bc748baa48219eabb8bf8fdea Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:21:42 -0700 m1pp.M1: port %str stringification builtin

m1pp.M1: port strlen expression op

2026-04-24T00:20:31Z

commit d0c704cfef090603c28aae04be29a82233d07761 parent ce22557fc8cb164acd92288a244669cc4651ec03 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:20:31 -0700 m1pp.M1: port strlen expression op

m1pp.M1: port local labels (:@name, &@name)

2026-04-24T00:20:19Z

commit cf5649c34ff07930d8d5bff5b9c6f37d332d5aba parent 48971a6c16a856b6b0fd6beafa72d2a170cab648 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:20:19 -0700 m1pp.M1: port local labels (:@name, &@name)

m1pp.M1: port braced block arguments

2026-04-24T00:19:25Z

commit 0948ad4bcc40d174469f825a24cf35415eea6f44 parent 968ba690fb1865bd72df6f2486c58db5807445f6 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:19:25 -0700 m1pp.M1: port braced block arguments

m1pp.M1: allow paren-less invocation of 0-arg macros

2026-04-24T00:17:40Z

commit 61dd8df03718c8672dbc70ace7d9d846e2f06a6d parent 09299304297f9244d866289eafc912422be489e7 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:17:40 -0700 m1pp.M1: allow paren-less invocation of 0-arg macros

m1pp: add %str stringification builtin

2026-04-24T00:06:38Z

commit ba016341a809dc6bc748baa48219eabb8bf8fdea parent dca693c3073799b8d46cc9f6479839ea4c8d913d Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:06:38 -0700 m1pp: add %str stringification builtin

m1pp: allow paren-less invocation of 0-arg macros

2026-04-24T00:05:40Z

commit 09299304297f9244d866289eafc912422be489e7 parent dca693c3073799b8d46cc9f6479839ea4c8d913d Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:05:40 -0700 m1pp: allow paren-less invocation of 0-arg macros

m1pp: add braced block arguments

2026-04-24T00:05:37Z

commit 968ba690fb1865bd72df6f2486c58db5807445f6 parent dca693c3073799b8d46cc9f6479839ea4c8d913d Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:05:37 -0700 m1pp: add braced block arguments

m1pp: add strlen expression op

2026-04-24T00:04:40Z

commit ce22557fc8cb164acd92288a244669cc4651ec03 parent dca693c3073799b8d46cc9f6479839ea4c8d913d Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:04:40 -0700 m1pp: add strlen expression op

m1pp: add local labels (:@name, &@name)

2026-04-24T00:04:31Z

commit 48971a6c16a856b6b0fd6beafa72d2a170cab648 parent dca693c3073799b8d46cc9f6479839ea4c8d913d Author: Ryan Sepassi Date: Thu, 23 Apr 2026 17:04:31 -0700 m1pp: add local labels (:@name, &@name)

m1pp: spell XOR as ^ (was \$); drop == alias for =

2026-04-23T23:58:19Z

commit dca693c3073799b8d46cc9f6479839ea4c8d913d parent 6761b0608d69551ae4bfe6e9b5a932ddcafb7b26 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 16:58:19 -0700 m1pp: spell XOR as ^ (was \$); drop == alias for = XOR is now ^ — the conventional spelling, no longer collides visually with the \$(...) 8-byte hex builtin marker. Touches both m1pp/m1pp.c and m1pp/m1pp.M1 (renamed op_dollar -> op_caret in rodata, swapped the dispatch comparison). = and == previously both returned EXPR_EQ. The expression language has no assignment so there's no ambiguity to resolve, but two ways to spell one thing is just noise. Dropped == (eliminated the dispatch block in expr_op_code and the op_eq_eq rodata). Updated tests/m1pp/04-expr-ops.M1pp accordingly and regenerated the .expected. All 11 m1pp parity fixtures and the P1 hello fixture pass.

m1pp extensions doc

2026-04-23T23:53:48Z

commit 6761b0608d69551ae4bfe6e9b5a932ddcafb7b26 parent ce7dbc4c1ff6509f28aa721ebffdffda6b6bb051 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 16:53:48 -0700 m1pp extensions doc

m1pp/m1pp.M1: scrub evolutionary comments

2026-04-23T23:48:36Z

commit ce7dbc4c1ff6509f28aa721ebffdffda6b6bb051 parent a52dc7712af059af01946616204fd6aebefc7652 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 16:48:36 -0700 m1pp/m1pp.M1: scrub evolutionary comments Comments now describe current state and intent only. No phase numbers, no track / layer / oracle references, no "this used to do X" narratives. Section dividers renamed from "Phase N STUBS:" to topic-named blocks (e.g. "Stream stack + expansion-pool lifetime", "Argument parsing"). Inline `Oracle: m1pp.c:NAME.` per-function refs deleted; oracle-anchored prose ("matches C oracle", "the oracle spells xor as $", "the C oracle uses 512-byte tmp") rephrased to state the rule directly. Stale sizing notes (32 macros, 256 expand slots) updated to match current caps. Behavior unchanged: all 11 m1pp oracle-parity fixtures and the P1 hello fixture still pass.

tests/p1: hello.P1 + harness exercising the full P1 -> ELF pipeline

2026-04-23T23:27:51Z

commit a52dc7712af059af01946616204fd6aebefc7652 parent adbf3758d169db6ab5e662c4b1e436fcdfd9da06 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 16:27:51 -0700 tests/p1: hello.P1 + harness exercising the full P1 -> ELF pipeline First end-to-end test that drives a P1-language program through every layer of the stack: cat p1/P1-aarch64.M1pp p1/P1.M1pp tests/p1/hello.P1 -> m1pp (M1 build) -> .M1 macro / builtin substitution -> m1pp/build.sh -> aarch64 ELF lint, M0, hex2-0 podman run binary -> diff stdout vs .expected hello.P1 is the smallest useful program: write(stdout, "Hello, World!\n") then exit(0), spelled with the P1 frontend's macros (%li, %la, %syscall, %sys_write, %sys_exit). Proves that: - the cat-three-files convention works (define-before-use holds across arch backend, portable frontend, and user program) - m1pp accepts inputs above its previous 8 KB cap (combined source is ~15 KB), and that its emit_hex_value output round-trips through M0 (the recent quoted-hex change makes this work) - the full toolchain produces a runnable ELF that the kernel actually invokes Filenames starting with `_` are skipped (parked), matching m1pp/test.sh.

m1pp: emit !@%$ hex builtins as quoted strings; bump buffers for P1

2026-04-23T23:27:38Z

commit adbf3758d169db6ab5e662c4b1e436fcdfd9da06 parent 611ad4d2c08121778de01ec03e8e59898617c458 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 16:27:38 -0700 m1pp: emit !@%$ hex builtins as quoted strings; bump buffers for P1 Two infrastructure changes needed to send the P1 frontend through the m1pp -> M0 -> hex2 pipeline: 1. emit_hex_value now wraps its output in single quotes, e.g. !(0xFF) emits 'FF' instead of FF. M0 treats bare alphanumeric tokens as decimal numerals (so a literal 40000058 from %((| 0x58000040 0)) was being mis-parsed as decimal 40000058 -> 1 byte 0x3A) but passes 'XX..' string literals through as raw hex bytes, matching the format DEFINE values use. Updated both m1pp/m1pp.c (emit kind = TOK_STRING; allocate +2 chars for the quotes) and the M1 port (same shape; ehv_scratch grown from 16 to 24 bytes for the two extra quote chars). Regenerated the affected .expected fixtures with the updated oracle: 03-builtins, 04-expr-ops, 05-int-atoms, 07-rescan, 10-full-parity. 2. Bumped the m1pp BSS caps so the expander can swallow realistic P1 inputs (the combined p1/P1-aarch64.M1pp + p1/P1.M1pp + a tiny user program is ~15 KB, well above the previous 8 KB INPUT_CAP): INPUT_CAP 8 KB -> 64 KB OUTPUT_CAP 8 KB -> 64 KB TEXT_CAP 4 KB -> 32 KB TOKENS_END 6 KB -> 96 KB (4096 tokens) MACROS_CAP 9472 B -> 76 KB (256 macros) MACRO_BODY_CAP 6 KB -> 96 KB (4096 tokens) EXPAND_CAP 6 KB -> 96 KB (4096 tokens) The corresponding ZERO32-padded BSS arenas were resized to match. All 11 m1pp oracle-parity fixtures still pass.

tests/m1pp: add 7 oracle-parity fixtures + unpark full-parity

2026-04-23T22:52:15Z

commit 611ad4d2c08121778de01ec03e8e59898617c458 parent 64c477b2cbf7684e884fa53252a8d89b2a09300c Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:52:15 -0700 tests/m1pp: add 7 oracle-parity fixtures + unpark full-parity Each .expected was generated by running m1pp/m1pp.c (the oracle) on the fixture, so a passing test means the M1 implementation is byte-for-byte identical to the C reference on that input. 03-builtins - !@%$ at 1/2/4/8-byte sizes, including high-bit values 04-expr-ops - every operator: variadic folds, unary -, /, %, shifts, ~, all 6 comparisons, nested expressions 05-int-atoms - decimal/hex literals, negatives, multi-digit, high-bit wrap 06-paste - multi-paste chain, prefix/suffix patterns, mix on one line 07-rescan - macro-in-macro, macro arg holding a call, macro-in-expression 08-select - empty branches, multi-token branches, branch with macro call, nested %select 09-args - nested parens, depth-1-only commas, 0-param, 16-param, string token containing a comma Unparks the existing _full-parity fixture (now 10-full-parity) - it has been passing oracle parity since the integration completed.

Integrate fix: spell negative addi immediates as negN (lint blind spot)

2026-04-23T22:43:37Z

commit 64c477b2cbf7684e884fa53252a8d89b2a09300c parent 84ebd78783a6bbc55059ea0cac96188ac7334c3d Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:43:37 -0700 Integrate fix: spell negative addi immediates as negN (lint blind spot)

Merge branch 'worktree-agent-aecb01d58b541506d' into integrate-m1pp

2026-04-23T22:37:46Z

commit 84ebd78783a6bbc55059ea0cac96188ac7334c3d parent fbf22f4bdb7af44f88f457580832a4d6b925229d Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:37:46 -0700 Merge branch 'worktree-agent-aecb01d58b541506d' into integrate-m1pp # Conflicts: # m1pp/m1pp.M1

Merge branch 'worktree-agent-a04674a83d5e97483' into integrate-m1pp

2026-04-23T22:36:42Z

commit fbf22f4bdb7af44f88f457580832a4d6b925229d parent 4d156315d79d8c10153cdde943b0edb4bd13c603 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:36:42 -0700 Merge branch 'worktree-agent-a04674a83d5e97483' into integrate-m1pp # Conflicts: # m1pp/m1pp.M1

Merge branch 'worktree-agent-a63654c7cd13786ed' into integrate-m1pp

2026-04-23T22:36:03Z

commit 4d156315d79d8c10153cdde943b0edb4bd13c603 parent e0a71b2419b2d73e5e8c8c5b453e290c69df7d54 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:36:03 -0700 Merge branch 'worktree-agent-a63654c7cd13786ed' into integrate-m1pp # Conflicts: # m1pp/m1pp.M1

Merge branch 'worktree-agent-a55e267523e077156' into integrate-m1pp

2026-04-23T22:35:28Z

commit e0a71b2419b2d73e5e8c8c5b453e290c69df7d54 parent 1c9de593e0c3ec9dcc53712e78aa6353f0aaf41a Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:35:28 -0700 Merge branch 'worktree-agent-a55e267523e077156' into integrate-m1pp

Track E: builtins (! @ % $ %select)

2026-04-23T22:35:15Z

commit 2f746287e685ff08457e14c57f847d7a8d6ffc31 parent 9507a3668cd1218bae1aa0f171e7991cc8768afd Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:35:15 -0700 Track E: builtins (! @ % $ %select)

Track D: expression evaluator

2026-04-23T22:35:15Z

commit edcc64c29178d89a6da0b97ec38e588c79690ece parent 9507a3668cd1218bae1aa0f171e7991cc8768afd Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:35:15 -0700 Track D: expression evaluator

Track C: macro expansion (parse_args, find_macro/param, expand_*)

2026-04-23T22:35:15Z

commit 49216446b528bf525f65f87772edea67dbe3d5d6 parent 9507a3668cd1218bae1aa0f171e7991cc8768afd Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:35:15 -0700 Track C: macro expansion (parse_args, find_macro/param, expand_*)

Track B: ## token paste compaction

2026-04-23T22:35:14Z

commit 02cef1ca5ab2161f25f7761d5e54e8bffd0ea57d parent 9507a3668cd1218bae1aa0f171e7991cc8768afd Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:35:14 -0700 Track B: ## token paste compaction

Track A: stream stack + pool lifetime + process_tokens rewrite

2026-04-23T22:35:09Z

commit 1c9de593e0c3ec9dcc53712e78aa6353f0aaf41a parent 9507a3668cd1218bae1aa0f171e7991cc8768afd Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:35:09 -0700 Track A: stream stack + pool lifetime + process_tokens rewrite

m1pp stubs

2026-04-23T22:01:53Z

commit 9507a3668cd1218bae1aa0f171e7991cc8768afd parent de636ed71441d5dd25d5e73eb0bc1cb808acdc17 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 15:01:53 -0700 m1pp stubs

m1pp: Phase 2 — store %macro definitions in arenas

2026-04-23T21:45:59Z

commit de636ed71441d5dd25d5e73eb0bc1cb808acdc17 parent fa9eb5967ef6ac61d3711e3687f7b2ffdbaf8767 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 14:45:59 -0700 m1pp: Phase 2 — store %macro definitions in arenas Replace the structural %macro/%endm skip in m1pp.M1 with a real define_macro that parses the header (name, comma-separated params, trailing newline) and copies body tokens into macro_body_tokens[] until a line-start %endm. Records live in a 32-slot macros[] arena (296 B/record) with running tail pointers. Macros are still not invoked; defs-only input continues to match the C oracle byte-for-byte. Adds tests/m1pp/02-defs.{M1pp,expected} as the Phase 2 parity fixture.

docs/M1M-IMPL.md: reconcile with post-Phase-1 state

2026-04-23T21:32:34Z

commit fa9eb5967ef6ac61d3711e3687f7b2ffdbaf8767 parent 2eb0c7d9d788f25032944b92823867115b2402d7 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 14:32:34 -0700 docs/M1M-IMPL.md: reconcile with post-Phase-1 state Update the implementation sketch to match what's now on disk: - m1pp.M1 description reflects the P1v2 port landing in Phase 1. - Manual-run example and Phase 0 blurb use localhost/distroless-busybox:latest (built from Containerfile.busybox) instead of the old localhost/lispcc:aarch64 alpine alias. - Oracle-compare commands and the Supported Features / Phase 7 / Phase 10 bullets use .M1pp instead of the old .M1M extension, matching the fixture rename and the p1/*.M1pp sources. - Layer 9 pseudo-code advances s.pos += 24 (matching the 24-byte Token record laid out in Core Data Structures) instead of the old 32.

post updates

2026-04-23T21:30:07Z

commit 2eb0c7d9d788f25032944b92823867115b2402d7 parent 992c8cc15c1e7ae225711195b82fbadcf384327a Author: Ryan Sepassi Date: Thu, 23 Apr 2026 14:30:07 -0700 post updates

m1pp: switch build.sh and test.sh to a distroless-busybox image

2026-04-23T21:29:51Z

commit 992c8cc15c1e7ae225711195b82fbadcf384327a parent 7f8ceb9a19fdfb5f894827e596261da1d3c5fb87 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 14:29:51 -0700 m1pp: switch build.sh and test.sh to a distroless-busybox image Replace the alpine base with a two-stage distroless-static image that pulls only busybox from another distroless layer. Both digests are pinned. Gives us the minimal sh/cp/chmod surface needed by build.sh's container step without shipping apk or any of alpine's userland. build.sh now `podman build`s the image from Containerfile.busybox on first run instead of aliasing a pinned alpine digest. test.sh shares the same tag so parity runs and build-pipeline smoke runs execute under one image.

m1pp: annotate m1pp.M1 with pseudo-C inline comments

2026-04-23T21:28:39Z

commit 7f8ceb9a19fdfb5f894827e596261da1d3c5fb87 parent 19c3c3bd465f5e192537c247239a4d27497536e7 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 14:28:39 -0700 m1pp: annotate m1pp.M1 with pseudo-C inline comments Each logical chunk of assembly now carries a single-line # comment in pseudo-C form: register/field loads as "name = expr", branches as "if (cond) ...", loops as "for (i = 0; i < len; i++) ...", token field accesses as "tok->kind" / "tok->text_ptr", etc. Function-header ## comments call out args, return value, leaf/non- leaf status, and relevant invariants (token layout, the shared shape of the lparen/rparen/comma dispatch, the caller-saved register juggling in fatal). Distinct hash levels (# vs ##) keep inline annotations from colliding with the structural section dividers. No code changes; tests still pass.

m1pp: add structural section comments in m1pp.M1

2026-04-23T21:00:31Z

commit 19c3c3bd465f5e192537c247239a4d27497536e7 parent c0792835d2bf76ee366e38128102228745deceb0 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 14:00:31 -0700 m1pp: add structural section comments in m1pp.M1 Ten section dividers (constants, runtime shell, helpers, lexer, output, main processor, %macro skip, error paths, rodata, BSS) each with a short note on what the block does and how it composes with its neighbors. The file header docstring now sketches the pipeline from _start through lex_source, process_tokens, skip_macro_def. No code changes; tests still pass.

m1pp: Phase 1 — port lexer + pass-through skeleton to P1v2

2026-04-23T20:55:03Z

commit c0792835d2bf76ee366e38128102228745deceb0 parent cddd94f916f41fad819f4905b6508be0378368d1 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 13:55:03 -0700 m1pp: Phase 1 — port lexer + pass-through skeleton to P1v2 Rewrite m1pp/m1pp.M1 against the P1v2 ISA (a*/t*/s* registers, enter_0/leave, la_br &label). Covers _start, read/write, lex_source, emit_token, emit_newline, process_tokens, structural %macro/%endm skip, plus the append_text / push_source_token / tok_eq_const helpers the oracle lifts for this slice. BSS now precedes :ELF_end so the LOAD segment's filesz = memsz covers it (the stage0 aarch64 header doesn't support memsz > filesz). Simplify tests/m1pp/ fixture format: shape is picked by extension. .M1 is a standalone P1v2 program; stdout is diffed against .expected. .M1pp is expander input; runner builds m1pp/m1pp.M1 once, runs it as `m1pp .M1pp `, diffs the output file against .expected. Filenames starting with `_` are skipped — used here to park the pre-existing full-parity fixture until Phases 2–10 land. tests/m1pp/01-passthrough.M1pp is the Phase 1 parity fixture: a definition-only input that exercises the lexer, pass-through emit path, and structural %macro skip against the C oracle.

m1pp: alias container image as localhost/lispcc:aarch64; add port pointers

2026-04-23T20:25:10Z

commit cddd94f916f41fad819f4905b6508be0378368d1 parent 049c1ea756097264ad88f7a22003d26259543ee3 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 13:25:10 -0700 m1pp: alias container image as localhost/lispcc:aarch64; add port pointers Tag the digest-pinned alpine arm64 image as localhost/lispcc:aarch64 on first run from m1pp/build.sh; both build.sh and test.sh now run podman against the short tag. The Makefile keeps the digest as the canonical pin; the tag is just an alias for ergonomics so the long SHA doesn't have to appear in commands or docs. Expand docs/M1M-IMPL.md with a "Working on the port" section: file map, command cheatsheet (build, test, manual container run, regen defs, C-oracle parity, lint), and a P1v2 quick-reference. Each phase now lists the oracle entry points in m1pp/m1pp.c that the M1 port should lift, so a session picking up mid-port can find the relevant C routines by symbol name.

m1pp: Phase 0 build infra; split aarch64 LI/LA literal prefix

2026-04-23T20:19:28Z

commit 049c1ea756097264ad88f7a22003d26259543ee3 parent 92f3cc91a2b73dc490ae6048309beb6da7382c96 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 13:19:28 -0700 m1pp: Phase 0 build infra; split aarch64 LI/LA literal prefix Add m1pp/build.sh and m1pp/test.sh: build a P1v2 .M1 fixture against build/p1v2/aarch64/p1_aarch64.M1, run it in the aarch64 alpine container, diff against .expected. Wired in via `make test-m1pp`. Phase 0 fixture tests/m1pp/00-hello.M1 is a hello-world that exercises the full pipeline end to end. Split the aarch64 literal-pool prefix. LI stays 64-bit (ldr x, b PC+12) so any one-word constant fits. LA and LA_BR move to a 32-bit zero- extending prefix (ldr w, b PC+8). Label addresses fit in 32 bits under the stage0 layout, so source now writes `la_a2 &msg` without padding the literal pool entry to 8 bytes. Codified in docs/P1v2.md. Moves the full-parity fixture from tests/m1m/ to tests/m1pp/ so per- phase fixtures live together. The prior 64-bit prefix had b PC+8 where it needed b PC+12 and would have SIGILL'd any call site; no in-tree code had run against it, so the bug surfaced only when Phase 0 drove it.

move files around

2026-04-23T18:55:22Z

commit 92f3cc91a2b73dc490ae6048309beb6da7382c96 parent 90d6018d786c42cf3a053508647144dceca2167e Author: Ryan Sepassi Date: Thu, 23 Apr 2026 11:55:22 -0700 move files around

post update

2026-04-23T18:52:00Z

commit 90d6018d786c42cf3a053508647144dceca2167e parent add01712f59b79acb21ef42924bdb260357ef6b8 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 11:52:00 -0700 post update

drop i386, expand register set, add BLTU

2026-04-23T18:42:13Z

commit add01712f59b79acb21ef42924bdb260357ef6b8 parent 932b2491b2cd4b0d21da8e12eb8796f6deb74da5 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 11:42:13 -0700 drop i386, expand register set, add BLTU

post update

2026-04-23T18:15:21Z

commit 932b2491b2cd4b0d21da8e12eb8796f6deb74da5 parent 67f4606e7e4c31112760ac9855384a4315e3a7df Author: Ryan Sepassi Date: Thu, 23 Apr 2026 11:15:21 -0700 post update

rename to m1pp

2026-04-23T18:06:35Z

commit 67f4606e7e4c31112760ac9855384a4315e3a7df parent f0abcca8fab5acb18dbfbfcea898ffc3bdc6d176 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 11:06:35 -0700 rename to m1pp

Use !@%$ vs %le32/le64 in m1pp

2026-04-23T18:03:01Z

commit f0abcca8fab5acb18dbfbfcea898ffc3bdc6d176 parent c40935b691eabb80b07e5bb9d65d5d86c402dbe1 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 11:03:01 -0700 Use !@%$ vs %le32/le64 in m1pp

Regularize M1M macro oracle

2026-04-23T15:58:37Z

commit c40935b691eabb80b07e5bb9d65d5d86c402dbe1 parent a0b1dde7214f0ea5574565c1c0fbcdd16fedef90 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 08:58:37 -0700 Regularize M1M macro oracle

m1m simplify spec

2026-04-23T15:07:10Z

commit a0b1dde7214f0ea5574565c1c0fbcdd16fedef90 parent 608c2bdc5926f697bace3c4d9881226acd06f792 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 08:07:10 -0700 m1m simplify spec

Implement initial P1 m1m tokenizer slice

2026-04-23T14:19:12Z

commit 608c2bdc5926f697bace3c4d9881226acd06f792 parent 6408a180a619872e0e96da5be7871cb7ad81ad54 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 07:19:12 -0700 Implement initial P1 m1m tokenizer slice

Describe m1m bootstrap path

2026-04-23T13:54:52Z

commit 6408a180a619872e0e96da5be7871cb7ad81ad54 parent 4ea652af1d9ea5e7dc9869386da837c95ff76f49 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 06:54:52 -0700 Describe m1m bootstrap path

Document m1macro P1 port plan

2026-04-23T13:54:48Z

commit 4ea652af1d9ea5e7dc9869386da837c95ff76f49 parent c24e386d1f9da71804e861ccfa961b6db0f6d924 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 06:54:48 -0700 Document m1macro P1 port plan

Add P1 m1m bootstrap stub

2026-04-23T13:54:42Z

commit c24e386d1f9da71804e861ccfa961b6db0f6d924 parent b67850fbf7de470673af34368fe0342900082d26 Author: Ryan Sepassi Date: Thu, 23 Apr 2026 06:54:42 -0700 Add P1 m1m bootstrap stub

p1.m1m

2026-04-23T04:19:56Z

commit b67850fbf7de470673af34368fe0342900082d26 parent c34eadeb5a6d1e48435c97d722fcc15a677a6a54 Author: Ryan Sepassi Date: Wed, 22 Apr 2026 21:19:56 -0700 p1.m1m

p1v2

2026-04-23T00:51:03Z

commit c34eadeb5a6d1e48435c97d722fcc15a677a6a54 parent 1b11e9dca9efcdb09fd2917b96b240e43c8ec91d Author: Ryan Sepassi Date: Wed, 22 Apr 2026 17:51:03 -0700 p1v2

Vendor AArch64 catm hex2 source

2026-04-22T18:17:23Z

commit 1b11e9dca9efcdb09fd2917b96b240e43c8ec91d parent a84e4ca4303eb3dcad5cf34ca83c2f119019ef1e Author: Ryan Sepassi Date: Wed, 22 Apr 2026 11:17:23 -0700 Vendor AArch64 catm hex2 source

build: ~30x faster + 75% smaller lisp binary

2026-04-22T18:17:03Z

commit a84e4ca4303eb3dcad5cf34ca83c2f119019ef1e parent b97e3d36e52f0da1c9260fb3f5c56dcc35e6ee4f Author: Ryan Sepassi Date: Wed, 22 Apr 2026 11:17:03 -0700 build: ~30x faster + 75% smaller lisp binary Three orthogonal wins stacked into one build pass: - Prune unused DEFINEs: awk filter keeps only the ~350 of 7,293 p1_.M1 DEFINEs that lisp.M1 actually references, cutting M0's input ~50% and its O(macros × tokens) Set_Expression scan ~20x. - tmpfs staging: M0/hex2-0 issue one read()/write() syscall per byte (custom unbuffered fgetc/fputc, not libc). Going through the virtiofs bind mount on Apple Silicon podman costs ~90s per 340KB stream; the same stream through the container's own overlayfs /tmp costs ~2s. All per-byte I/O now lives on /tmp; virtiofs only sees bulk cp traffic. - BSS via the ELF header: new src/elf/ELF-.hex2 declares p_memsz = :ELF_bss_end - :ELF_base (kernel-zero-filled) while p_filesz stays at :ELF_end. lisp.M1 splits accordingly: :ELF_end lands before the ZERO32-heavy BSS region, :ELF_bss_end sits at the very end. :mark_stack_next drops its &mark_stack init (gc_mark_all re-seeds it every pass). A post-link truncate reads p_filesz from byte offset 0x60 and drops the trailing zero bytes hex2 had to emit to advance IP. The stock stage0-posix ELF-*.hex2 files under build/upstream/ stay untouched because bootstrap.sh still uses them to link M0 itself (no BSS there). Measured (aarch64, clean build): ~124s -> ~5s; binary 159,293 -> 38,541 bytes. All 28 lisp tests pass on aarch64/amd64/riscv64.

lisp.M1: move 14 primitives to Scheme prelude for reviewability

2026-04-22T16:16:36Z

commit b97e3d36e52f0da1c9260fb3f5c56dcc35e6ee4f parent 1721f75a69bfd518d1f4c1c7327118b7353b1393 Author: Ryan Sepassi Date: Wed, 22 Apr 2026 09:16:36 -0700 lisp.M1: move 14 primitives to Scheme prelude for reviewability Promoted all fixed-arity helpers that can be derived from the remaining primitives: arithmetic: <=, >=, zero?, negative?, positive?, abs list: list?, length, reverse, assoc, member vector: vector->list, list->vector structural: equal? (+ removes equal_helper and eq_* sub-labels) Also adds Scheme-side nice-to-haves that were never P1 primitives but are commonly expected: not, caar/cadr/cdar/cddr/caddr, list-ref, for-each. Variadic primitives (list, append, min, max) stay in P1 because env_extend doesn't yet support (lambda args body) or dotted-tail params; revisit after that lands. Remaining primitive code ids are renumbered contiguously 0..45. lisp.M1 shrinks ~560 lines; src/prelude.scm grows ~150. Tests 28/28 on aarch64 and amd64 (riscv64 in progress at commit time).

lisp update TODO

2026-04-22T16:16:29Z

commit 1721f75a69bfd518d1f4c1c7327118b7353b1393 parent 07de70b206e96bc65e95c530a6b487101f218640 Author: Ryan Sepassi Date: Wed, 22 Apr 2026 09:16:29 -0700 lisp update TODO

P1.md: clarify leaf-function semantics (CALLable in, not out)

2026-04-22T14:19:54Z

commit 07de70b206e96bc65e95c530a6b487101f218640 parent e05d709edfb9f6f42bff02097a80ebb58282b4e6 Author: Ryan Sepassi Date: Wed, 22 Apr 2026 07:19:54 -0700 P1.md: clarify leaf-function semantics (CALLable in, not out) The prior rewording left ambiguous whether a leaf could be CALLed at all. Split the leaf contract into the three operations it may (or may not) perform internally: RET is fine, tail-branch via li_br+B is fine and lets the target's RET skip the leaf in the return chain, only inner CALL is forbidden. Being CALLed is always fine — the restriction is on what the leaf does, not on who invokes it.