The work on StarForth-v4.0.0 since #197 (52 commits):
v4 init.4th and the ACL. The ACL policy is loaded after POST, then the search order (steps 1 and 2). init.4th pins the words that create or end nodes. The ACL countdown runs at call, as in v3, and the call hook may redirect (plan ruled first).
ADR 2026-11-19: F18 binary artifacts, ABI/API and live runtime updates. Recorded as given, with rulings 1–4. docs/v4.0.0/ABI.md (ADR step 1) and docs/v4.0.0/ARTIFACT.md (step 2) were drafted, then ruled choice by choice.
ADR step 3: artifacts in the build (docs/v4.0.0/plans/2026-11-19-f18-artifact-build.md, ruled §6 and §10):
task 1: text.c's address record (§8);
task 2: effect.c, stack effects derived from the code (§9);
task 3: mkf18a, the host tool that builds a train's .f18a, and make -C v4 artifacts (§11). This includes text.c's word attributes and import lines, and the artifact check in artifact.c.
Each was test-first, seen to fail; the evidence is in logs/20261119-*.
Acceptance
logs/20261119-230627/PROTOCOL.md for the head commit. The protocol ran on bd8997ff, with every stage passing. The head, 99851075, differs from it only in the evidence; scripts/check_protocol.sh accepts it.
amd64, aarch64 and riscv64 reach POST 990/990, with dict_hash=0x0ffb1d31c4aa7857 on all three (405 words);
stick-high reaches POST 990/990;
test-32: 42 programs, 0 failures;
the proofs, the images and the documents pass.
Logs from other runs made by hand are committed in logs/:
the earlier protocol on dc3803ab, whose boots and test-32 passed. It was not completed, because an evidence commit moved HEAD.
its two aarch64 runs that hit the 570 s foreground timeout, kept as stopped runs, not failures;
the make -C v4 test, test-32 and sanitize gates for each task.
Hardware this was tried on: none.
The record
docs/fabric/FABRIC-3.md has the ruling and what was built: not in this PR. The rulings and what was built are in docs/v4.0.0/ADR, ABI.md, ARTIFACT.md and the step 3 plan.
CHANGELOG.md: not a milestone.
Nothing in this PR is a fix nobody asked for. Two findings are recorded, not fixed:
the nucleus's UM*, D+ and DNEGATE have no fixed effect;
mkf18a's store check does not see stores below here.
## What and why
The work on `StarForth-v4.0.0` since #197 (52 commits):
- **v4 init.4th and the ACL.** The ACL policy is loaded after POST, then the search order (steps 1 and 2). init.4th pins the words that create or end nodes. The ACL countdown runs at call, as in v3, and the call hook may redirect (plan ruled first).
- **ADR 2026-11-19: F18 binary artifacts, ABI/API and live runtime updates.** Recorded as given, with rulings 1–4. `docs/v4.0.0/ABI.md` (ADR step 1) and `docs/v4.0.0/ARTIFACT.md` (step 2) were drafted, then ruled choice by choice.
- **ADR step 3: artifacts in the build** (`docs/v4.0.0/plans/2026-11-19-f18-artifact-build.md`, ruled §6 and §10):
- task 1: `text.c`'s address record (§8);
- task 2: `effect.c`, stack effects derived from the code (§9);
- task 3: `mkf18a`, the host tool that builds a train's `.f18a`, and `make -C v4 artifacts` (§11). This includes `text.c`'s word attributes and `import` lines, and the artifact check in `artifact.c`.
Each was test-first, seen to fail; the evidence is in `logs/20261119-*`.
## Acceptance
- [x] `logs/20261119-230627/PROTOCOL.md` for the head commit. The protocol ran on bd8997ff, with every stage passing. The head, 99851075, differs from it only in the evidence; `scripts/check_protocol.sh` accepts it.
- amd64, aarch64 and riscv64 reach POST 990/990, with `dict_hash=0x0ffb1d31c4aa7857` on all three (405 words);
- stick-high reaches POST 990/990;
- test-32: 42 programs, 0 failures;
- the proofs, the images and the documents pass.
- [x] Logs from other runs made by hand are committed in `logs/`:
- the earlier protocol on dc3803ab, whose boots and test-32 passed. It was not completed, because an evidence commit moved HEAD.
- its two aarch64 runs that hit the 570 s foreground timeout, kept as stopped runs, not failures;
- the `make -C v4 test`, `test-32` and `sanitize` gates for each task.
- [ ] Hardware this was tried on: none.
## The record
- [ ] `docs/fabric/FABRIC-3.md` has the ruling and what was built: not in this PR. The rulings and what was built are in `docs/v4.0.0/ADR`, `ABI.md`, `ARTIFACT.md` and the step 3 plan.
- [ ] `CHANGELOG.md`: not a milestone.
- [x] Nothing in this PR is a fix nobody asked for. Two findings are recorded, not fixed:
- the nucleus's `UM*`, `D+` and `DNEGATE` have no fixed effect;
- `mkf18a`'s store check does not see stores below `here`.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled "ACL at boot", loaded "From init.4th", the include
"Compose[d] in FORTH".
- capsules/v4/init.4th, at block 3980: S" v4:ACL.4th" CAPSULE-INCLUDE,
then STARSHIPOS.
- capsules/v4/ACL.4th is the policy, git mv'ed from v4/capsule/ACL.fth
and reflowed into chained 64x16 blocks, its text unchanged.
- CAPSULE-INCLUDE in centre.4th places a capsule and LOADs its lowest
block. It is composed from CAPSULE-OPEN and CAPSULE-LINE, with no new
C word.
- The mesh loads v4:init.4th after POST and COLD, its one fixed step
there.
Test first in test_mesh_boot: the parity line, DOE without naming
STARSHIPOS, and ACL-BOOT pinned. The before-run failed.
After: 113 checks pass and POST is 990/990. All 52 tests, test-32 and
ASan/UBSan pass. The dictionary has 405 words. One amd64 session passes
18/18 with "ACL: active".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled: "Also pin the kernel words". After the policy loads,
init.4th pins 11 words: BIRTH, BIRTH-AS, UNIT, KILL, UNIT-KILL, PAUSE,
RESUME, NODE-BORN, NODE-KILL, NODE-SLEEP and NODE-WAKE. Their fields
cannot be changed, so no policy can deny them. v4 has no CAPSULE-BIRTH.
The wiring words, NODE-WIRE and NODE-UNWIRE, are not pinned; they
neither create nor end a node.
Test first in test_mesh_boot: each of the 11 is pinned. The before-run
had all 11 unpinned. After: 114 checks pass, POST 990/990.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled "As v3: add at call", "Engine: call redirects" and
"Build it".
- The engine's call hook returns where P goes.
- The kernel checks every call from code defined at run time. A TTL is
counted down at the call. At a TTL of 0, or a denied word, the call is
sent to (ACL-CALL), with the target on the data stack, which rechecks
or refuses it by the word's own name, then goes on to it.
- A word being compiled is no longer checked, as in v3.
Two things were added while building, both from v3:
- (ACL-BUSY) turns the check off while the policy rechecks. Without it
the recheck recursed.
- (ACL-CALL) does its own refusal and leaves WORD's buffer alone. A first
version broke POST's CONVERT.digits.
Test first: test_mesh_boot's five new checks failed before; now 120 pass
and POST is 990/990. test_exec tests the redirect. test_host_quit's two
v4-only expectations now follow v3: a denied word compiles.
All 52 tests, test-32 and ASan/UBSan pass. An amd64 session passes 18/18.
Five alternating host boots: +1.4% median, inside the noise.
DECOMPOSITION.md 5.20 is updated, and the FPGA's call opcode is recorded
as owing the same redirect.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every cell laid down that holds an address of the assembly is recorded,
with its branch slot or 7 for the whole cell, as ARTIFACT.md's
relocations need. Test first: test_text 13 checks failed before.
test_host_rom_nucleus runs with the record on: 1825 branches, 664 whole
cells, the ROM still equal word for word. v4 test, test-32 and
ASan+UBSan pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Walks a word's code, opcode by opcode with the engine's own counts, and
gives what it takes and leaves on both stacks, or why it has none.
Called and in-line words; recursion, computed jumps, paths that meet at
different depths and loops that leave a count have none. Test first:
23 checks failed against a stub. SWAP', UM/MOD, C@, C! and ROUNDTRIP
agree with their stack comments. On the nucleus: 214 of 315 headers
fixed. v4 test, test-32 and ASan+UBSan pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
As ruled (plans §10.2, §10.3, §10.5): ': NAME inline pinned mode N
ports request N' records the word's attributes and lays nothing down;
'import TRAIN NAME' declares another train's word, its calls, jumps and
&NAME recorded as that import. test_text: 11 checks failed before; 274
pass at both widths. The check that ': T immediate ;' is refused is
replaced, as ruled. Full gates run at the end of task 3.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
The artifact's layout and its check (artifact.h, artifact.c), the tool
(tools/f18a.c, mkf18a.c) and make -C v4 artifacts. test_mkf18a, 83
checks at both widths, failed against a stub first. Gates: test,
test-32 and sanitize pass, POST 990/990. Plan section 11.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
accept-amd64, accept-aarch64, accept-riscv64 and accept-stick-high
pass, POST 990/990, dict_hash 0x0ffb1d31c4aa7857 on all three (405
words); test-32, 42 programs, 0 failures. Two aarch64 runs stopped at
the 570 s foreground cap, not failed (212934 after the heartbeat,
213939 at node 11's SEND), are kept; the third ran in the background,
alone (rules, 2026-11-18), and passed. Proofs, images and docs to come.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every stage passes: the four boots at POST 990/990, dict_hash
0x0ffb1d31c4aa7857 on all three (405 words), test-32 (42 programs, 0
failures), the proofs, the images and the documents.
logs/20261119-230627/PROTOCOL.md. test-32 ran twice: a container
restart lost the first run, with no result.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
What and why
The work on
StarForth-v4.0.0since #197 (52 commits):docs/v4.0.0/ABI.md(ADR step 1) anddocs/v4.0.0/ARTIFACT.md(step 2) were drafted, then ruled choice by choice.docs/v4.0.0/plans/2026-11-19-f18-artifact-build.md, ruled §6 and §10):text.c's address record (§8);effect.c, stack effects derived from the code (§9);mkf18a, the host tool that builds a train's.f18a, andmake -C v4 artifacts(§11). This includestext.c's word attributes andimportlines, and the artifact check inartifact.c.Each was test-first, seen to fail; the evidence is in
logs/20261119-*.Acceptance
logs/20261119-230627/PROTOCOL.mdfor the head commit. The protocol ran onbd8997ff, with every stage passing. The head,99851075, differs from it only in the evidence;scripts/check_protocol.shaccepts it.dict_hash=0x0ffb1d31c4aa7857on all three (405 words);logs/:dc3803ab, whose boots and test-32 passed. It was not completed, because an evidence commit moved HEAD.make -C v4 test,test-32andsanitizegates for each task.The record
docs/fabric/FABRIC-3.mdhas the ruling and what was built: not in this PR. The rulings and what was built are indocs/v4.0.0/ADR,ABI.md,ARTIFACT.mdand the step 3 plan.CHANGELOG.md: not a milestone.UM*,D+andDNEGATEhave no fixed effect;mkf18a's store check does not see stores belowhere.🤖 Generated with Claude Code
https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
dc3803abso far -- four boots and test-32 passbd8997ff(mkf18a, ADR step 3 task 3): PASS