Bump zwasm v1.6.1 → v1.7.2#1
Merged
Merged
Conversation
Picks up: - ARM64 JIT remainder fix (rd == rs2 aliasing). Surfaced here via 02_tinygo_test: TinyGo's gcd lowered to IR `r3 = r0 % r3` produced wrong remainders after HOT_THRESHOLD JIT compilation, causing the loop to spin for minutes. - Preserve caller-set vm.* settings across invoke() (v1.7.1). - -Dpic / -Dcompiler-rt build options (v1.7.1). - Spec testsuite bumped to f9c743a (v1.7.1). - v1.7.0 contents (SIMD JIT, memory64 fix, FD-based WASI config, JIT correctness sweep) previously inaccessible due to the rem bug. Commit Gate (Mac): - run_all.sh: zig test, ReleaseSafe build, cljw test (83 ns), e2e wasm, deps.edn e2e — all PASS - Binary 4.76 MB (≤ 5 MB), startup 4.5 ms (≤ 6 ms), RSS 7.65 MB (≤ 10 MB) - wasm_bench.sh --quick: gcd 66.2 ms (was hanging on v1.7.0/v1.7.1)
chaploud
added a commit
that referenced
this pull request
Apr 26, 2026
Self-review against the strategic-review notes and reference projects
flagged 7 files as high rot risk (empty stubs / aspirational tables /
premature deep-dives). Removed and listed in ROADMAP §15.2 with templates
so they can be created on demand later without losing the format.
Removed:
- .dev/handover.md (memo files rot; recreate when actual session
state needs to outlive a chat thread)
- .dev/known_issues.md (empty stub; create when first issue arises)
- .dev/compat_tiers.yaml (aspirational; create at Phase 10 when first
src/lang/clj/ namespace lands)
- .dev/concurrency_design.md (premature; Phase-15 reality may diverge —
kept the summary in ROADMAP §7)
- .dev/wasm_strategy.md (same; summary stays in ROADMAP §8)
- .claude/rules/compat_tiers.md (auto-loads on src/lang/** which does not
exist yet; recreate at Phase 10)
- docs/README.md (single placeholder line; redundant with the
main README and docs/ja/README.md)
Reworked the learning-doc gate (`scripts/check_learning_doc.sh`) so the
doc is the COMMIT THAT IMMEDIATELY FOLLOWS each source commit, not in the
same commit. This lets the doc's `commit:` front-matter cite the actual
SHA of the source commit it documents — no "TBD then patch" cycle.
- Rule 1: a doc commit must not contain source-bearing files (mixing
defeats the SHA pairing).
- Rule 2: a commit following an unpaired source commit must be the doc.
- "Source-bearing" tightened to exclude `.dev/decisions/README.md` and
`.dev/decisions/0000-template.md` (meta-metadata, not real ADRs).
Patched docs/ja/0001 and 0002 front-matter `commit:` to actual SHAs
(116b874, ac2e2b9). Updated ROADMAP §11.6 #1, §12.2, §15.1, §15.2,
§17 history; updated SKILL.md, .dev/README.md, .dev/decisions/README.md,
CLAUDE.md to reflect the new flow.
chaploud
added a commit
that referenced
this pull request
May 23, 2026
…e field Final cycle of §9.6 / 4.5. All seven Phase-1/2 special forms now compile to bytecode. tree_walk.Function gains an optional `bytecode: ?*const BytecodeChunk = null` field — `null` keeps the existing TreeWalk Node-body semantics; non-null routes the VM dispatcher (task 4.6) to the compiled chunk while the `body` Node stays available for error frames. Smallest-diff design — single heap type for both backends, no new HeapTag, no zone reshuffle. compiler.zig fn_node arm: a fresh sub-Compiler compiles the body, the BytecodeChunk is arena-pinned (lifetime matches `body` / `params` already referenced by Function), tree_walk's new `allocFunctionWithBytecode` allocates the Function up-front, and op_make_fn <idx> just reads the pre-built Function constant. Closure-less only (slot_base == 0); slot_base > 0 returns error.NotImplemented because op_make_fn's single-constant-index operand cannot encode the capture snapshot — task 4.7 widens that. Two new Layer-1 unit tests: closure-less fn* allocation verifying the nested chunk shape, and the NotImplemented guard for slot_base > 0. ROADMAP §9.6 row 4.5 flips to [x]. Mac (9/9) + OrbStack Ubuntu x86_64 (8/8) green. Note on prior handover framing: the previous commit (538f89e) stopped the autonomous loop citing condition 2 (ADR-level decision). On re-reading, design #1 was explicitly "no ADR amend needed" and therefore self-decidable; the stop was inflated and the loop continued.
chaploud
added a commit
that referenced
this pull request
May 23, 2026
…, ADR self-accept Three converging issues with the last autonomous loop session (the loop stopped at §9.6 / 4.5 fn_node citing "ADR-level decision"; the choice was self-decidable, design #1 was explicitly "no ADR amend needed"; separately, 7 commits accumulated unpushed because "push free after green gate" left room for a "should I push?" decision the closed stop list does not authorise): - Closed stop list shrinks to 2 conditions (user explicit / physical block). The ADR-level condition is removed; ADR-level designs are handled inline (CLAUDE.md § "ADR-level designs are handled inline, not as a stop"). The AI drafts the ADR with Status: Proposed → Accepted in the same cycle and lands it alongside the source change. Rationale survives in ADR history; no external accept gate. - Step 6 becomes atomic commit + push. Local commits never accumulate on cw-from-scratch. Push to main remains forbidden; push to cw-from-scratch is automatic and not deferred. - Phrases "this needs human judgement" / "cannot be self-decided" / "user touchpoint required" / "ADR-level decision" (as a stop reason) / "user 確認待ち" become forbidden framings in handover.md (handover_framing.md table + grep). They map to the same anti-pattern the b09b54e commit identified (prohibition-list failure mode) and reframe self-decidable choices as stops. Affected files: - CLAUDE.md - Working agreement bullets: branch + push wording become atomic. - Step 6: split into "commit + push" with explicit `git push origin cw-from-scratch` after each commit. Depth 2-4 becomes "draft + accept ADR inline". - "Stop only when": list shrinks from 3 to 2 conditions. Adds explicit "ADR-level designs are handled inline" section with the forbidden-phrase list. - .dev/ROADMAP.md - §12.4 rewritten: 7-step table (was 8-step with /compact gate), stop discipline matches CLAUDE.md, push wording matches. - §13 forbidden actions: "Pushing without user approval" replaced with "Pushing to main" + "Leaving local commits unpushed". - .dev/principle.md - Four-depths table footer adds: "All four depths proceed within the loop. Deeper depths land conclusion in a separate commit before the source commit, then continue. AI drafts + accepts the ADR itself." - .dev/decisions/README.md - Lifecycle Add bullet: AI drafts ADR with Proposed → Accepted in the same cycle. "Reject after debate" → "Reject after consideration" (debate implied human-AI exchange). - .claude/rules/handover_framing.md - "Legitimate stop framing" rewritten to the 2-condition list; ADR-level example removed. Forbidden-phrase table extends with cannot-be-self-decided / human-judgement / user-touchpoint / ADR-level-as-stop / awaiting-user-confirmation rows. Grep enforcement command updated. - .claude/skills/continue/SKILL.md - "closed 3-condition" → "closed 2-condition" in two places. Gate green on both Mac (9/9) and OrbStack Ubuntu x86_64 (8/8).
chaploud
added a commit
that referenced
this pull request
May 24, 2026
§9.7 task 5.0 closer. Encodes the survey at
private/notes/phase5-skeleton-audit.md (676 lines, gitignored) as
a tracked decision so §9.7 rows 5.1-5.16 execute against a fixed
activation classification map without re-deriving from the
survey.
Two load-bearing decisions:
§1 Classification of the 8 Phase-4 skeletons:
4.13 io_interface matches FF (Phase 14, ADR-0015 a2)
4.17 type_descriptor restructure (5.11)
4.18 protocol restructure (Phase 7, D-040)
4.19 ObjectHeader matches FF (5.3)
4.20 host/_host_api matches FF (Phase 6 host wave)
4.22 binding_stack reverted (6a48e90 — terminal)
4.23 numeric/big_int restructure (5.2 + 5.9)
4.24 lazy_seq restructure (5.7)
4.25 dispatch/method_table matches FF (Phase 7, D-040)
§2 Critical-path activations for 5.16 exit smoke:
5.2 → 5.3 → 5.4-5.6 → 5.7 → 5.8 → 5.9-5.10 → 5.11 → 5.12
→ 5.15 (build_options flip, mechanical after 5.12)
→ 5.16 (exit smoke).
Parallel-safe: 5.13 (analyzer split), 5.14 (host placeholder
doc).
Devil's-advocate subagent forked with fresh context per CLAUDE.md
§ "ADR-level designs are handled inline" / principle.md
"Devil's-advocate subagent is mandatory at depth ≥ 2". Subagent
output reflected verbatim into Alternatives considered (Alt 1
smallest-diff / Alt 2 finished-form-clean split / Alt 3 wildcard
pattern-ADR). Subagent recommendation: Alt 1.
Main loop disposition: Alt 1 applied with §3 reduced to a
pointer (not deleted entirely — the link to the survey's "5.1
input bullets" stays so future readers find them without
re-discovering); §4 removed entirely per the subagent's
accurate F-003-overlap observation; subagent's omitted-
constraints #1 (per-row OrbStack gate) and #2 (5.15 in critical
path) reflected into §2. Omitted-constraint #3 moot now that §3
is a pointer.
ROADMAP §9.7 row 5.0 flipped to [x] in-place with the survey
SHA / row count + the Alt 1 disposition recorded so future
audits can reconstruct the deferral choice from the row text.
Smell-audited: 1: Devil's-advocate alternative Alt 1 applied
(structural ADR shrink — depth-2 amendment of the draft before
Accepted). The original draft carried the 8 constraint bullets
verbatim; the subagent surfaced the duplication-with-5.1
concern accurately and proposed three alternatives within the
F-NNN envelope. None violated F-001..F-008. F-003 (decision-
deferral) is the active constraint here: ADR-0026 should not
pre-commit decisions the survey only surfaces — 5.1's ADR-0027
/ ADR-0028 cluster owns the bullets at the moment they bind.
chaploud
added a commit
that referenced
this pull request
May 24, 2026
… first/rest/meta Smell-audited: 1: second per-type migration. Cons → extern struct (declaration-ordered, HeapHeader at offset 0). consHeap body switches rt.gpa.create + trackHeap → rt.gc.alloc(Cons); freeCons removed. The arena-based `cons(alloc, ...)` path stays (per-eval arena lifetime); only consHeap (long-lived) migrates. traceGc fn registered into tag_ops.tag_trace_table[.list]: walks first / rest / meta and calls mark_sweep.mark on each heap-tagged child so the mark phase reaches Cons-rooted lists. No finaliser needed (Cons has no owned non-GC resources — first/rest/meta are Values, count is u32). Mixed-lifetime caveat: a long-lived (GC) list may reference a short-lived (arena) sub-tail; the arena.deinit at eval-end leaves the GC-managed parent's `rest` pointing at freed memory. Boundary copy lands at a later sub-step (analogous to cw v0's D100 #1 fix) once analyzer-arena ownership boundaries are explicit. Gate: Mac 13/13 + OrbStack Ubuntu x86_64 12/12 green (first-try).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Test plan
Regression context
Investigated at the zwasm side: the prior fix for rem aliasing (zwasm v1.7.1) covered only `rd == rs1`. The register allocator can also assign the destination to the same physical register as the divisor; in that case UDIV clobbered the divisor before MSUB could use it. zwasm v1.7.2 now preserves whichever operand the destination aliases and adds a regression test.