From 1a30928fc52c4c1a9e351a5e55b406889fb73e46 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Ralph=20K=C3=BCpper?= Date: Thu, 6 Aug 2026 06:20:08 +0200 Subject: [PATCH 1/2] =?UTF-8?q?docs(engine-plan):=202026-08-06=20ticket=20?= =?UTF-8?q?sync=20=E2=80=94=20#7472/#7396=20closed,=20#7475=20sole=20basel?= =?UTF-8?q?ine=20blocker,=20repsel=20stack=20position?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Step 7 now records: the stringify crash fixed (#7472 -> #7473, verified against node), #7396 closed as fixed-by-#7421, the public artifact down to one blocker (#7475), and the #7481 residual with the probe-coverage lesson. Adds the repsel stack position: phases 1-4a merged, 4b in flight, element-shape route decided in #7480 (shared-prerequisite sequencing: invariant bit -> versioned-loop consumer -> element Ptr), all gated on the full gap suite per #6377. --- docs/engine-plan.md | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) diff --git a/docs/engine-plan.md b/docs/engine-plan.md index 508859c298..4aa1e1c553 100644 --- a/docs/engine-plan.md +++ b/docs/engine-plan.md @@ -762,6 +762,15 @@ landed (#7314) and became *reachable* (#7339) and *selectable* (#7340). The spin the tape's memcpy path. Any tape change must keep it (#7476 pins the numbers). + **Ticket status (2026-08-06 sync):** the `JSON.stringify` crash that + blocked the artifact is FIXED (#7472 → #7473, default path verified + byte-equal to node's checksum); #7396 is CLOSED as fixed-by-#7421. The + public artifact now has exactly ONE blocker: **#7475** (two app-pattern + kernels fail only under the auto-optimize runtime — isolated to the + feature-stripped archive, scale-dependent, pre-existing). A residual + non-default-cap fault from #7472's investigation is tracked as #7481 — + a large-Eden ratchet-probe arm would have caught both it and #7472. + **The JSON workstream is sequenced, with a correctness gate in front:** #7477 first — the DirectParser parses floats differently from the tape materializer *and node agrees with the tape*, which is silently wrong for @@ -782,6 +791,19 @@ landed (#7314) and became *reachable* (#7339) and *selectable* (#7340). The spin host load hit 55 and its disassembly check matched an absent symbol; the sweep's own gate (≤25% CPU for 60 s, AC power) is the standard. + **The repsel stack (the "unbox-by-default" campaign this step feeds):** + phases 1 / 2 / 3a (#6909) / 3b (#6911) / 4a (#6915, plus #7421/#7425 + follow-ups) are **merged**; #6904's 26× histogram is closed. Phase 4b + (field-store note/addref elision, prompt-file scoped to two items) is + **in flight**. The next gap after it is element-shape proofs through + array reads — `keep[j].v` measured at **6.2× vs node** on the pure + shape — and the route is decided in **#7480**: the two candidate routes + share one prerequisite (a per-array homogeneous-element-shape + invariant, construction-maintained and self-healing like 4a's dense + bit), consumed first by the #5093 versioned-loop clone and then by + element `Ptr`. Every phase gates on the FULL gap suite against a + same-session main baseline (#6377's lesson), never its own microbench. + 8. **CI hygiene is a correctness input, not housekeeping.** Three of four RS4GC arms had never executed (#7393). Before citing any matrix as platform coverage, confirm its arms actually reach a runner. From bc345a36b448ca7f438737e74c3f885bccaa662a Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Ralph=20K=C3=BCpper?= Date: Thu, 6 Aug 2026 06:20:09 +0200 Subject: [PATCH 2/2] docs: changelog fragment for 7482 --- changelog.d/7482-plan-ticket-sync.md | 1 + 1 file changed, 1 insertion(+) create mode 100644 changelog.d/7482-plan-ticket-sync.md diff --git a/changelog.d/7482-plan-ticket-sync.md b/changelog.d/7482-plan-ticket-sync.md new file mode 100644 index 0000000000..56cdb3f08e --- /dev/null +++ b/changelog.d/7482-plan-ticket-sync.md @@ -0,0 +1 @@ +- **`docs/engine-plan.md` synced to ticket reality (2026-08-06):** #7472 fixed by #7473 and closed, #7396 closed as fixed-by-#7421, the public benchmark artifact down to a single blocker (#7475), residual #7481 recorded with the large-Eden probe-coverage lesson, and the repsel stack position added — phases 1–4a merged, 4b in flight, element-shape route decided in #7480. (#7482)