chore(memory): V1-2.2 reserved ids evidence, mutations, contract, TODO

This commit is contained in:
bchanot committed 2026-10-30 17:31:01 +01:00
1 parent 6ed4d21eb3
commit 81f7639c75
5 files changed
+155

No files matched your search

+6
View File
@@ -587,6 +587,12 @@ Order (BDR-071): P0 docs + decisions → V1-1 clients and animals → V1-2 sheet
- [x] J4 FLOOR false positive: reword `.claude/audits/2026-10-29-v1-2-2-arbitrations-security.md:225`, guard kept
- [x] J5 memory: BDR-087 checked, LRN-100 superseded after J2; per-document prefix requirement kept
- [ ] J6 gates: full cargo, svg suites, mutations, wasm + ts-rs, size < 600 000, check:all (DEV-isolated steps only), fresh independent review; commit + push; STOP (A6 open: Illustrator). Done 7d9ab8c/f976f33: check:all green, verifier ECARTS(2) (A6 Illustrator; §12.4 drift fixed after), security PASS (new pre-existing MEDIUM: PDF halo id `osteo-halo` reusable by a drawing, owner decision); left open until Illustrator
- [ ] K reserved ids and A6 Illustrator (brief « V1-2.2 : correction de sécurité PDF et poursuite A6 », 2026-10-30, on 2f1f076): fixtures and isolated stacks only, DEV untouched; no merge; STOP at checkpoint, no V1-2.3
- [x] K1 audit: `osteo-halo` written by `compose_highlight` only (PDF view keeps the drawing's ids); `has_halo_filter` skips the injection when the drawing defines `filter#osteo-halo` (hijack), a non-filter `#osteo-halo` before the first `<defs>` wins resolution; historical drawings carry `osteo-overlays` (DEV equine object included) → exact reserved set, never an `osteo-*` namespace; canvas scoped `fv.<view>.`, layer writes no id, filter `result` names local to the filter
- [x] K2 reader: `SvgRejection::ReservedId` (`reserved_id`), `refs::check_reserved_ids` (ASCII case folded, the id grammar is ASCII) in `read_svg` and in `clean_svg` after the active classification; reaches verdict, preview, loaded sheet, upload tool, PDF snapshot gate
- [x] K3 composer fail-closed: `compose_highlight` refuses a tree holding a reserved id (`ComposeError::ReservedId`), always writes its own halo (`has_halo_filter` removed); ops mapping; TS-era "never duplicated" test becomes the refusal
- [x] K4 tests: valid, filter + other elements (root included), references, adversarial spellings (case, entities, whitespace, `xml:id`, prefixed names, look-alikes), composition, PDF session (refusal + internal halo kept), equine + sha256, declared divergence for `osteo-halo-id.svg`, pixel proof in Chromium (before / after)
- [ ] K5 docs (RUST.md §12, contract §9.x hard rejections, ANNOTATION-RENDERING), Illustrator `presentation` / `internal-css` owner steps; gates (cargo, svg, mutations, wasm size, check:all DEV-isolated), fresh verifier + security on final HEAD; memory; commit + push; STOP (A6 open until Illustrator). State: check:all green, verifier ECARTS(1) LOW fixed (leaked `id` expansion in a comment), security PASS + pre-existing MEDIUM root `visibility` hides every marker (owner decision), LOW root stroke, LOW halo defs placement
- [ ] V1-2.3 drawings and worker verification (M6) · V1-2.4 wireframes (editor, review, admin entry) · V1-2.5 catalogue, admin space, rights, species, MFA · V1-2.6 editor level 1 · V1-2.7 preview, review, publication UI · V1-2.8 equine working copy and review preparation · V1-2.9 FR/EN acceptance and closure (owner checkpoints V1-2.3, V1-2.4, V1-2.6, V1-2.7, V1-2.8, V1-2.9 + each DEV operation of contract §16.11)
## Release readiness track (owner 2026-10-25, distinct from the functional roadmap; nothing here authorises a release or exposure)
@@ -0,0 +1,48 @@
# Contract: V1-2.2 reserved ids (PDF halo collision)
Owner brief « V1-2.2 : correction de sécurité PDF et poursuite A6 », 2026-10-30, on `2f1f076`
(branch `feature/v1-2-anatomical-sheets`). Diff under review: the working tree against `2f1f076`.
## Problem
`crates/core/src/anatomy/compose.rs::compose_highlight` (PDF view) keeps the drawing's ids and
writes its halo filter `osteo-halo` next to them. Before the fix it skipped writing its filter when
the drawing already held a `filter#osteo-halo`, and a non-filter element holding the id before the
first `<defs>` won `url(#osteo-halo)` resolution. A sheet drawing could thus replace, divert or hide
the PDF markers.
## Criteria (each must hold on the diff)
- C1 The shared SVG reader (`anatomy/svg/mod.rs::read_svg`) refuses any kept element (root included)
whose `id` is a reserved id, with a closed, stable reason `reserved_id`
(`SvgRejection::ReservedId`).
- C2 The decision is exact-name (reserved set, not a prefix): historical drawings holding
`osteo-overlays` (the equine draft drawing, sha256 `0ea5202f…`, which is the DEV object) still read
with no removal; no historical asset changes byte-wise.
- C3 No bypass: letter case (ASCII), XML character references, whitespace, look-alike or invisible
characters, `xml:id`, prefixed attributes or prefixed SVG elements, CSS escapes and percent-encoded
`url(` targets either refuse the drawing or never leave a reserved id in the tree a composer
receives.
- C4 `clean_svg` (upload cleaning) refuses it (`reserved_id`) with no stored form, after active
content is classified; the verdict, the preview, the loaded sheet, the upload tool and the PDF
snapshot gate (`pdf/write/input.rs`, `snapshot_invalid` / `asset reserved_id`) answer it through
the same reader.
- C5 Composers are fail-closed: `compose_highlight` and `compose_canvas` refuse a tree holding a
reserved id (`ComposeError::ReservedId`), and the highlight always writes the engine's own filter;
no path remains where a drawing-supplied element named `osteo-halo` reaches the PDF view.
- C6 No existing control weakened: every other reader rule, cap, reference check, scoping and the
PDF/archive readers unchanged; frozen oracles, R4 digests, step6 goldens, N-1 copies untouched
(declared divergences applied in tests only).
- C7 Other internal collisions audited (ids the composition writes, canvas scope `fv.`, layer
attributes, filter `result` names) and either covered or argued absent.
- C8 Tests cover: a valid drawing; `filter id="osteo-halo"`; the id on other elements; references
to it; adversarial spellings; the PDF keeping its own halo (behavioural, through the export
session); equine non-regression and historical fingerprints; the shipped wasm module.
- C9 Docs updated (RUST.md §12.9, V1-2 contract §9.2 table and closed codes, ANNOTATION-RENDERING);
the R5 ledger target renamed with the test.
- C10 A6: the owner README gives precise Illustrator steps for `…-presentation.svg` and
`…-internal-css.svg` (names and variants unchanged), no fabricated Illustrator file, A6 not claimed.
## Out of scope
No SQL, no Supabase DEV, no worker, no merge, no V1-2.3.