U2 Task 3 (A6, A14, R-2). A fixed-height strip over the drawing and the
right column replaces the tool line, the sr-only selection tag of the
edit body and TracingBar: lead (name + status, type + detail), notice
(refusal, first geometry issue + n, unmeasured view with its link, no
category, no child view), commands (Fermer le trace, Retirer le dernier
sommet, Annuler le trace). A fixed-height line under the drawing holds
the add buttons and the gesture message: a kept drop says « Point
deplace » with « Defaire » (same store undo as the header), a refused
drop says « Deplacement refuse : ... » as a red alert. 6 s, held by
mouse hover and focus inside, closed by any commit, undo or redo.
trace_pop and Backspace remove the last vertex. The busy « Aller a »
target is the strip's tracing commands. scrollbar-gutter stable, scoped
to the editor. e2e helper expectNoShift (boxes, scroll height, gutter,
client width, layout-shift sum). DraftHeader untouched: no shift seen.
Rewritten assertions (old -> new):
- tab unit « Sélection : point » on annotation-selection-tag -> strip
detail « Point »
- tab unit tag sr-only, not live -> strip lead outside any live region
- tab unit « 1 sommet posé. » -> « 1 sommet posé » (board 04 wording)
- tab unit D6b tag present / « zone » -> strip detail « Point » /
« Zone · 3 sommets »
- keep-alive, sections, escape units: tracing-bar -> tracing-commands
- e2e selectionTag helper -> strip title with data-selection; the
twin's tag is twinSelectionTag (deselect 375 px describe)
- e2e deselect « tag position absolute » (edit) -> strip height 64 px,
drawing top unchanged; twin Escape test keeps tag 0, properties 0,
overview visible
- e2e deselect and journey tracing-bar / annotation-selection-tag ->
tracing-commands / selectionTag
- e2e historyButton scoped to history-buttons (the message's
« Défaire » has the same accessible name)
useFraming takes an optional panning gate read on each one-pointer move;
the editor passes the press machine's drag-with-pan phase, so a zoomed
pen within its 6 px slop or a mouse wiggle no longer pans and clicks.
Catalogue: no options, behaviour unchanged.
Test added: « a press is either a pan or a click, never both (R-8) »
(pen 5 px clicks without pan, pen 8 px pans without click, mouse wiggle
clicks without pan). No assertion removed.
Split onClick (place / select click), onPointerDown (targetAt) and
useBlockDrag (dragActions) to respect the 5-locals limit. No behaviour
change; no test changed.
press.ts (pure: slop per pointer kind, nearest handle, press target, click
and drag actions, double click, stepPress) is now the editor canvas's only
decider; useCanvasGestures no longer uses clickHandler, useEmptyTap,
swallowClick nor useMoveGesture (deleted), and the box has no onClick.
canPan = dragAction(press target).type === 'pan' at pointer-down; framing
keys only when the box itself is the target. workStore.ts holds the U2
work with stale stamps: dropStale on every documentText change, reset on
a new session (editorStore subscription). The catalogue keeps useFraming
defaults and useEmptyTap, unchanged.
Named change (A13): a finger drag on a selected zone's (or circle link's)
interior moves nothing; the selected point is its own handle and still
moves by any pointer. Second observable change: touch and pen taps are
decided at the pointer-up from the press target, no longer from the
following click (no browser touch adjustment toward a marker).
tests/setup.ts: happy-dom has no elementsFromPoint; a shim answers the
pointerdown target and its ancestors.
Rewritten tests (editor-annotations-tab.test.tsx, D6b):
- "a touch tap deselects in its own click, not at the pointer-up" ->
"a touch tap deselects at the pointer-up" (deselected after pointerup)
- "a touch tap whose click lands on a marker (adjustment) selects that
marker" -> "a touch tap is decided by its press: a click adjusted toward
a marker selects nothing" (deselected, announced, marker b not current)
- "a pen tap deselects in its own click, like a touch tap" -> "a pen tap
deselects at the pointer-up, like a mouse tap"
- "a pen tap whose click lands on a marker selects that marker" -> "a pen
tap is decided by its press: a click landing on a marker selects
nothing"
Added: "a finger drags the selected point (its handle), never a selected
zone (A13)".
Drive-only changes, assertions unchanged: canvas fireEvent.click(x, y)
-> pressAt (pointerdown + pointerup) in the tab suite (C26, tracing,
paths, H3, H6, H8, filter reselect), editor-keep-alive and
editor-sections; drag moves carry buttons: 1 (C34, polygon block move,
H8 lostpointercapture and read-only, editor-no-loss drag in flight),
since a move without a button is a hover.
PreviewState.min_point_radius maps a view id to the smallest point
hitbox radius that passes TooSmall on the smallest preview, so the
sheet editor can mark it on a size slider. Twin constants named in
layer.rs; frame.rs shares one first-view drawing lookup.
Busy look, stopped-engine tone, focus when no retry is left, board 30
deviation fixed, cleanup name check, write-budget row as R4's own, row C
built by R4 pending the owner's validation, exact results of the wave.
D4 identity check: the integrity spec reads each own sheet's name by its
captured id and archives it only when it is exactly the seeded name.
seedGrant checks the account id as a captured id, now shared from
records.ts.
Board 30: in a three-column card at 1280 px the two buttons wrapped by a
few pixels. The failure row narrows their padding only there; the wrap
stays as the fallback. The network e2e now checks both on one row.
A retry ending on a failure with no retry (network to integrity, engine
stopped for good) keeps the polite region and moves focus to Consulter.
Busy look on Réessayer, error tone once the engine stopped, one attempt
for two calls in one tick; tests for the stopped engine, two core cards
and no blob URL after unmount.
- CHANGELOG: the six shell failures split into the four Target crashed
of BLK-015 and the two A0 lookup timeouts (counted as BLK-015, not A/B
proven); pgTAP 014 / 021 not proven by a run without M9; the intro
names the R3 entry's branch
- owner trial 4.15: steps strictly in order (step 5 shows what two
sheets prove, the move back to the top is step 9's last action), how
to open the working copy, the trimmed text, no Claude sentence, the
re-save sentence removed
- deploy order (database first, then the application) and its
fail-closed consequence in the SQL checkpoint and SUPABASE.md
- design review, code review and security gate results, the fix wave
run, the owner decisions left open; verifier, check:all and the owner
trial still pending
- TODO: fix wave line, three follow-ups
- the three fixture statements pass every value through literal(); the
sheet, draft and actor ids are refused unless they have the uuid shape
- test.use({ timezoneId: 'UTC' }): a row still in Paris time proves the
page sets the zone itself, whatever the host's
- the home ordering loop first asserts one <time> per item
- ratifiedDate escapes the dictionary text (escaped, now exported by
support/admin.ts)
- the long-search check types the text again and presses Escape: field
empty and focused, count back (Chromium's native type=search included)
R-24: no assertion removed or rewritten. Added: 'one date per home item'
(toHaveLength) before the unchanged ordering loop; the Escape block after
the unchanged button block. ratifiedDate builds the same pattern for the
current dictionaries (no RegExp character in them).
Run: admin-sheets-list 4 passed (fr, en x desktop, 375).
- the grid aligns its cards at the start; Espèces and Accès et
habilitations stack in one right-hand column at their own height
- when the sheets card is refused, that column spans the grid and lays
its cards side by side, so no half of the page stays empty
- the card's sheet name links are 44 px tall on the phone, unchanged
from sm up
- the card item keeps the list row's rhythm (space-y-1), a larger gap
only above the working copy action
No test changed.
- useSearch holds the query, its field and clearSearch; useListView keeps
the filters and the sort (8 bindings down to 5)
- the search term is folded once per search (sheetNameMatcher beside
matchesSheetName, whose signature and behaviour stay), each date read
once before the sort, the shown rows memoised on list, filters, search,
sort and language
- long className strings hoisted into module constants, no class changed
- the no-result block's clear button takes the page's 44 px phone height
- labels.ts: the SheetSummary alias removed; ResultsProps declared first
No test changed: the existing unit suites are the oracle.
The timestamp shape check lets an offset such as +99:99 through, which
Date.parse reads as NaN and the date formatter would throw on. The key
now also requires a readable date, so the case folds into unavailable.
Test added beside the existing refusal test; no assertion changed.