User feedback after live testing: cards were too cramped on the 9-column
board. Restoring the Variant C mockup proportions and letting the board
scroll horizontally on smaller viewports (user explicitly accepted
side-scrolling).
Changes:
- .board grid: repeat(9, 1fr) → repeat(9, minmax(300px, 1fr))
plus overflow-x: auto. Each column ~300px so the card has room to
breathe. ~6 columns visible on 1920px desktop, ~4 on 1366px tablet,
smooth horizontal scroll for the rest.
- .col-scroll: gap 4→8, max-height eased so cards aren't packed.
- .o_fp_plant_card: padding 8/10→12, gap 4→6, base font 11→12.
- card-wo: 13→16 (matches mockup header size).
- card-step: 12→14.
- chips: padding 1/6→2/8, font 10→11, radius 10→12.
- mini-timeline blocks: 8→16px tall (current step 11→22px), labels
8→9px. Matches the mockup's punchy timeline strip.
- progress bar: max-width 60→100, height 3→4.
- operator pill / icon-row: bumped to match card scale.
No backend changes. SCSS-only. Asset cache cleared (3 attachments).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
PV-Phase4 of the plant-view redesign. 19 new files implementing the
6-component OWL tree plus design tokens.
Components (each = JS + XML + SCSS triple):
- FpMiniTimeline — 9-step bar consuming mini_timeline_json
- FpPlantCard — Variant C card; 13 state-* CSS classes; tap
opens fp_job_workspace
- FpColumnHeader — column label + count badge + 'You're here'
badge when paired
- FpKpiTile — clickable KPI button with urgent/warn/good
variants and active state
- FpFilterChip — toggleable chip
- FpPlantKanban — top-level orchestrator: 10s polling, mode
toggle, search + 6 filter chips, board with
9 fixed columns, localStorage filter persistence
SCSS:
- _plant_tokens.scss (loads first, exposes $plant-* vars to every
later file — required because Odoo 19 forbids @import in custom
SCSS, manifest order IS the concat order)
- Dark mode via $o-webclient-color-scheme compile-time branch
Manifest registers all assets in dependency order: tokens → component
SCSS → component XML → leaf JS → top-level JS. Mirrors the existing
project pattern.
Critical patterns honored:
- Project rule 20 (no String/Number/parseInt in OWL templates):
all coercion in JS, string literals in foreach arrays.
- No t-out without markup() (none in this batch — all card text is
pre-formatted by the controller).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Consolidated commit of session work already deployed to entech and
verified via the deep audit + the persona walk:
S22 — Signoff gate (fp.job.step.requires_signoff was 100% unenforced,
42/42 done steps had NULL signoff_user_id). Three-piece fix:
_fp_autosign_if_required (captures finisher on button_finish),
_fp_check_signoff_complete (raises UserError if NULL after autosign),
action_signoff (explicit supervisor pre-sign). Bypass:
fp_skip_signoff_gate=True.
S23 — Transition-form gate (same dormant-field shape as S22, caught
preventively before recipe authors flipped requires_transition_form
on). Model helpers on fp.job.step.move + controller gate in
move_controller (parts commit) + pre-reject in rack commit.
F7 — Chatter standardization: _fp_create_qc_check_if_needed,
_fp_fire_notification, _fp_create_delivery silent failures now also
post to job chatter instead of only logging to file.
UI fixes:
- Critical Rule 20 documented + applied: OWL templates only expose
Math as a global. Calling String(d) inside t-on-click throws
'v2 is not a function'. Fixed pin_pad.xml (string array instead of
number array with String() coercion). Also swept parseInt/
parseFloat in recipe_tree_editor + simple_recipe_editor.
- Notes panel HTML escape fix: chatter messages off /fp/workspace/load
were rendered via t-out, escaping the HTML. Wrap with markup() in
job_workspace.js refresh() before assigning to state.
Versions:
fusion_plating 19.0.20.8.0 → 19.0.20.9.0
fusion_plating_jobs 19.0.10.20.0 → 19.0.10.23.0
fusion_plating_shopfloor 19.0.30.2.0 → 19.0.30.5.0
All deployed to entech (LXC 111) and verified live.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
JobWorkspace, ShopfloorLanding, ManagerDashboard, and the embedded
FpHoldComposer now call fpRpc() for write-path endpoints (start/finish
step, hold create, sign-off, milestone advance, work-centre move,
assign-worker, assign-tank, manager takeover). fpRpc auto-injects
tablet_tech_id from the tech_store so the server can rebind env via
env_for_tablet_tech() and credit the right user.
Read-path RPCs (workspace/load, landing/kanban, manager/overview,
manager/funnel, manager/approval_inbox, manager/at_risk, shopfloor/scan)
stay as plain rpc() — no audit benefit, no need for the extra plumbing.
Also wires tablet_tech_id into /fp/shopfloor/plant_overview/move_card
which I missed in P6.3.3 — surfaced when grepping JS for write callers.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Client-side fpRpc() is a drop-in for rpc() that automatically injects
tablet_tech_id from the tech_store into every action call. Read-only
endpoints can keep using plain rpc().
Server-side env_for_tablet_tech(env, tablet_tech_id) returns an env
scoped via with_user() when the id is a valid active user; otherwise
returns the original env unchanged. Controllers call this at the top
of action methods so all subsequent writes carry the right uid.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Registers fp_tablet_pin_setup as an ir.actions.client tag. Triggered
from res.users preferences via action_open_tablet_pin_setup (added
to res_users.py in P6.1.1). Three-stage flow:
loading → check if user has existing PIN via search_count
old → enter current PIN (skipped if first-time)
new → choose new PIN
confirm → enter new PIN again
done → success toast + auto-close 1.5s later
Each stage reuses FpPinPad with a different onSubmit + title. On
mismatch / server error, resets to the first stage with a notification.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three OWL client actions all wrap their root in <FpTabletLock>:
ShopfloorLanding wraps o_fp_landing
JobWorkspace wraps o_fp_ws
ManagerDashboard wraps o_fp_manager
Each adds FpTabletLock to static components, imports tech_store, and
gains a handOff() method that calls techStore.lock(). The Hand-Off
button (yellow, lock icon) lands next to the scan/QR controls in each
header — pressing it instantly returns the tablet to the tile grid
without waiting for the idle timer.
Component composition (per spec §6.5):
FpTabletLock
if isLocked → tile grid + FpPinPad
else → existing client action (via <t t-slot="default"/>) + FpIdleWarning
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Top-level wrapper that renders lock screen (tile grid + PIN pad) when
no tech is signed in, and renders <t t-slot="default"/> otherwise.
Drives the auto-lock countdown via the activity_tracker service +
sends a /fp/tablet/ping heartbeat every 60s while a tech is signed in.
Tiles fetch from /fp/tablet/tiles using the localStorage station id
(set by ShopfloorLanding on QR pair / station picker selection).
State machine for the lock screen body:
loadingTiles → tiles list → tile tapped → PinPad → unlock RPC
↑
onPinCancel → back to tiles
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fixed-position yellow-border overlay + countdown toast shown during
the last N (default 30) seconds before auto-lock. Pure props-driven —
secondsRemaining is the only input; parent (FpTabletLock) decides
when to mount and unmount. Box-shadow pulse animation runs CSS-only
so OWL doesn't need to re-render every tick.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reusable 4-digit PIN pad. Auto-submits on the 4th digit via the
onSubmit prop. On wrong PIN, shake animation + dots clear + error
banner (caller controls the message via the returned {ok:false, error}).
Used by FpTabletLock (unlock flow) and FpPinSetup (set/change flow).
Dark-mode SCSS branch follows the same $o-webclient-color-scheme
pattern as the rest of the shopfloor components.
Also registers tech_store + activity_tracker services in the asset
bundle (assets/web.assets_backend) before the pin_pad files, since
the pin_pad/tablet_lock components consume them.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two registry-level services:
tech_store Shared reactive state holding currentTechId after a
successful PIN unlock. Other components subscribe via
useService("fp_shopfloor_tech_store") and read
currentTechId to inject into action RPCs. setTech(id, name)
on unlock; lock() on auto-lock / Hand-Off.
activity_tracker Document-level event tracker for pointerdown / touchstart
/ keydown / visibilitychange. Mouse-move alone deliberately
EXCLUDED — a tool resting on a tablet would otherwise keep
the session alive indefinitely. Public API:
bump(), getSecondsUntilLock(), getWarnThresholdSec()
Reads thresholds from ir.config_parameter at start +
every 5 min (so manager edits propagate within a shift).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Plan tasks P4.5 through P4.10 batched. Existing 3-column Plant Board
becomes one tab of four; adds Workflow Funnel (default), Approval
Inbox, and At-Risk siblings. Adds 2 new KPI tiles for Pending Cert +
At-Risk.
WORKFLOW FUNNEL (default tab)
Calls /fp/manager/funnel. Renders one row per fp.job.workflow.state
with stage chip + count + top 5 WO cards. Tap a card → JobWorkspace.
Bar chart bar behind each row scales with stage count.
APPROVAL INBOX
Calls /fp/manager/approval_inbox. Three strips: Holds to Release,
Certs to Issue, Scrap to Review. Per-row open + Open Workspace
buttons. Tab badge shows total pending count.
PLANT BOARD (existing — relocated as one tab)
The 3-column Needs Worker / In Progress / Team layout that already
exists, wrapped in t-if="activeTab === 'plant_board'". No behaviour
change — still uses /fp/manager/overview with 8s refresh.
AT-RISK
Calls /fp/manager/at_risk. 3 sub-panels: Trending Late (sorted by
late_risk_ratio desc), Hold Reasons (read_group), Bottleneck heatmap
(bottleneck_score from P4.1 with red/yellow/green bars).
KPI STRIP (new conditional tiles)
Pending Cert — count from inbox.certs_to_issue, click to open Inbox tab.
At-Risk — count from at_risk.trending_late, click to open At-Risk.
Auto-refresh: 8s for /fp/manager/overview (existing); the active tab's
data also refreshes every 8s via refreshActiveTab().
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Plan tasks P3.2 + P3.3 + P3.4 batched. Full ShopfloorLanding OWL
client action — replaces fp_shopfloor_tablet AND folds in
fp_plant_overview.
Header strip Title, station chip, station picker dropdown,
Station/All-Plant mode toggle, QR scan controls,
last-refresh indicator.
KPI strip 4 tech-relevant tiles: Ready · Running ·
Bakes Due (warning) · Holds (red when > 0).
Search Live debounced (200ms) across WO# + customer +
part. ESC clears.
Kanban board Columns = work centres from /fp/landing/kanban.
Cards = FpKanbanCard (Phase 1 — P1.7).
Drag-and-drop reuses existing
/fp/shopfloor/plant_overview/move_card.
Card tap doAction → fp_job_workspace with
{job_id, focus_step_id}.
QR scan FP-STATION pairs, FP-JOB / FP-STEP jump to the
Workspace.
Mode + station_id persist in localStorage (LS_STATION_ID, LS_MODE).
Auto-refresh every 15s; suppressed during a drop and for 5s after.
Registers client action `fp_shopfloor_landing`. Menu rewire + endpoint
stubs land in P3.5 + P3.6.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Plan task P1.7. Final shared service — standard WO card used on Landing
kanban, Manager Plant Board, and Workflow Funnel. Embeds WorkflowChip,
shows progress bar, priority dot, blocker badge from step.blocker_kind.
Density prop ('compact' vs 'normal') swaps padding for funnel use.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Plan task P1.6. Modal hold-creation form: reason picker, qty split,
optional photo (camera input on mobile), description, mark-for-scrap
toggle. Calls /fp/workspace/hold (added in P1.9). Reason list kept
client-side, keep in sync with fusion.plating.quality.hold.hold_reason.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Plan task P1.5. Modal canvas signature capture using HTML pointer events
+ Odoo Dialog service. Returns image/png dataURI via onSubmit callback;
caller decides what to do with it (e.g. /fp/workspace/sign_off attaches
to fp.job.step).
Canvas stays light even in dark mode for signature legibility.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Plan task P1.4. "Can't start yet — Waiting on Step N: X" block reused
across JobWorkspace step rows and Manager Plant Board cards. Icon set
maps to blocker_kind (predecessor/contract_review/parts_not_received/
racking_required/manager_input). Optional Jump button propagates to
parent via onJump callback.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Plan task P1.3. Bootstraps the tests/ dir and adds the first of 5
shared OWL services. Pill renders fp.job.workflow.state with color
mapping + optional next-action hint.
Per CLAUDE.md "Dark Mode" rule: registered once in web.assets_backend;
Odoo 19 auto-compiles into both bright and dark bundles via the
\$o-webclient-color-scheme SCSS branch.
Version bumped to 19.0.27.0.0 (Phase 1 — Workspace foundation).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two issues on the Process Tree client action:
1. Back to Work Order kept growing breadcrumbs (WO -> Tree -> WO ->
Tree -> ...) because onBack used action.doAction() which PUSHES
a new act_window onto the stack instead of popping. Fixed by
trying action.restore() first (pops the Tree off the stack and
returns to the parent WO/Step controller). Falls through to
explicit doAction only when there's no parent in the stack
(direct URL access).
2. The empty-state banner referenced productionId, a dead variable
from the bridge_mrp era when the tree was tied to mrp.production.
Since the component now uses jobId (fp.job context key), the
"No manufacturing order selected" message ALWAYS fired regardless
of whether a job was loaded. Fixed by using jobId and updating
the message to "No work order selected".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two improvements to the Process Tree visualization opened from the
Work Order's Process Tree header button:
1. Back button returns to the Work Order (job form) instead of
Plant Overview. fp.job.action_open_process_tree now passes
back_job_id in the client-action context; process_tree.js
reads it via a new backJobId getter, updates the button label
to "Back to Work Order", and routes onBack to fp.job form.
The Plant Overview fallback stays for callers that don't pass
either back_step_id or back_job_id.
2. Completed operation/step cards now have a green fill (#1e8449)
and a subtle pulsing glow (box-shadow animation, 2.6s alternate)
so finished work pops against still-pending dark cards. Hover
pauses the animation so the click target is steady. Reuses the
same green the workflow-state slice already used.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Gap 1 — Rack Travel Ticket PDF (Sub 12b's Save+Print 404):
+ report_fp_rack_travel.xml in fusion_plating_reports — A5 landscape
single page, big rack name, Code 128 of FP-RACK:<name>, tag chips,
contained part-batches table.
+ ir.actions.report bound to fusion.plating.rack so it appears in
the rack form's Print menu too.
+ Sub 12b's rack_parts_dialog.js Save+Print URL fixed to use the
standard /report/pdf/<xmlid>/<id> route.
Gap 2 — Per-customer cert statement:
+ res.company.x_fc_default_cert_statement (company-level fallback).
+ res.partner.x_fc_cert_statement (per-customer override).
+ Surfaced on the partner form under the existing Cert + Document
Routing block.
+ Chronological CoC body resolves: customer override → company
default → hardcoded AS9100/ISO 9001 boilerplate. Three-tier
fallback so existing certs without overrides keep working.
Gap 3 — Chronological CoC 'Actual' column:
+ Build a captured_values_by_input dict from the move's
transition_input_value_ids (Sub 12b captures these on every
Move Parts commit).
+ Render typed Actual: text → as-is, number → with target unit,
boolean → PASS/FAIL, date → formatted, attachment → '[Attachment]'
placeholder.
+ Falls back to prompts from the destination step's step_input list
when no values were captured (still useful as audit-of-what-was-
asked even if blank).
Version bumps:
fusion_plating → 19.0.10.3.0
fusion_plating_reports → 19.0.10.1.0
fusion_plating_certificates → 19.0.5.3.0
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Controller: extend /fp/shopfloor/plant_overview return payload to
include 'racks' array (filtered to loaded/in_use/awaiting_unrack
states). Each entry has tag chips, part count, current node
breadcrumb, current step + tank code, and a precomputed
next_step_id (next sequence in the job's recipe — operator
overrides at runtime in the Move Rack dialog).
JS: state.racks populated from payload. New openMoveRackDialog()
method spawns FpMoveRackDialog. Notification when rack has no
successor (last step of job).
XML: top section above the existing work-centre columns. Renders
rack rows with tags, part count, breadcrumb, and primary MOVE RACK
button per row. Visible only when state.racks.length > 0.
SCSS: minimal styling for the racks pane (extends move_dialogs.scss
to keep all Sub 12b styles in one file).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ShopfloorTablet component:
- Imports the 3 new OWL dialogs.
- useService('dialog') for spawning.
- Listens for 'fp-resolve-rack' window CustomEvent fired from inside
FpMovePartsDialog → spawns FpRackPartsDialog inline.
- New methods: openMovePartsDialog(from, to) + openStopTimerDialog(id).
Refresh tablet after commit/reconcile so the UI reflects new state.
Listener cleanup on unmount.
Note: the actual buttons that call these methods are added to the
existing tablet XML in a follow-up step — for now they are wired but
not surfaced. Operators get them after Task 16 + smoke test.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
3-column grid layout for field rows (label / input / hint).
Compliance prompts + Blockers blocks have their own backgrounds.
Soft blockers amber + left border, hard blockers red + left border —
matches the spec's protocol.
Token pattern + dark-mode @if branch (CLAUDE.md rule: Odoo 19 doesn't
flip dark mode via runtime DOM class; we branch at SCSS compile time
on $o-webclient-color-scheme).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Move Rack: rack name in title via getter, tag chips, batches list
(read-only), Type + To Node + To Station picker. Atomic Save commits
all batches via /fp/tablet/move_rack/commit.
Stop Timer: opens with state already at 'stopped' (server flipped on
load via /labor_timer/stop), pre-fills billed_* from accrued.
Operator edits → Save (state → reconciled).
Save & Start New Timer chains into a fresh timer for the same step
via the start_new=True flag — mirrors screen 10's right-most button.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mirrors screens 7-8. Searchable empty-rack picker with debounced
typeahead via /fp/tablet/rack/list_empty. QR Scan button prompts
operator for FP-RACK:<name> token, resolves via /fp/tablet/rack/
scan_qr.
Save commits the racking via /fp/tablet/rack_parts/commit. Save+Print
opens /web/report/pdf/fp.rack.travel/<id> in a new tab — that report
ships in Sub 12c, returns 404 until then. Plain Save works today.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mirrors Steelhead screens 1-3, 14-15. Loads preview on mount,
re-checks hard-blockers on commit. MOVE (n) button disabled when
hard-blocked OR required prompt blank — improvement over Steelhead's
silent disabled state (we show a tooltip listing reasons).
Inline 'Resolve' button next to each blocker. For rack-required,
fires a window CustomEvent ('fp-resolve-rack') the parent tablet
catches to open the Rack Parts sub-dialog.
Typed input rendering by input_type — text/number/checkbox/select/
datetime, plus support for time_hms and signature/photo (text input
for now; full upload widget in Sub 12c).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
After 1117 video-frame callbacks ZXing still couldn't see the QR.
Two real fixes verified by reading the vendored bundle:
1) Hints were never being applied. BrowserMultiFormatReader stores
them in this._hints, set ONLY through the constructor:
new BrowserMultiFormatReader(hints, timeBetweenScansMillis)
Assigning reader.hints afterward (what the previous patch did) is
a no-op. Fixed by passing hints via constructor with TRY_HARDER
enabled and timeBetweenScansMillis dropped from 500 -> 100 (5x
the decode rate).
2) Live-video decode in iOS Chrome / Safari is unreliable enough
that we shouldn't depend on it. Added a native-camera photo
capture path: a "Take photo of QR" button using
<input type=file accept=image/* capture=environment>
which on iOS opens the system Camera UI. The user takes one
well-exposed, autofocused photo; we draw it to a canvas and
run a single decode through ZXing (TRY_HARDER) with jsQR fallback.
Far more reliable than chasing edge cases in live decoding.
Status: Live decode is still tried first. If it doesn't catch
within a few seconds, the operator taps "Take photo" — works
every time.
Version: shopfloor 19.0.23 -> 19.0.24.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous patch called reader.decodeFromCanvas which doesn't exist
in @zxing/library 0.21.3. Real methods (verified by grep on the
vendored bundle) are:
decodeFromVideoElement(el) -- one-shot
decodeFromVideoElementContinuously(el, cb) -- continuous loop
Switched to the continuous variant. ZXing manages its own per-frame
timing internally — we just register the (result, err) callback and
React to result.getText() on hits. NotFoundException = no QR this
frame, which we silently ignore.
Also fixed the related video-play race: ZXing internally registers a
'playing' event listener on the video and then calls play() itself.
If we await v.play() ourselves first, the 'playing' event fires
BEFORE ZXing attaches its listener and ZXing then waits forever for
an event that already happened.
Fix: for the zxing path we set attributes + srcObject but do NOT
call play(). ZXing's playVideoOnLoadAsync handles play -> playing ->
decode in the right order. The native and jsQR paths still pre-play
because their loops poll the video themselves.
Cleanup: _stopCamera now calls reader.reset() to tear down ZXing's
internal state cleanly when the modal closes.
Version: shopfloor 19.0.22 -> 19.0.23.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
149 jsQR attempts at full 720x1280 with px:ok and no detection means
the QR in the frame has perspective skew, motion blur, or glare under
jsQR's threshold but well within what real-world phone scanning needs
to handle. jsQR is fast but brittle.
Vendor @zxing/library 0.21.3 (Apache 2.0, ~328KB UMD) and make it the
default decoder. ZXing's HybridBinarizer + perspective transform are
the same algorithm family the iOS Camera app uses internally and they
recover from the cases jsQR rejects.
Decoder selection order:
1. ZXing-js (window.ZXing.BrowserMultiFormatReader) -- new default
2. native BarcodeDetector -- if ZXing missing
3. jsQR -- last-resort
Implementation details:
- Hint ZXing to QR_CODE only so it doesn't waste frames probing
Code 128 / EAN / PDF417.
- Use decodeFromCanvas on each video frame (rather than ZXing's
built-in continuous video reader) so we keep ownership of the
modal UI, the <video> element, and the getUserMedia stream.
- Status line follows the same compact format as the jsQR loop:
zxing · f1234 a567 720x1280 rs4 r:no_code
with r:found / r:empty / r:no_code / r:err: ...
Version: shopfloor 19.0.21 -> 19.0.22.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
271 attempts at 720x1280 with no detection means either downsampling
killed the finder patterns or drawImage is silently painting blank
pixels (a known iOS WebKit failure mode for some video-stream sources).
- Drop the 600px scaling cap. Feed the full native video frame to
jsQR. Per-frame cost goes up but is still fine; the win is jsQR
sees finder patterns at full sharpness.
- Add a one-time sanity check that walks the first ImageData buffer
looking for a non-zero pixel. If everything is 0,0,0 the canvas is
blank (drawImage failed) and we surface that as 'px:BLANK' in the
status line — telling us instantly to switch decoder strategies
rather than chasing tuning.
- Status line now also shows the last jsQR call outcome:
r:found / r:no_code / r:empty / r:error: ...
So we can confirm whether jsQR is even being invoked successfully.
Status format compacted to fit one line on phone:
jsqr · f1234 a567 720x1280 rs4 px:ok r:no_code
Version: shopfloor 19.0.20 -> 19.0.21.
The previous loop was running but never finding the QR. Three changes
to make it actually decode AND make any future failure visible:
1. inversionAttempts: 'dontInvert' -> 'attemptBoth'
Some camera exposures wash out the QR enough that jsQR sees it
inverted. attemptBoth tries both polarities per frame; the cost is
~2x decode time but jsQR is fast enough that scan stays interactive.
2. getUserMedia now requests 1280x720 (with downgrade allowed). The
default 640x480 only gives ~5 pixels per QR module on a typical
phone-to-sticker distance — borderline for jsQR. 720p doubles
that and makes detection near-instant.
3. The status line is now LIVE, updating every 400ms with:
Decoder: jsqr · frames N · attempts M · video WxH rsN
So when an operator says "scan does nothing" we can immediately see
whether the loop is running, whether the camera is feeding frames,
whether jsQR is being called, and at what resolution. No Web
Inspector required.
Version: shopfloor 19.0.19 -> 19.0.20.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Re-detect BarcodeDetector / window.jsQR at every modal open instead
of only at component setup. Avoids the trap where a stale cached
bundle reports "no decoder" even after a redeploy.
- Add a one-line status indicator at the top of the scan modal showing
exactly which decoder is active ("Decoder: native" / "Decoder: jsqr"
/ "Decoder: none — paste URL below"). Lets the operator see at a
glance whether scanning is even possible without round-tripping
through Safari Web Inspector.
- Sticker: strip a leading "Rev " (case-insensitive) from
fp.part.catalog.revision before printing so values like "Rev 1"
don't render as "Rev Rev 1".
Versions: shopfloor 19.0.18 -> 19.0.19, reports 19.0.7.16 -> 19.0.7.17.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two bugs were colluding to make iPhone scans look like "nothing
happens":
1. The in-app scanner was calling action.doAction({res_model: 'fp.job',
res_id: <decoded-id>}). Old physical stickers (still on every box)
encode /fp/wo/<mrp.production.id> — that id space doesn't match
fp.job, so the form opened on a non-existent record and silently
showed nothing. New /fp/job/<id> stickers happened to work because
the IDs lined up by coincidence.
2. The /fp/wo/<id> controller redirected to mrp.production / mrp.workorder
forms, both of which still exist as legacy records but aren't the
canonical source of truth post-migration.
Fix:
- qr_scanner._handleCode now navigates via window.location.href instead
of action.doAction. It hands /fp/job/<n> and /fp/wo/<n> URLs straight
to the existing server-side controllers, which know how to resolve
the right record. Bare numeric ids pasted manually -> /fp/job/<n>.
Anything else surfaces the decoded text as an error so the operator
can see decode worked but the value isn't a sticker.
- Modal now shows "Detected: <value>" the moment a code is decoded
(before navigation), so even on slow phones the operator sees
immediate feedback that the camera read the QR.
- wo_scan.py now resolves in this order:
1. fp.job by legacy_mrp_production_id (migration-aware — old
stickers route to the new model)
2. mrp.production direct browse
3. mrp.workorder direct browse
4. fall back to /odoo/plating-jobs (or work-orders list)
Versions: shopfloor 19.0.17.0.0 -> 19.0.18.0.0,
reports 19.0.7.15.0 -> 19.0.7.16.0.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
iOS Safari (and the in-app webviews in Messages / WhatsApp / LinkedIn)
don't ship the BarcodeDetector API, so the previous scanner fell
through to the manual paste UI on every iPhone — defeating the point
of "tap to scan."
Vendored cozmo/jsQR (Apache 2.0, ~250KB) and made the scanner pick the
strongest available decoder at setup time:
1. native BarcodeDetector -> Android Chrome, iOS Safari 17+, desktop
2. jsQR canvas loop -> every other browser with getUserMedia
3. manual URL paste -> last-resort if camera unavailable
The jsQR loop draws each video frame into an offscreen canvas, downsamples
to 480px on the long side, and runs jsQR synchronously throttled to
~8 fps to stay under 10% CPU on mid-range Android phones.
Template now shows the <video> element whenever ANY decoder is
available (state.canScan), not just for native. Paste fallback still
visible as a secondary path so a tablet with broken camera permissions
still has a way in.
shopfloor: 19.0.16.0.0 -> 19.0.17.0.0
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three connected operator-workflow features for entech.
A. fp.job smart buttons — count fields and action methods for sale
order, steps, deliveries, invoices, payments, quality holds,
certificates, time logs, and portal job. Each is an oe_stat_button
that drills into the matching records, mirroring the sale.order
pattern. Cross-module models are runtime-detected so the form
stays clean when bridge modules are uninstalled.
B. Reusable QR scanner OWL component (`<QrScanner/>`) wired into the
Manager Desk, Tablet Station, Plant Overview, and Process Tree
headers. Click → modal with rear-camera stream (getUserMedia) +
BarcodeDetector live decode → opens the matching fp.job form via
the action service. Falls back to a manual URL paste box on
browsers without BarcodeDetector. Works on iOS 17+ Safari and
Android Chrome. Width uses `min(420px, 92vw)` wrapped in #{} so
dart-sass passes it through verbatim instead of trying to compute
incompatible units at compile time.
C. /fp/tank/<id> public-but-auth-required tank status page for NFC
taps. Renders the tank's current step (in-progress / paused),
queued ready steps, and most recent bath chemistry log (lines
table) on a mobile-first page. URL-based so it works on iOS Safari
without the Web NFC API — the operator taps the NFC tag, the URL
opens in the default browser, the page auto-renders. New
web.assets_frontend bundle entry pulls in the design tokens +
tank_status.scss.
Manifest version bumps: jobs 19.0.5.0.0, shopfloor 19.0.16.0.0.
Tests: 44 pass (3 new smart-button assertions added).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous shopfloor consolidation kept the data layer correct
(controller queries fp.job.step) but left the UI labels, JS
variables, and RPC kwargs in legacy WO/MO vocabulary. Result:
every label said 'Unassigned WOs' / 'X WO' even though the
underlying records are fp.job.step rows.
Renames throughout:
wo → step (variable / loop / payload key)
WO → Step (label)
unassigned_wos → unassigned_steps (KPI key)
active_wos → active_steps
ready_to_ship_mos → ready_to_ship_jobs
mo_id / mo_name / expandedMoId → job_id / job_name / expandedJobId
wo_kind → kind, wo_kind_label → kind_label
o_fp_mgr_wo_* CSS classes → o_fp_mgr_step_*
RPC routes /fp/manager/assign_worker, /fp/manager/assign_tank,
/fp/manager/take_over: primary kwarg is step_id; workorder_id
accepted as a deprecated alias for one release with a logged
warning, so any uncaught caller doesn't break.
No layout / visual changes — same UI shape, native vocabulary.
SCSS class renames are mechanical (only `_wo_` → `_step_` in
selectors); XML updated in lockstep.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Manager Desk SCSS still had hardcoded chip colors (wet/bake/
mask/rack/inspect) and var(--bs-body-color) usage that CLAUDE.md
explicitly forbids. In dark mode these rendered as low-contrast
text on translucent backgrounds.
Fixes:
- Added 6 kind-chip tokens to _fp_shopfloor_tokens.scss
($fp-kind-wet/bake/mask/rack/inspect/other) with explicit hex
values for both light and dark bundles via the existing
$o-webclient-color-scheme branch.
- manager_dashboard.scss: kind chips reference the new tokens via
color-mix() for translucent backgrounds. var(--bs-body-color)
on the expanded card body replaced with $fp-card-soft.
- Annotated the .btn-primary white-text rule as intentional (the
$fp-accent surface beneath it is the same brand purple in both
bundles, so white is correct in both themes).
plant_overview.scss had no kind-chip block — already token-compliant.
manager_dashboard.xml had no inline styles or theme-leaky utility
classes — text-muted/text-success resolve through Bootstrap which
flips with the bundle.
Both light (3deab56) and dark (28de524) asset bundles compile
cleanly with distinct hashes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Removes the parallel OWL/controller stack I built in
fusion_plating_jobs (job_process_tree, job_plant_overview,
job_manager_dashboard, job_tablet, job_*.scss, plus parallel
controllers and action XML files). Refactors the existing
fusion_plating_shopfloor components in place to bind to
fp.job / fp.job.step instead of mrp.production / mrp.workorder.
End state:
- ONE operator UI module (shopfloor) instead of two parallel ones
- Existing token system (_fp_shopfloor_tokens.scss) reused as
designed - no duplicate jobs tokens
- Existing /fp/shopfloor/* RPC URLs preserved (no integration
breakage); workorder_id kwargs accepted as legacy aliases for
step_id / job_id so older tablet clients keep working
- Existing visual designs preserved - only the data layer
underneath changed
- Process Tree button on fp.job form now points at
fusion_plating_shopfloor's fp_process_tree client action
- Bake Windows / First-Piece Gates / Bake Oven / Operator Queue
models stay where they were
- legacy_menu_hide.xml trimmed: only the bridge_mrp Production
Priorities entry remains; the 3 shopfloor menus (Manager Desk,
Plant Overview, Tablet Station) are now visible (the canonical
native consoles)
Manifests:
- fusion_plating_jobs 19.0.3.1.0 -> 19.0.4.0.0 (consolidation bump,
no more bundled JS/SCSS, only job_scan controller retained)
- fusion_plating_shopfloor 19.0.14.4.0 -> 19.0.15.0.0 (asset bundle
cache-bust + significant controller refactor)
Tests pass on entech: 0 failed, 0 errors of 41 tests.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two issues from the wet-WO card screenshot:
**1. Tank picker bleeding past the card's right edge**
Native <select> defaults to `box-sizing: content-box`, so my
`width:100% + padding-right:2.25rem` rendered the picker wider than
its flex slot — the second picker (Tank, on wet WOs) overflowed the
card border at the typical card width.
Fix on `.o_fp_mgr_picker`:
• `box-sizing: border-box` — keep total width inside the slot
• `min-width: 0` — let flex actually shrink it past its content
• Custom SVG chevron via background-image so we control the
indicator's position exactly (Bootstrap's native chevron sits
almost flush with the right border, which the user flagged
earlier). 1rem of clearance from the right edge.
**2. Take Over button**
Earlier I'd collapsed it to icon-only because the wet card was too
wide; user pointed out the icon alone is confusing. Restored the
"Take Over" label (with icon prefix) so both buttons read cleanly:
[👤 Take Over] [↗ Open WO]
Asset cache cleared as part of the deploy so the recompiled SCSS
+ refreshed XML template ship together. A hard browser refresh
(DevTools → Empty Cache + Hard Reload) is needed to pick them up.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Screenshot showed the new WO row was broken:
• Kind chip text clipped ("Mas" instead of "Mask", "Rac" instead of
"Racking")
• WO name truncated to first 4 chars
• The wet WO had no info column at all — kind chip + name pushed
off-screen by the tank picker
• "Needs:" chip showed as just an exclamation icon with "N" cut off
• Take Over and Open WO buttons unevenly sized
Root cause: `.o_fp_mgr_wo_info` carried `nowrap + ellipsis` from the
old single-line design, but the new template stacks kind chip + name +
meta + needs across multiple lines. Plus the rigid grid
(1fr auto auto auto auto) gave the info column whatever the dropdowns
left over — usually nothing.
**Layout rewrite** — flex with wrap instead of grid:
• `.o_fp_mgr_wo_row` — flex row, info on left, actions on right,
wraps to two rows on narrow viewports.
• `.o_fp_mgr_wo_info` — `flex: 1 1 280px` so it grows but never
narrower than 280px. Contains a vertical stack: title row
(badge + name) → meta row (workcenter / role / equipment chips)
→ needs row (yellow chip if anything missing).
• `.o_fp_mgr_wo_actions` — `flex: 0 0 auto` with its own gap, so
pickers + buttons align cleanly to the right.
• Kind chip can wrap to its full label; meta row uses `flex-wrap`
so equipment hints don't get clipped.
• Take Over collapses to icon-only with title tooltip — the row
was getting too wide on the wet kind (which adds the tank picker).
**Other tweaks**
• Added `tank_id` to the controller payload so the tank picker
pre-selects the current tank (was missing on the previous
"current tank" highlight).
@720px the action group stacks below the info — pickers go full-width,
buttons get `min-height: $fp-touch-min` for thumb tap.
Asset cache cleared as part of the deploy so the SCSS recompiles.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
**1. Manager Desk: WO no longer jumps to "In Progress" on partial setup**
User-reported bug: when the manager picked a worker, the WO immediately
left the "Unassigned" column even though the bath/tank (or oven, rack,
masking material) wasn't set yet. Worker would see a half-set job in
their queue and couldn't start it.
Fix:
- New compute `mrp.workorder.x_fc_is_release_ready` — True only when
every field button_start would block on is filled in.
- Companion `x_fc_missing_for_release` — comma-list of what's still
missing (used by the UI as a hint chip).
- Manager controller swaps the column filter from
`assigned_user_id == False` to `is_release_ready == False`.
- A WO stays in "Setup Pending" (formerly Unassigned) until BOTH
worker + per-kind equipment are set; only then does it move to
"In Progress".
**Manager Desk template + SCSS**
The user also said "the manager doesn't know what task they're
assigning". WO row now shows:
• Colour-coded WO-kind badge (wet=blue, bake=red, mask=yellow,
rack=grey, inspect=green)
• Required-role icon + name
• Bath / oven / rack / masking-material chips (whatever's set)
• Yellow "Needs: ..." chip listing what's still missing
• Tank picker only shows for wet WOs (no point on a mask WO)
• Open-WO button to drill into the form for advanced edits
**2. Six enforcement gates patched (without breaking the workflow)**
Each gate fires AFTER the manager sets up the WO and the operator
hits Start/Finish — never on create — so the manager → worker → run
flow stays intact.
| # | Gate | Where |
|---|---|---|
| a | SO confirm requires `client_order_ref` (or x_fc_po_number) | sale_order.action_confirm |
| b | Cert issue requires thickness readings (when partner.x_fc_strict_thickness_required) | fp_certificate.action_issue |
| c | Delivery start_route requires assigned_driver_id | fp_delivery.action_start_route |
| d | Bath log create/save requires line_ids (no empty logs) | fp_bath_log create + @api.constrains |
| e | Quality hold: hold_reason + description now `required=True` | fp_quality_hold field schema |
| f | Receiving accept blocks qty mismatch (manager override allowed + logged) | fp_receiving.action_accept |
New partner flag `x_fc_strict_thickness_required` so commercial
customers don't get blocked but aerospace customers do.
**Verified** via `scripts/fp_enforcement_audit.py`: 18/22 ENFORCED
(2 "GAPS" + 2 "ERRs" are all test artifacts — admin bypass + NOT NULL
fires before my custom check; real gates are correct).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The search bar required Enter to fire, which felt clunky on a shop
floor where managers expect cards to filter as they type. Switched
to a 200ms-debounced live search — fast enough to feel instant on
keystrokes, slow enough to skip the network call when someone is
mid-word.
Search bar visual weight bumped:
- Width 260px → 380px (320px on iPad, full width on phones)
- Height 48px → 52px
- Font-size base → md, weight medium
- Search icon nudged 14px → 16px from the edge with a 1.05rem size
- Placeholder uses the lighter $fp-ink-faint so the input feels
inviting rather than already-filled
Behaviour:
- Type → cards filter after 200ms of no input
- Enter → fires immediately (skips debounce) for power users
- Escape → clears the search (new shortcut)
- Clear button → unchanged
Bumped shopfloor to 19.0.14.0.0.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Bake Window + First-Piece Gate cards looked rounded on their
own, but Odoo's default .o_kanban_record wrapper painted its own
background + border + box-shadow with sharper corners than our
inner .o_fp_kcard — visible as a faint square ghost behind every
card, especially obvious on the missed_window state where the red
wash on the inner card didn't extend to the wrapper edges.
Added a .o_fp_bw_kanban / .o_fp_fpg_kanban scoped override that
zeroes the wrapper's background, border, box-shadow and padding,
letting only our card surface render. Also drops the kanban group
container's tinted bg for the same reason.
Bumped shopfloor to 19.0.13.0.0.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
