The changed-file maintainability gate caps functions at 150 lines;
VendorPortalDocuments reached 156. Move the per-document row into
VendorPortalDocumentRow with no behavior change.
A work order with a zero granted total rendered $0.00 in the Pending
table's Approved on WO column and in the detail modal's Total line.
Both now share isGrantedAmount, so zero or unknown totals render a dash
and the column always matches the modal Total.
Persisted HEIC files and local files with an empty or octet-stream type
were classified as "other", so they did not count toward the 10 photo /
3 video limit. Both classifiers now recognise them by extension.
The Photos & Videos uploader and Extra Docs ignore a new selection while
the previous one is still being screened, and the vendor portal drops an
earlier pick whose video check finishes after a newer pick.
Also removes ticket keys from source comments.
The tile link sent the board API token __unassigned__ as the dispatcher.
The drill-down parser stores that as an Advanced Filters dispatcher id,
while the filter UI only knows __unassigned, so the banner showed a raw
__unassigned__ chip, Assigned to did not tick Unassigned, and the option
could not clear it. Send ASSIGNEE_FILTER_UNASSIGNED, as the notification
queue link already does; toBoardSearchQueryParams still maps it to
__unassigned__ for /board/search, so the rows are unchanged.
The review contract bars Jira keys in source comments and test names;
ticket identity lives in the PR, commit and branch. The previous commit
cleaned the source files; this strips the six (SH-391) suffixes the
branch added to test names and a test comment, keeping the behavioural
wording.
The vendor completion upload and uplift evidence checked type, size and
duration but not the 10-photo / 3-video work-order limit. Use the counts the
dispatch detail now reports (shoc-backend#173), on the same basis as the
server: a new completion version does not count the document it replaces.
When the backend does not report counts yet, the server check still applies.
The browser pre-check decoded the picked file through a video element and
an object URL, which CodeQL flags as DOM text reinterpreted as HTML. Parse
the moov/mvhd movie header from file slices instead, the same way the
server enforces the 90-second limit, so both sides read one duration.
Unreadable headers still never block an upload.
Extra Docs advertised the 90-second limit but only checked type and size.
Both dispatcher surfaces now share one screening step (type, size, count,
duration), and the Completion Doc media tab and Extra Docs count the whole
work order's photos and videos rather than only their own tab's share.
Failed local uploads no longer count toward the limit.
Shows open work orders with no dispatcher for the selected period and
drills into the same list on the board (dispatcher Unassigned, open
statuses, same dates; All time includes undated work orders).
With no range selected, or with Unscheduled, the board search adds only
open undated work orders; completed and canceled ones without a date are
never unscheduled. The board page kept them. It now applies the same rule,
so both return the same rows.
A foreign declared type (e.g. video/3gpp on a .jpg) gave the file the video
size allowance in the browser while the server sizes it as a photo. Use the
same rule as the server: an allowlisted type decides, otherwise the extension.
A week-only row that still carries a scheduled date outside the range
passed the board search (its target week overlaps) but failed the
on-page filter, which judged it by the date alone. Both now accept a row
dated in the range or a week-only row whose target week overlaps it.
Any image/* or video/* type passed the client check, so GIF, WebP and WebM
were only rejected after upload. Resolve the kind from the same allowlist
the server uses: an allowlisted browser type, otherwise the extension.
Photos up to 10 MB (JPG/PNG/HEIC), videos up to 100 MB and 90 s (MP4/MOV),
at most 10 photos and 3 videos per work order, pre-validated with stable
generic messages on the Photos & Videos modal, the Completion Doc media tab,
Extra Docs and the vendor portal. Video duration is read from metadata when
the browser can; unreadable metadata never blocks. Mobile MIME variants
(empty type, octet-stream with a video extension, QuickTime) stay accepted.
The signed completion PDF keeps its 50 MB cap.
The Advanced Filters default was This week, and that value doubled as no
range, so an explicit This week searched every week. Other ranges passed
every undated WO through on the server. No range is now its own state
(nothing selected, like the prototype; clicking a selected range clears it).
It searches every week with undated rows, via includeDateless. Any selected
range narrows strictly. The pinned Unassigned queue and the WO# duplicate
lookup keep spanning every week.
React Query keeps the last successful data after a failed refetch, so an
earlier true kept the entry visible. Require a non-error query per tier,
and cover the real query path with a mocked API.
Uplift Approvals appears in a new APPROVALS group only when GET
uplifts/can-approve is true for tier 1 or tier 2, the same rule the
approve/reject API enforces. Services is linked from a DATA MANAGEMENT
group for every user. Both entries highlight on their routes.
The completed-date cell editor had no Clear control. It now offers Clear
when a date is set and the row is editable, patching completedDate to an
empty value, which the board PATCH persists as null. Completed and
Canceled rows stay locked, matching set/change.
TaskTemplateItemsField owns newItemText locally but stays mounted
across selectTemplate/startNewTemplate (which only call form.reset()),
so a typed draft survived switching templates - the base page cleared
this draft explicitly in both handleSelect and handleNew.
Remount TaskTemplateItemsField on selection change via key={selectedId}
instead of lifting the draft into the editor hook: it's transient
input-only state, not form data, so this keeps the fix local and lets
React's own remount semantics reset it. Confirmed the two new
regression tests fail without the key and pass with it.
