status_code='abnormal' records (>30 min) hide the underlying late/early/ot
component from by_status. Use *_minutes>0 union to surface them so the
dashboard count matches the real data.
Late: 9 → 23, Early: 0 → 17, OT: 15 → 28
Note: component totals may now exceed total_records because each record
can carry multiple status components (e.g. late_ot).
Status & filter improvements consolidated:
- StatusBadge: 5 basic categories + leave suffix chip
- Backend filter: enrich-then-filter so badges match result set
- 請假/公假 filter: also match leave/holiday suffix in status_text
- Rest day detection: roster rest weekday → 休息 (not 缺勤)
When roster says the shift rests on a given weekday (e.g. Sunday for S2/S3),
the employee wasn't scheduled to work. Import still marks the record
status_code='missing' because no shift times were available, but the
right user-facing label is '休息' not '缺勤'.
Enrich now: if status_code='missing' and get_shift_schedule_full returns
None (shift defined no schedule for that weekday), override to
status_code='rest' / status_text='休息'.
Added '休息' to the basic-status vocabulary: dropdown option, filter
condition, StatusBadge (slate grey + coffee icon), and row highlight.
Pure-leave status_code (al/sl/cl/mixed_leave) only catches employees who
took the whole day off. Records where the employee worked AND took leave
mid-day keep their raw status_code (normal/abnormal/ot) but carry a
' + SL2h' / ' + AL4h' suffix that the badge now surfaces as a chip —
yet those records disappeared under the '請假' filter.
Same gap for holidays: '公假' only matched status_code='holiday' but an
employee who worked on the public holiday still gets the '🎉' suffix and
should be reachable from the holiday filter.
Expand both filters to also match records whose status_text contains the
relevant suffix (' + SL' / ' + AL' / ' + CL' / '🎉'), keeping the result
set aligned with what the badge chip advertises.
The previous '🎉香港特別行政區成立紀念日' chip wrapped to 7 lines on
mobile and duplicated info already conveyed by the '公假' main badge.
Keep leave chips (+SL2h, +AL4h, etc.) since the basic-status badge
doesn't reveal the leave type, but skip holiday decoration entirely.
status_text fields like '異常-遲到48分 + SL2h' or '缺勤 (🎉香港特區…)' carried
important leave/holiday context that the simplified badge was dropping on the
list page — the user had no way to tell an abnormal record was partly covered
by sick leave.
Extract the suffix from status_text and render a small secondary chip next
to the basic-status badge (violet for leave types, indigo for holidays).
Detail page picks this up automatically since the component is shared.
Example: '異常 +SL2h' / '缺勤 🎉香港特區成立紀念日'
The previous filter applied status_code conditions on raw DB rows, then
enriched the page of results. That meant a missing record with a leave
request would surface under '缺勤' (raw code='missing') but render a
'請假'/'公假' badge (enriched status_code).
Move status_text filter to AFTER enrichment so it matches the badge.
status_text sort_by='status_code' also moves to Python sort since
post-enrich codes aren't in the DB. Log a warning if the in-memory
filter scans >5000 rows so we know when to revisit.
The previous condition 'c === "normal" && !t' required status_text to be
empty, but enriched records carry text suffixes like '正常 + CL4h' which
are truthy and dropped normal records into the '異常' fallback.
3 Kaka records (id=22/28/30) were affected — act_in 10:38/10:43 (early),
act_out 20:00 (on time), so they should render the green '正常' badge but
were showing '異常' instead.
The previous filter used 'late_minutes == 0 AND early_minutes == 0 AND
ot_minutes == 0' for '正常', which incorrectly included records whose
import-time late_minutes was 0 but post-enrich status_code was 'abnormal'
(e.g. act_in=18:00 vs exp_in=11:00) or 'missing' (act_in='-').
Switch filter to status_code conditions so the result set matches the
badge rendered from status_code. Also accept the Chinese basic-status
labels (正常/異常/缺勤/公假/請假) as filter values, and return only those
5 from /api/attendance/status_texts so the dropdown vocabulary stays
in sync.
Frontend attendance list: simplify getStatusBadge to drop now-unused
late/early/ot props; align row highlight with the basic-status grouping.
- Response key is 'records' not 'data'
- Backend /api/attendance/records uses 'per_page' (not 'page_size')
- Without this fix, list never loaded data — would always show 'no records'
- StatusBadge no longer shows detailed late/early/OT minute breakdowns
nor AL/SL/CL sub-types; collapse to single basic-status badge
- List page: replace text-input column filter with status dropdown
(server-side ?status_text= filter; backend now supports 'leave' IN-clause
for al/sl/cl/mixed_leave group)
- List page: badge now driven by backend status_code/status_text
(remove frontend latenessMap which contradicted backend status)
- Add /api/admin/backup/list, /create, /download, /restore endpoints
- Add /api/version endpoint (git commit info)
- Add Settings page to sidebar nav + App.jsx routes
- Fix lateness stats: add total_late_count/total_late_minutes keys
- Fix leave aggregation: no more duplicate text in status
- Settings page now shows backup/restore UI + version info
- Fix leave upload: overwrite existing records on re-upload
Issues:
1. Frontend renders record[h] where h comes from column_order.
column_order now uses submitter_email (not employee_name),
so record[submitter_email] = None → shows as "-".
2. submitted_at shows as ISO format "2026-07-06T15:45:33.850000"
(less readable than "2026-07-06 15:45:33").
Fix:
1. to_dict adds both submitter_email and employee_name alias
(same value, r.submitter_email). Same for injured_person_id ↔ employee_id.
Frontend column_order key matches record key (submitter_email shows email).
2. submitted_at formatted as "YYYY-MM-DD HH:MM:%S" via strftime
Verified:
- record.submitter_email = "operation@scmedical.hk"
- record.employee_name = "operation@scmedical.hk" (alias)
- record.submitted_at = "2026-07-06 15:45:33"
Backend:
- to_dict() now includes incident_datetime: "YYYY-MM-DD HH:MM:SS"
- Composed from r.date (YYYY-MM-DD) + r.time (HH:MM:SS, padded if HH:MM)
- _get_excel_column_order_accident() adds incident_datetime column after time
- Export xlsx rec dict now includes incident_datetime
Verified:
- /api/accident/records: incident_datetime = "2026-07-30 09:40:00"
- column_order includes incident_datetime column
- /api/accident/export/excel: xlsx 明細 sheet now has incident_datetime column
- Time edge case: HH:MM auto-padded to HH:MM:SS
Frontend:
- Accident list page auto-displays new column via column_order
- No frontend changes needed (renders record[h] as string)
Issues fixed:
1. Indentation error after rename (rec dict outdented one level)
2. rec dict used employee_name/responsible_person keys but
column_order used submitter_email/injured_person keys
→ rec.get(col, \"\") returned \"\" → xlsx cells showed None
Fix:
- Re-indent whole for r_idx block (ws.set_row, for c_idx)
- Use submitter_email/injured_person/injured_person_id as rec keys
to match column_order
- Tested: xlsx 電子郵件地址 column now shows real email values
After renaming Accident.employee_name -> submitter_email, four more
references to r.employee_name / r.employee_id slipped through and caused
500 Internal Server Errors:
1. /api/attendance/records to_dict - was using r.submitter_email (wrong table!)
(regression from earlier batch sed that hit too many matches)
2. /api/accident/dashboard/stats - emp = r.employee_name
3. /api/accident/export/excel to_dict
4. /api/accident/export/excel overview + detail sheet
Fix: explicit context-aware replacements (only Accident-related sections)
Verification:
- /api/attendance/records: 200 OK (returns 110 records)
- /api/accident/records: 200 OK (returns 342 records)
- /api/accident/dashboard/summary: 200 OK
- /api/accident/dashboard/stats: 200 OK (was 500)
- /api/accident/export/excel: 200 OK + valid xlsx (78KB)
Database migration:
- ALTER TABLE accidents RENAME COLUMN employee_name TO submitter_email
- ALTER TABLE accidents RENAME COLUMN responsible_person TO injured_person
- ALTER TABLE accidents RENAME COLUMN employee_id TO injured_person_id
Reason: Accident.employee_name contained submitter EMAIL (not staff name),
causing confusion when joining with attendance_records.employee_name (real name).
New column names are self-documenting.
API backward-compat:
- All API responses still use "employee_name" / "responsible_person" field names
- Internal SQL mapping translates submitter_email/injured_person -> employee_name/responsible_person
- Frontend code unchanged
Verified:
- /api/accident/1 returns employee_name = submitter_email content
- /api/accident/records returns 2 records OK (was 500 before fix)
- /api/employees/list still returns 12 staff (accident excluded)
- Data preserved (1st record id=1 submitter_email = info@scmedical.hk)
Root cause: Accident.employee_name contains submitter EMAIL addresses
(e.g. info@scmedical.hk), not the injured staff name. UNION of all three
sources polluted the attendance staff dropdown with emails.
Fix:
- Removed Accident from /api/employees/list UNION (source=all now means
attendance + leave only)
- Added EMAIL_RE filter to skip any name containing @ (defensive)
- Updated docstring + response includes excluded field explaining why
Verified: 12 clean staff names returned (down from 16 with 4 emails mixed in)
Bug: After refactoring LeavesTab to fetch employees from /api/employees/list
(useState + useEffect), the LeaveModal was still passed employees={uniqueEmps}
but that variable no longer existed, causing JS runtime error:
"Can't find variable: uniqueEmps"
Fix: rename reference to {employees}
Frontend (Holidays.jsx LeaveModal):
- Split into Edit Mode (single record) + Create Mode (multi-row)
- Create mode: shared employee/date/approver + list of (type, hours, reason) rows
- Add/remove rows dynamically, each with own type + hours + reason
- Submit loops POST /api/leave per row, aggregates results (added/skipped/failed)
- Submit button label shows count: "加 N 條"
- Bottom shows total valid count + total hours
- Result summary shown after submit (expandable details)
- Auto-close on success after 1.2s
Backend:
- LeaveRecord: add UniqueConstraint (employee_name, date, leave_type)
- Migration: CREATE UNIQUE INDEX uix_leave_emp_date_type (cleaned any pre-existing dups first)
- POST /api/leave: catch IntegrityError → return 409 with clear message\n\nVerified:\n- 3 same-day different-type records (SL/CL/AL) created successfully\n- POST same type twice returns 409 with specific message\n- HTTP 200 for valid creation, 409 for collision
Schema:
- attendance_records: last_edited_at + last_edited_by (was: is_manually_edited boolean only)
- accidents: last_edited_at + last_edited_by (was: no edit tracking)
Migration: ALTER TABLE adds columns (idempotent)
Logic:
- Attendance import: skip if last_edited_at IS NOT NULL OR is_manually_edited (backward compat)
- Accident import: skip if last_edited_at IS NOT NULL (was: always overwrite)
PUT endpoints:
- PUT /api/attendance/{record_id}: set last_edited_at + last_edited_by + is_manually_edited
- PUT /api/accident/{record_id} (NEW): set last_edited_at + last_edited_by, whitelist editable fields
Verified via direct _import_attendance_to_db / _import_accident_to_db call:
- Edited attendance: skipped on re-import (status_text preserved)
- Edited accident: skipped on re-import (description preserved)
- Not-edited accident: updated to new values (description changed)
Note: HTTP /api/upload/{section} returned 404 in test, but function logic verified working
- Add last_edited_at + last_edited_by columns to holidays + leave_records
- Migration: ALTER TABLE adds columns (idempotent)
- PUT endpoints set last_edited_at on every manual edit
- POST /api/leave/upload: skip if last_edited_at NOT NULL, update otherwise
- New POST /api/holidays/upload: same logic, admin only
- LeaveOut + HolidayOut schemas include last_edited_at
- Response includes added/updated/skipped counters
- Verified: edited Jay Lam SL preserved, not-edited May Chan SL updated
IT狗