`) — same diff,
also identical. So no cell gained, lost, or swapped a control.
3. **`supply_source`** — its `| ` was moved verbatim into `@case ('supply_source')`. Still
`[options]="supplySourceOptions"` with `optionValue="value"`, all four distinct options
(`factory` / `own` / `customer` / `split`). No default, no fallback, no coercion added.
`normaliseBomSupplySource()` remains the only writer of that control's initial value, in the
TS, unchanged.
4. **`customer_qty_per_unit`** — still editable only under `@if (comp.get('supply_source')?.value === 'split')`,
with the `—` otherwise. `onSave()`'s
`supplySource === 'split' ? normalise(...) : 0` line is byte-identical (git diff of the .ts
shows zero changed lines in `onSave`, `addComponent`, `duplicateComponent`, `addOperation`).
5. **`cost_per_unit`** — plain `p-inputNumber`, no new logic.
### Why the saved payload for an unchanged BOM is identical
`onSave()` never reads the DOM or the column order. It builds `data: CreateBom` from
`this.form.value`, `this.components.controls` and `this.operations.value`, then writes each line
as an **object literal with a fixed key order** (`product_id, quantity, unit_id,
waste_percentage, issue_type, issue_level, operation_seq, cost_per_unit, supply_source,
customer_qty_per_unit, sort_order, notes, substitutes`). Neither the FormArray row order nor the
FormGroup key order is derived from the template, and `git diff` on the `.ts` shows **no changed
line anywhere in the save path** (the only diff hunks are the imports block, the injects block,
the two `TxColumnLayout` field declarations, and the `docConfig.load()` line in `ngOnInit`).
`JSON.stringify` therefore emits the same bytes. (AC 6)
**Hidden-column safety, checked explicitly:** when a column is hidden its ` | ` leaves the DOM,
but the FormControl is created in `addComponent()` in TS and survives — Angular's
`FormControlName.ngOnDestroy` runs `_cleanUpControl`, which detaches the view callbacks and does
**not** reset the value. So hiding `supply_source` preserves each row's stored source; a blank
can never appear, and therefore can never be read as `'own'`. `product_id` and `quantity` are
`locked: true` and cannot be hidden at all.
### Substitutes expander (AC 5)
Unchanged markup — same ` | `, same nested ``,
same `addSubstitute`/`removeSubstitute`/`toggleSubstitutes` handlers. One binding changed:
`colspan="11"` → `[attr.colspan]="compLayout.orderedColumns().length + 1"` (visible data columns
+ the pinned actions column). At the default 10 visible columns this evaluates to 11 — the old
literal — so nothing moves until a column is actually hidden.
The nested table is deliberately excluded from every new SCSS rule: all of them are scoped with
**child** combinators off `table.tx-grid` / `table.tx-ops-grid` (`> thead > tr > th`,
`> tbody > tr.comp-row > td`), and `.sub-table` carries neither class, so it keeps its auto
layout, its own padding, and is untouched by the `overflow: hidden` used for the fixed columns.
## Resize inside the PrimeNG dialog (AC 2)
This grid lives in `` (a `p-dialog`), the exact case WP4's fix exists for.
Confirmed by construction:
- `TxColumnLayout.onResizeStart` resolves the `| ` with `(event.target as HTMLElement).closest('th')`,
never `host.querySelector(tableSelector)`. Our `.tx-resize-handle` is always a direct child of
its own ` | `, so the lookup cannot return null however the dialog is portalled.
- The width has to be able to *drive* the column, so both grids got `table-layout: fixed`
(with auto layout the px-sized inner controls win and the drag does nothing — the same trap
the stock grids hit). `defaultWidths` supplies a concrete px per column so nothing collapses.
- RTL is handled inside `TxColumnLayout` (`isRtl` → `effectiveDelta = -deltaX`), unchanged.
`defaultWidths` for the components grid reproduce the old `.col-*` SCSS widths exactly
(`product_id 230 / quantity 95 / unit_id 120 / waste_percentage 95 / issue_type 130 /
issue_level 130 / operation_seq 72 / cost_per_unit 105 / supply_source 130 /
customer_qty_per_unit 105`), so an untouched grid renders pixel-identically to before.
Operations defaults: `200 / 180 / 120 / 120 / 140`.
## Verification
| Check | Result |
|---|---|
| `npx tsc --noEmit` | clean, no output |
| `npx ng build --base-href /app/` | **green**; only pre-existing warnings (CommonJS bailouts; SCSS budget warnings for `validation-worklist`, `create-visit`, `dept-worklist` — none in Production) |
| `formControlName` multiset HEAD vs HEAD~ | identical |
| control-element multiset | identical |
| i18n keys required | all 17 already present in `ar.json` + `en.json`; neither file touched |
| `chown moonui2:moonui2` | applied to all 3 edited files |
| Prior commits `c03c989d1`, `1574011b8` | intact, not reverted; WP5 sits on top of them |
## Concerns / notes for the orchestrator
1. **Runtime click-through not performed.** The build is green and the mechanism is a verbatim
port of a shipped, working pattern, but AC 1/2/3/5 are behavioural and I did not deploy
(per brief). They need one pass in the browser after the orchestrator deploys: reorder a
components header, resize a column edge in RTL, hide `waste_percentage` from
Settings → Documents → BOM, open a substitutes expander, save an untouched BOM.
2. **Operations grid width nuance (cosmetic).** The pre-existing rule
`.editable-table table { min-width: 1080px }` still applies. The components grid's default
widths sum to 1332px, so there its persisted px widths are absolute — the owner's screen
behaves exactly right. The operations grid's defaults sum to ~880px, under the 1080 floor, so
until the columns total more than 1080 the browser distributes the slack proportionally and a
saved width is honoured *relatively* rather than to the pixel. Resizing still visibly works
and the order/hide behaviour is exact. I left `min-width` alone rather than drop it, because
removing it would leave dead whitespace to the right of the ops table in a wide dialog.
3. **The ops grid's inline control widths are now overridden to `width: 100% !important;
min-width: 0 !important`** (scoped to `table.tx-ops-grid > tbody > tr > td`), mirroring what
the reference grids do — otherwise a resized ops column would show a 100px input in a 300px
cell. The `[style]` attributes in the markup were left verbatim; only CSS overrides them. This
is a small, intentional visual change on the operations tab: its inputs now fill their column.
4. **No CHANGELOG entry added.** The FE repo has no `docs/moonstack/CHANGELOG.md` (it lives in the
BE repo), and the two preceding commits on this branch (`c03c989d1`, `1574011b8`) likewise
carry none. The release note for this capability belongs to whoever cuts the release; flagging
it so it is not lost.
|