@smartbit4all/ng-client 7.0.0 → 7.0.2

package/MIGRATION-7.0.md

@@ -24,6 +24,28 @@ Measured across the 12 hosts in scope on 2026-07-28, so you know what to expect:
 `Smart*Module` imports per host, 1–20 string-keyed providers, `.cssClass =` on an action
 entry at 72 sites in 7 hosts, `this.uiActionModels =` in 5 hosts, `UseUiAction` in 6.
 
+**How big is this for *your* host?** Codemod dry-run counts for the five hosts in scope
+(2026-07-30). The first column the codemod applies for you; the second is the list it refuses to
+guess at, which is the real work:
+
+| host | mechanical | needs a decision |
+|---|---|---|
+| p014 *(done — the reference)* | 121 | ~100 |
+| p043 | 121 | 96 |
+| app-tournament | 86 | 34 |
+| app-finance-ai | 62 | 18 |
+| app-fitnessmirror | 61 | 14 |
+
+The spread is not about host size — all five are within 274–841 source files — but about which
+APIs they happened to use. p043 is p014's twin and repeats its whole decision list; the other
+three have **no `.cssClass =` sites at all**, and their lists are dominated by `smart-modules`,
+the 3-line `moment-adapter` swap, and `zone-js` (which is **optional** — dropping zone.js is not
+required). Take a small one first if you want the process shaken out cheaply.
+
+**One number that is now zero: the library install.** p014 was migrated against an unpublished
+build and paid for it in hand-copies, tarballs and `ERESOLVE`. 7.0.0 is on npm, so a host today
+writes `"@smartbit4all/ng-client": "^7.0.0"` and runs `npm i`. None of that thread applies to you.
+
 ## Migrating a host: the codemod
 
 It lives in the `platform-angular2` repository — `tools/ng-client-7-codemod/` — and you run
@@ -281,7 +303,7 @@ The replacements, for what the hosts actually import:
 | `DialogService` | `MatDialog`. There is **no `header` option**: put `<h2 mat-dialog-title>` in the component |
 | `DynamicDialogRef` | `MatDialogRef` — `afterClosed()` replaces `onClose` |
 | `DynamicDialogConfig` | `MAT_DIALOG_DATA` for the data, `MatDialogRef` for the rest |
-| `PrimeNGConfig` | **delete it.** `setTranslation()` of day and month names is already covered by `MAT_DATE_LOCALE` + `SmartDateFnsAdapter` — the date-fns `hu` locale carries both the names and the Monday week start (see [Dates](#dates-moment--date-fns-and-the-timezone-contract-phase-38)) |
+| `PrimeNGConfig` | **delete it.** `setTranslation()` of day and month names is already covered by `MAT_DATE_LOCALE` + `SmartDateFnsAdapter` — the date-fns `hu` locale carries both the names and the Monday week start. **Confirmed on a running host** (p014, 2026-07-29, against a live backend): the datepicker opens on `2026. JÚL.` with `H K Sz Cs P Sz V`, i.e. Hungarian and Monday-first, exactly what the deleted `firstDayOfWeek: 1` + `dayNamesMin` used to buy. See [Dates](#dates-moment--date-fns-and-the-timezone-contract-phase-38) |
 | `MessageService` | `MatSnackBar` |
 | `ConfirmationService` | the library's `UiActionConfirmDialogService` |
 | `MenuItem` | the library's `UiActionModel`, or a plain `<mat-menu>` item |
@@ -675,6 +697,36 @@ functioning in TypeScript 7.0"). Two ways out:
   strict, so nothing changed. If your host was not strict, the `ng update`
   migration adds `"strictTemplates": false` to keep it that way.
 
+### The third-party packages the library does *not* own (Phase 2)
+
+Every hop section above lists the library's own peer dependencies. Your host almost
+certainly carries packages the library never sees, and those have their own Angular
+peer ceilings — each one can stop a hop dead. Measured on p014 (2026-07-29), which
+went 17 → 22 with the library still installed:
+
+- **`@swimlane/ngx-charts` needs a ladder of its own**: `20 → 22 → 22 → 23 → 24 → 25`
+  across the five hops. At the Angular 21 hop it peer-caps the CDK, and because
+  `ng update`'s own `npm` step is **not** `--force`d (even when you passed `--force`
+  to `ng update`), the install fails there while `package.json` has already been
+  written correctly. The fix both times was to redo the install by hand, not to
+  re-run the update. `p043` and `app-tournament` both carry ngx-charts `^20.4.1`
+  and will hit this; `app-fitnessmirror` and `app-finance-ai` do not.
+- **`ngx-extended-pdf-viewer` moved to `.mjs`** at the 21 hop and stopped shipping
+  `pdf-3.10.560-es5.min.js`. If your `angular.json` names that file under `scripts`
+  or `assets`, the build fails on a missing file rather than on anything Angular.
+- **`karma` and `@angular/build`**: `@angular/build` 22 declares
+  `peerOptional karma@^6.4.0`. A host pinned to `~6.3.0` gets a wall of
+  `npm warn ERESOLVE overriding peer dependency` — **warnings, exit code 0**, so it
+  is not blocking. It becomes a *hard* install failure the moment `@angular/build`
+  is a direct devDependency rather than a nested one, which is what happened in the
+  library workspace. `karma@~6.4.4` clears it.
+
+The general shape: **`ng update` needs `--force` at every hop** (the library and
+PrimeNG both peer-cap at Angular 17 until the very end), and `--force` covers the
+Angular schematics but not the `npm install` they trigger. When that install fails,
+check `package.json` before assuming the hop did nothing — it is usually already
+correct.
+
 ### Housekeeping (Phase 3.0)
 
 #### The dialog's built-in Ok/Cancel buttons are gone
@@ -946,7 +998,7 @@ record behind it is ADR-0005 in the `platform-angular2` working copy.
 |---|---|
 | `this.useQueryLists = true;` | The property is gone. Its setter rejected anything but `true`, so every branch behind it was unreachable — this was a no-op that logged a warning. |
 | `[parentSmartComponent]="this"` on `<smart-component-layout>` | The input is gone; the layout injects the client. `strictTemplates` reports leftovers as **NG8002**. |
-| `this.addForm(id, smartForm, formComponent)` | Gone. A `<smartform>` that displays the client's model registers itself. Two call styles existed; where the return value was used, it was the `smartForm` argument — use that. |
+| `this.addForm(id, smartForm, formComponent)` | Gone. A `<smartform>` that displays the client's model registers itself. Two call styles existed; where the return value was used, it was the `smartForm` argument — use that (which makes those lines self-assignments, so they just go). **Delete the scaffolding around the call too**: in p014 the 21 calls left behind 9 empty `if (this.formX) { }` guards and 3 `new Promise((r) => setTimeout(r, 500)).then(() => { })` wrappers whose only content had been the registration. The compiler does not complain about either. |
 | `this.widgets.delete(…)` / `this.widgets.set(…)` / `this.formWidgets…` | The maps are gone (they were only written by the legacy path and read by nothing). |
 | the 4th argument of `addGrid(smartGrid, options, useAsDefaultGrid, componentReference)` | `addGrid()` keeps the options and the default-grid role; the services it used to copy into the grid model (`uiActionService`, `uiActionDescriptorService`, `serviceToUse`, `viewContextService`) are the grid's own DI job now. |
 | `override getSmartXxxQL()` (e.g. the two `throw new Error` overrides in a `FilterPageService`) | The eight abstract getters are gone from `SmartComponentApiClient`. |
@@ -1304,6 +1356,20 @@ npm install @angular/material-date-fns-adapter@^22.0.0 date-fns@^4
 If your own code imports moment for something unrelated, keep it — nothing here forces its
 removal, only the adapter's.
 
+**But check *how* it imports moment.** `import * as moment from 'moment'` stops being callable
+under `moduleResolution: "bundler"`, which the Angular 20 hop turns on: moment is an
+`export =` module, and a namespace import of one is not a callable value. The symptom is
+**`TS2349: This expression is not callable`** on every `moment(...)` call, and it arrives at the
+Angular 20 hop rather than at this step. Two ways out:
+
+- `esModuleInterop: true` plus `import moment from 'moment'` — correct, but it changes how every
+  CommonJS import in the app resolves, which is a wide blast radius for one file.
+- Replace the calls. Worth checking first how much moment is actually doing: in p014 the three
+  surviving calls were `moment(new Date()).toDate()` (i.e. `new Date()`) and two moments compared
+  with `<=` (`Date` compares through `valueOf()` identically). All three became plain `Date`s, and
+  **nothing under `src/` imported moment any more** — so the `npm uninstall … moment` above was
+  right for that host after all, just for a reason nobody predicted.
+
 #### 2. Leave your `MAT_DATE_LOCALE` provider alone
 
 If you have this line — p014 and p043 both do — **keep it as it is**:
@@ -1399,6 +1465,21 @@ not less: `2026.07.28.`, `2026.07.28`, `2026.7.28.`, `2026.7.28` and ISO-8601 ar
 The month picker is unchanged (`07/2026`), and grid/table date columns were never affected —
 `SmartDatePipe` and friends extend Angular's own `DatePipe` and never used moment.
 
+**Verified on a running host** (p014, 2026-07-29, live backend, `Europe/Budapest`). Picking
+2026-07-15 in a `Kezdődátum` field gave, end to end:
+
+| | |
+|---|---|
+| input | `2026.07.15.` |
+| widget value | `instanceof Date` — not a moment, not a luxon `DateTime` |
+| `JSON.stringify` of that value | `"2026-07-14T22:00:00.000Z"` |
+| `getHours()` | `0` |
+
+That third row is the whole contract in one line: the value's `toJSON()` **is** the wire format
+(see §4), and it is zulu, while the screen and `getHours()` stay local — 15 July 00:00 CEST is
+14 July 22:00 UTC. A grid column on the same host rendered `2026.04.20 0:03`, confirming the
+untouched pipe path.
+
 ### HTTP: the library no longer forces the XHR backend (Phase 3.8)
 
 `provideSmartNgClient()` used to pass `withXhr()` to its internal `provideHttpClient()` call,