feat(groups): add 'Select all' to web group permission flags

[Rushaway]

What

Ports upstream issue sbpp#1436 to this fork.

On Group Management -> Web admin groups, the permission flag grid
had no bulk control, so operators had to tick every flag individually.

This adds two ghost buttons next to the Permission flags label in
the groups-list master-detail editor:

• Select all — checks every enabled flags[] checkbox
• Select none — clears them

Both refresh the live bitmask preview (SbppFoldFlags), and the folded
OR-sum round-trips through Save exactly like a manual toggle. The
buttons are gated on permission_editgroup, matching the checkboxes
themselves, and disabled checkboxes are left untouched.

Changes

• web/themes/default/page_admin_groups_list.tpl — buttons + a small
  SbppGroupsToggleAllFlags(checked) helper.
• web/tests/e2e/specs/flows/admin-groups-select-all-flags.spec.ts —
  new flow spec covering select-all, save round-trip, and select-none.

Notes

Fork's current code already handled the unsigned-bitmask concern
(sbpp#1272); this change is purely additive UI. No SourcePawn changes; no
local SP compiler, relying on CI.

🤖 Generated with Claude Code

[Rushaway]

Review — approve with fixes pushed (7ffb12f)

Reviewed the feature against upstream sbpp#1436, the whole page_admin_groups_list.tpl, SbppFoldFlags / SbppGroupsSave, web/pages/admin.groups.php, and api_groups_edit. The approach is right and the gating is correct; I found one real bug and pushed a fix plus a hardened spec.

Bug found: a bulk toggle did not arm the unsaved-changes guard

SbppGroupsToggleAllFlags set input.checked from script. A programmatic .checked = fires no change event, so the master-detail markDirty listener (form.addEventListener('change', markDirty), further down the same page tail) never ran. Consequence: press Select all, then click another group in the left rail — the selection handler saw dirty === false, called paintGroup() straight away, and the operator's bulk change vanished with no "Unsaved changes" confirm. Ticking the same 30 boxes by hand does raise the prompt, so this was a behavioural asymmetry the button introduced.

Fixed by dispatching one bubbling change from the grid after the loop (the grid's own preview listener ignores non-checkbox targets by design, so the preview is still repainted explicitly through SbppFoldFlags). The helper now also skips checkboxes already in the target state and returns early when nothing changed, so a no-op press doesn't arm the guard spuriously.

Also added aria-label="Select all permission flags" / "Select no permission flags" — "Select all" alone doesn't say which list it applies to out of context, and the label is a superset of the visible text so label-in-name still holds.

What checked out as correct in the original patch

• Gate. {if $permission_editgroup} is the same flag the checkboxes use for disabled (line 179) and the Save block uses (line 188), and AdminGroupsListView already carries it — no view change needed. Belt-and-braces input.disabled skip in the helper is the right defensive layer.
• Real buttons. <button type="button"> inside <form data-testid="group-detail"> — no accidental submit, no <a href="#">.
• Unsigned 32-bit. $all_flags strips ALL_WEB and ADMIN_OWNER (admin.groups.php L214-236), but the remainder still includes ADMIN_UNBAN_GROUP_BANS (bit 31), so select-all folds to 4278189567 — squarely the Web admin groups bitmask preview goes negative when high-bit flags toggled — JS |= returns Int32 sbpp/sourcebans-pp#1272 case. Routing the repaint through the shared SbppFoldFlags (>>> 0) is exactly right; no sign or precision bug, and SbppGroupsSave folds the same helper over the form so Save round-trips the same value.
• Escaping / Smarty. No new interpolation, no new nofilter, {if}/{/if} balanced, and the JS lives inside the existing {literal} block so the { bubbles: true } object literal is safe.
• No-JS. Not a regression: admin.groups.php never reads $_POST, so the whole master-detail editor (Save included) is already JS-only. The SSR bitmask preview is untouched.
• Scope. [data-testid="flag-grid"] is unique — page_admin_groups_add.tpl deliberately has no flag grid (it points at the List tab), and page_admin_overrides.tpl's override_flags[] is a text input, so there's no parity gap and no selector collision.

Spec hardening

The original spec asserted the badge was "not 0 bitmask" and, after the Save reload, only that the first checkbox was checked — weak enough that a partial fold or a wrong-but-nonzero value would pass. Rewritten to:

• recompute the expected bitmask in-spec from each data-flag-value using non-bitwise arithmetic (an independent oracle rather than re-running the page's own fold), assert the badge equals it exactly, and re-assert it after the Save + reload round-trip with every checkbox verified;
• guard the guard: expect(expected).toBeGreaterThanOrEqual(2 ** 31) fails loudly if a future web.json edit quietly drops the bit-31 flag and stops exercising the Web admin groups bitmask preview goes negative when high-bit flags toggled — JS |= returns Int32 sbpp/sourcebans-pp#1272 path;
• assert type="button" on both controls, and that a second Select all press is idempotent;
• add a second test seeding two groups that presses Select all, clicks the other group, and asserts the shared [data-testid="sbpp-confirm-dialog"] opens with "Unsaved changes" and that cancelling keeps the selection — the regression test for the bug above.

Both tests stay desktop-chromium-only with truncateE2eDb() per test, matching admin-groups-bitmask.spec.ts.

CI

All green on 7ffb12f: PHPUnit, Playwright + axe, Static analysis, tsc --checkJs.