Files
core/docs/cart.md
T

15 KiB

Cart Admin Visibility

Modules\Core\Cart\Filament\Resources\CartResource gives staff read-only visibility into customer/user carts in the Filament admin panel. Lunar itself ships no cart admin view at all — no Filament resource for Cart/CartLine exists anywhere in lunarphp/lunar or lunarphp/core — this is a from-scratch addition, not an extension of something Lunar half-built. See docs/lunar.md's "Cart and Checkout" section for the underlying Lunar cart mechanics this resource reads from.


Scope: only carts with a known customer or user

CartResource::getEloquentQuery() filters to Cart::whereNotNull('user_id')->orWhereNotNull('customer_id') — an anonymous guest's session cart is excluded entirely.

This was a deliberate call, not an oversight: an anonymous cart carries no identity a staff member could act on — no name, no email, nothing to follow up with — so listing every guest session cart would be noise, not a real admin capability. This does not mirror Shopify's admin (Shopify has no "all carts" view at all — only "Abandoned checkouts," gated on a shopper reaching checkout and entering contact info, a later/narrower stage than Lunar's Cart). Lunar's own Cart model already gets user_id/customer_id set the moment a shopper is authenticated (via Lunar\Listeners\CartSessionAuthListener on login), with no checkout step required — so scoping to "identifiable" here is broader than Shopify's equivalent, not a copy of it.


Four states, not two — and not Cart::completed_at

Lunar\Models\Cart::completed_at is declared and cast ('completed_at' => 'datetime') but never actually written anywhere in Lunar core — grep vendor/lunarphp/core/src for it; the only hits are the property declaration and the cast. It is not a real signal. Cart has no status column at all — every state below is derived from relations/timestamps, not a single field.

Cart::scopeActive() (Lunar's own "not yet converted to an order" scope) actually mixes two distinct states together: no order ever started, vs. a draft order exists (placed_at IS NULL) but was never placed — checkout was started, not finished. Those are different purchase-intent signals (see "Abandoned Cart vs Abandoned Checkout" below) and different reachability (checkout usually captures an email even for a guest), so ListCarts::getTabs() splits them into four tabs instead of scopeActive()'s two-state split:

  • Ongoing — scopeActive() and recent updated_at (within abandonedCutoff()). Default active tab on page load.
  • Abandoned Cart — whereDoesntHave('orders') and stale updated_at.
  • Abandoned Checkout — has an order with placed_at IS NULL, and stale updated_at.
  • Completed — has an order with placed_at IS NOT NULL.
// Ongoing
$query->active()->where('updated_at', '>', CartResource::abandonedCutoff());

// Abandoned Cart
$query->whereDoesntHave('orders')->where('updated_at', '<=', CartResource::abandonedCutoff());

// Abandoned Checkout
$query->whereHas('orders', fn ($q) => $q->whereNull('placed_at'))
    ->where('updated_at', '<=', CartResource::abandonedCutoff());

// Completed
$query->whereHas('orders', fn ($q) => $q->whereNotNull('placed_at'));

There is deliberately no "All" tab. Every row shown is always scoped to one of the four states above — the list never runs an unfiltered Cart::query()->get() over the whole (potentially large) table.

Abandoned Cart vs Abandoned Checkout — why they're not one bucket

Different purchase intent, different reachability, and different recovery strategy — see docs/recovery-strategies.md for the full marketing-strategy discussion. In short:

  • Abandoned Cart (no order started) is a weak intent signal — often window-shopping, not a near-purchase. Frequently unreachable (no email/identity at all for a true guest). Recovery leans on on-site retargeting and ad remarketing rather than email.
  • Abandoned Checkout (draft order, never placed) is a strong intent signal — the shopper committed to buying and something blocked completion. Checkout typically captures contact info even for a guest, so this state is usually reachable. This is the state the researched 1h/24h/72h recovery-email cadence targets specifically.

Modules\Core\Cart\Events\CartAbandoned and Modules\Core\Checkout\Events\CheckoutAbandoned mirror this same split (see "Events" below) rather than one combined event.


Why this scales fine at a large cart count

Two things keep this cheap regardless of how many carts exist (10,000+):

  • The list is always paginated. Filament applies LIMIT/OFFSET to whichever tab's query is active — a page only ever fetches one page's worth of rows, never the whole table, "All" tab or not (and there is no "All" tab — see above).
  • No per-row queries. lines_count/lines_sum_quantity use Filament's built-in ->counts('lines')/->sum('lines', 'quantity'), which fold into the same query as the rest of the list (one LEFT JOIN-based aggregate, not N separate lookups). There's no per-record getStateUsing() closure anywhere in this table doing its own query — that's the pattern to avoid if a future column needs derived data (see Modules\Core\Catalog\ Services\ProductIndexer for the general "compute once at index time / one aggregate query, never per-row" principle this project follows elsewhere).

The one thing that does scan more rows as the cart count grows is CartResource::getNavigationBadge() (see below) — but it's a COUNT(*), not a fetch, and runs once per admin page load, not once per cart row.


Navigation badge — abandoned cart count

public static function getNavigationBadge(): ?string
{
    return (string) static::getEloquentQuery()->active()->count();
}

Shows the number of abandoned carts (not all carts — a converted cart isn't something a staff member needs to keep noticing) next to "Carts" in the sidebar. ->count() compiles to a single SELECT COUNT(*) ... — confirmed via query log — no rows are ever loaded just to render the badge.


The view page runs the cart's full calculate pipeline — once

ViewCart::resolveRecord() calls $cart->calculate() before rendering, since CartLine's computed properties (unitPrice, total, etc.) and Cart's own totals (subTotal, total, ...) are plain public properties populated as a side effect of that pipeline — never persisted, so a plain Eloquent-fetched Cart has them all null/unset (see docs/lunar.md Gotchas). This only runs on the single-record view page, not per row in the list table — running the full 5-step pipeline for every row of a paginated list would be needless cost for data the list doesn't display.


Not built: staff editing a cart

The resource is deliberately read-only (canCreate() returns false, no edit page registered). A cart is owned by the storefront's own add/update/remove flow (CartSession/Cart::add()/etc.) — hand-editing cart contents from the admin panel isn't a supported use case here.


CartService — the storefront-facing API

Modules\Core\Cart\Services\CartService mirrors Modules\Core\Catalog\Services\ ProductService/CollectionService's shape — one boboko-owned API a storefront calls, so Lunar's own CartSession/Cart stay an implementation detail rather than something a consuming app depends on directly.

  • current() / currentOrCreate() — the latter force-creates a cart (CartSession::manager()), the former doesn't (CartSession::current(), returns null for a fresh visitor — see docs/lunar.md's Cart gotchas).
  • addLine() / updateLine() / removeLine() / clear() — thin wrappers over Cart::add()/updateLine()/remove()/clear(). No boboko-owned exception types wrap Lunar's own cart exceptions (InvalidCartLineQuantityException, CartLineIdMismatchException, etc.) — they propagate as-is; a wrapper would add indirection with identical semantics.
  • applyCoupon() / removeCoupon() — sets/clears Cart::coupon_code (there's no dedicated Lunar action for this, unlike add/update/remove). applyCoupon() validates via Discounts::validateCoupon() first and throws Modules\Core\Cart\Exceptions\ InvalidCouponException on a bad code — CouponString's cast only normalizes casing, it doesn't validate anything, so setting coupon_code directly would silently accept a bogus code and just not discount anything once calculated.
  • saveForLater() / moveToCart() / activeLines() / savedLines() — see "Save for later" below.

Every mutating method returns the recalculated Cart (matching Lunar's own Cart::add() etc., which already return $this after refresh()->recalculate()) and dispatches a matching domain event.

Events — Lunar dispatches none of its own

Lunar dispatches zero cart events — no "item added," no "cart created" (see docs/lunar.md's Cart gotchas). CartService fills that gap with its own, dispatched after the underlying Lunar operation completes:

CartLineAdded, CartLineUpdated, CartLineRemoved, CartCleared, CartCouponApplied, CartCouponRemoved, CartLineSaved, CartLineMovedToCart — all under Modules\Core\Cart\Events. CartAbandoned/CheckoutAbandoned live under Modules\Core\Recovery\Events instead, not Cart/Checkout — see "Abandonment detection" below for why.

None of these currently have a listener. They're dispatched-but-unconsumed by design — built so something downstream (reindexing, notifications, a future read-side reporting service) has a hook to attach to, not because a concrete consumer exists today. This was a deliberate decision, not an oversight — see the "don't build speculative infrastructure" calls made elsewhere in this project (e.g. not wrapping Lunar's cart exceptions).

Why not wired to Spatie's Activity Log: Cart/CartLine already use Lunar's own LogsActivity trait (Spatie's package, Lunar's defaults) — confirmed from source, this logs model saves/deletes automatically, independent of actor. Modules\Core\Logging\ ActivityLogService (this project's own wrapper, used by e.g. LogTranslationActivity) is hardcoded to the staff guard — correctly scoped for staff-driven writes (Filament admin actions), but wrong for customer-driven cart activity, which would resolve causedBy() to null every time. Both ActivityLogService and Cart/CartLine's native LogsActivity write to the same log_name = 'lunar' / activity_log table, with no built-in separation beyond reading causer_type per row — a real limitation worth knowing about, but not one this project is fixing by giving Cart a distinct log_name, since every other Lunar model logs to 'lunar' too and a Cart-only carve-out would just be inconsistent. The intended fix, if this becomes a real need, is a read-side service that queries activity_log and classifies by causer_type/log_name — not touching every write site.

Save for later

A CartLine can be moved out of the purchasable cart without being deleted — flagged via meta.saved_for_later, not a new column (matches the free-form-JSON pattern already used elsewhere, e.g. ProductOptionValue::meta). Modules\Core\Cart\Pipelines\ ZeroSavedForLaterPrice (registered in config('lunar.cart.pipelines.cart_lines'), after the stock GetUnitPrice) zeroes unitPrice/unitPriceInclTax for flagged lines before Lunar's own CalculateLines pipeline step sums the cart — CalculateLines sums every CartLine unconditionally with no meta-based exclusion of its own, so zeroing the price upstream is what makes Cart::subTotal/total naturally correct without a second pass or callers needing a different totals accessor.

Lunar\Actions\Carts\UpdateCartLine replaces the whole meta column on write (plain update(['meta' => $meta]), not a merge) — saveForLater()/moveToCart() read the line's existing meta and merge in the flag change before calling Cart::updateLine(), or an unrelated meta key set by something else would be silently wiped.

Coupons

See CartService::applyCoupon()/removeCoupon() above. Lunar\Base\Casts\CouponString just upper-cases the code; Lunar\Managers\DiscountManager::validateCoupon() (via the Discounts facade) is the actual check — does a matching Discount (type AmountOff or BuyXGetY) exist, active(), with max_uses not exhausted.


Abandonment detection

"Abandoned" is a derived state (Cart::updated_at older than config('core.cart.abandoned_after'), default 1 hour) — nothing transitions a cart into it via a normal Eloquent write, so there's no model-event hook to dispatch from directly. Modules\Core\Cart\Commands\DetectAbandonedCarts (registered on an hourly schedule by Modules\Core\Providers\CartServiceProvider) is the only place that moment gets detected: it queries the same two branches ListCarts::getTabs() uses (no order at all vs. draft order never placed) and dispatches Modules\Core\Recovery\Events\CartAbandoned/CheckoutAbandoned for anything currently stale.

DetectAbandonedCarts only dispatches — it never writes to Cart/Order at all. An earlier version recorded an "already notified" marker on Cart::meta/Order::meta to avoid refiring the same event every run, but that ->save() call bumped Cart::updated_at as an Eloquent side effect — since updated_at is also the field abandonment staleness is computed from, the write un-staled the very cart it had just marked abandoned: confirmed live, a cart that correctly fired CartAbandoned showed back up as "Ongoing," not "Abandoned Cart," on the very next tab-count check.

The fix wasn't to write the marker more carefully — it was to stop Cart/Checkout from having any way to write abandonment state at all. Deduplication ("has this cart already been notified") is deliberately not this command's job; it belongs to Recovery (not yet built — see docs/recovery-strategies.md), which will own its own tracking table, keeping Cart/Order permanently free of abandonment-related columns or meta keys.

Current tradeoff, accepted deliberately: until Recovery exists, every cart still matching the "abandoned" query refires its event on every hourly run — there is no dedup at all right now. That's fine today only because nothing consumes these events yet (see "Events" above); it would need addressing before anything real listens for them.


Recovery Sequences — design only, not built

See docs/recovery-strategies.md — a full marketing-strategy discussion and a first-pass feature design for an admin-configurable sequence of "touches" (delay + optional discount + label) per abandonment type. Explicitly parked as an open design question, not scoped for implementation yet — whether this belongs under Cart, a new Recovery/Marketing concern, and how far the touch model needs to flex (channel choice, value-based branching, segment targeting) are all still undecided.