feat: feed the record and arbitrary data through to Brick::toHtml()

[awcodes]

Closes the gap raised in #56.

Brick::toHtml() has always taken a second ?array $data parameter, and nothing has ever passed anything to it — MasonRenderer hardcoded data: [], both iframe renderers omitted the argument. The hook was in the signature, the generator stub and the docs, but unreachable. This wires it up.

What works now

Front end — a third argument on the helper, or data() on the renderer (both accept a closure on the renderer side):

@mason($post->content, BrickCollection::make(), ['record' => $post])
{!! mason($post->content, BrickCollection::make(), ['record' => $post])->toHtml() !!}

The @mason directive forwards its whole expression to the helper, so it picked the argument up for free, named arguments included.

Infolists — MasonEntry defaults to ['record' => $this->getRecord()], so a brick reading $data['record'] works with no wiring. data() overrides it.

public static function toHtml(array $config, ?array $data = null): ?string
{
    $record = $data['record'] ?? null;
    // ...
}

The entry crosses a request boundary

MasonEntry does not render in the Livewire request — it renders in an iframe fed by a POST to MasonController, so its data has to survive the trip. DataPayload reduces Eloquent models to a class-and-key marker, encrypts the JSON with Crypt, and re-resolves the model on arrival:

• model attributes never go over the wire;
• a payload we did not sign is treated as absent rather than fatal, so a forged one renders no data instead of a chosen record (there's a test for exactly this);
• a record deleted between render and request arrives as null.

Values that cannot survive the round trip are rejected at encode time rather than silently dropped — json_encode() turns a closure into {} instead of failing, so anything that is an object but neither JsonSerializable nor Arrayable throws, as does a resource.

The editor preview is deliberately untouched

A record-dependent brick renders its null branch in the editor for now, and the docs tell brick authors to write that branch on purpose.

Plumbing is not the blocker — the preview route could reuse DataPayload in about five lines. What it should show is the open question:

1. Which record? The last-saved row would contradict unsaved sibling fields in the same form. The infolist can't have this problem; the editor can, constantly.
2. There is no record on Create — getRecord() is null until the first save, so it would only ever be truthful on Edit.
3. Cost. The preview re-renders on every state change with no debounce, so each one would re-resolve the model plus any relations a brick touches.

Left for a follow-up, pending the answer in #56.

Workbench

PageMeta renders from the record rather than its own config, so the three call sites can be compared in a running app. It's seeded into PageFactory, so composer serve shows the contrast with no setup — open a page on the frontend, then edit the same page and watch that brick fall back to its placeholder.

Notes for review

• The PHPStan baseline gains one entry for the new workbench file. larastan can't resolve view names in this setup, so every view() call in the repo is already baselined. Confirmed file-specific rather than a real defect by swapping view names between Hero and PageMeta — the error follows the file, not the string. Appended by hand, because the baseline is included from phpstan.neon.dist and --generate-baseline would drop every existing entry.
• The README gains the @mason directive, which it has never documented at all. Since that section now describes three ways to render the same content, the helper and renderer got headings alongside it, matching docs/bricks/rendering.md.
• resources/dist and workbench/resources/dist are rebuilt; verified reproducible against a fresh npm run build.
• testbench.yaml now lists MasonServiceProvider explicitly. This is a no-op at runtime — the provider already loads through package discovery, and testbench route:list is identical with or without it — but it keeps the workbench's provider set readable in one place.

composer test is green — 201 passed, Rector/Pint/PHPStan clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BbLDdHUbgfXQ12eWn7zC68