Download the PHP package survos/schema-org-bundle without Composer

SurvosSchemaOrgBundle

Symfony integration for spatie/schema-org.

One request-scoped graph that any controller, listener, or service can contribute nodes to, rendered once as a single JSON-LD @graph script tag.

Why a graph and not one script tag per template

A page is rarely one thing. A movie page is a WebSite, a WebPage, a Movie, the Persons who directed and starred in it, an ImageObject, an AggregateRating — and the site-wide Organization that every page carries. Those come from different places in the code, so emitting them means either threading everything into one template variable or scattering <script> tags.

This bundle takes the first option and makes it painless: everyone writes into a shared SchemaOrgGraph, and render_schema_org() emits the lot.

Usage

Contribute nodes from anywhere:

Render it once, in base.html.twig:

It returns an empty string when the page collected nothing, so it is safe to call unconditionally. Pass a CSP nonce if the app needs one: {{ render_schema_org(csp_nonce) }}.

To see what a page collected, use the Schema.org item in the web debug toolbar — it shows the node count and opens a panel listing every node with the full JSON-LD. The panel also reports whether the graph actually reached the page, which is the one failure that is otherwise invisible: collecting nodes and publishing none looks exactly like a page with no structured data.

Without touching the constructor

Prefer normal constructor injection when you're editing the constructor anyway.

Identifiers and deduplication

add() keys each node by its own @id (spatie's identifier()), and the last write wins. Contributing the same @id from two places updates one node instead of duplicating it — which is what you want for the site-wide Organization, or a Person who is both the director and an actor.

Nodes with no @id are never deduplicated; each add() is a new node. To dedupe one anyway, pass an explicit key: $schemaOrg->add($image, 'hero').

When several places each fill in part of a node, use getOrCreate():

graph() returns the underlying Spatie\SchemaOrg\Graph for anything this wrapper doesn't expose (hide()/show(), a custom @context).

Attribute mapping

Annotate the entity and let SchemaOrgMapper build the node:

#[SchemaProperty] is orthogonal to #[Field] from survos/field-bundle — that one describes grid/search/API exposure, this one describes JSON-LD meaning. They coexist on the same property.

What it handles: scalars, DateTimeInterface, backed enums, lists, name-only relations (as:), nested #[SchemaOrg] objects, and cycles. Null, empty-string, and empty-list values are omitted rather than emitted as null — an absent property is correct JSON-LD; "name": null is not.

Property hooks, private(set) properties, constructor-promoted readonly properties, and zero-argument getters can all carry #[SchemaProperty]. #[SchemaOrg] is inherited up the parent chain (nearest declaration wins), so a DTO hierarchy needs it declared only where the type actually changes.

Full reference: docs/attributes.md — every parameter, what can carry the attribute, value handling, relations, references vs. embedding, cycle behaviour, and where the declarative half stops.

What it does not: cross-links, canonical-URL-derived @ids, and conditional rules stay hand-written. That's a deliberate boundary — expressing them in attributes means inventing a DSL. add() returns the node, so the rest is ordinary fluent spatie code:

Reflection runs once per class and is cached, so mapping 500 rows is one reflection pass, not 500.

References vs. embedding

A wrapped or nested node is by default added to the graph as its own top-level node and linked by @id — that's what keeps one Person one node when several entities share them. Pass reference: false to embed a copy instead.

A node with no @id can't be referenced, so it's embedded. The one exception is closing a cycle (A knows B knows A): embedding there would put the node inside itself and recurse until the stack ran out, so an un-@id'd back-link is dropped. Give anything reachable in a cycle an @id.

Worked examples

Four real consumers, in public repos, each exercising a different shape:

packages is the one to read first: it is the smallest, and it is the only one that uses the attribute mapper and hand-written composition side by side.

A note on picking types

packages deliberately models a Symfony bundle as SoftwareSourceCode, not SoftwareApplication — a bundle is source you compose into an app, not an application anyone installs and runs. That costs the Google software rich result, which only SoftwareApplication gets. It also declines to turn a GitHub star count into an aggregateRating, which would earn a rich result and would be a number nobody stated; stars become interactionStatistic with a LikeAction instead.

Both are the same trade: structured data is a claim about what something is. Reach for the type that is true, not the type with the SERP feature.

Commands

Prints what each mapped class maps to, property by property, and the public properties it doesn't map. That last list is the point: a mapping covering six of twenty-seven fields looks finished from the outside.

Fetches a URL, extracts every JSON-LD block, and checks it. Errors (exit 1): malformed JSON, missing @context, a node with no @type, duplicate @id. Warnings: a reference pointing at no node in the block, a null property value, and an @id/url whose scheme disagrees with the page it came from — that last one found a live reverse-proxy misconfiguration publishing http:// identities on an HTTPS site.

It deliberately does not check per-type "required" properties. Schema.org marks nothing required; those lists belong to Google's rich-result program, change without notice, and hardcoding them here would claim an authority this bundle doesn't have. Use Google's validator for eligibility, this for correctness.

JsonLdExtractor and JsonLdValidator are public services, so you can assert on a page's structured data from a functional test without shelling out.

Configuration

pretty_print defaults to %kernel.debug%, so there is normally nothing to configure.

auto_inject

With auto_inject: true a kernel.response listener inserts the script before the first </head> on successful, non-streamed text/html responses — no Twig call, whatever the layout.

It is off by default on purpose: {{ render_schema_org() }} is greppable, and injected output is not. Turn it on for apps whose layout you'd rather not edit, or that have several layouts.

A template that calls render_schema_org() suppresses the injection for that request, so enabling it can never produce two @graph blocks. The listener skips streamed, binary, compressed, attachment, non-HTML, redirect, and error responses, and drops a now-stale Content-Length.

Escaping

render_schema_org() encodes with JSON_HEX_TAG. A title or description containing </script> therefore cannot close the tag early — the angle brackets are emitted as JSON unicode escapes, which JSON-LD parsers read back as the original characters. This is not optional hardening; it is the difference between valid structured data and injected markup on any page with user-supplied text.

Worker mode

SchemaOrgGraph is a service, so under FrankenPHP worker mode or RoadRunner it outlives the request. SchemaOrgResetListener empties it on every main request (priority 4096) so one page's nodes never leak into the next. Under php-fpm this is a no-op.

Changelog

CHANGELOG.md. Note 2.24.12 removed schema_org_debug() and the debug_panel config node in favour of the profiler data collector; a template still calling it will 500.

Tests