Fork 0

ernstla added 6 commits 2026年07月05日 23:25:08 +02:00

Cosray-owned richtext storage format v1 (spec: hq/cosray/richtext-format.md):
Validator (writer-strict, config-declared classes/styles), Normalizer
(canonical form: key order, default omission, run merging, empty -> null),
Renderer (doc -> HTML, uid resolution via Resolver, render-time link rel
policy, reader-tolerant), Scanner (image.uid/link.node/link.asset), the
Envelope marker, and richtext.classes/richtext.styles config.

Field/RichText saves are writer-strict: the shape requires the envelope
(format/version) and validates every locale document against the spec
vocabulary and the declared richtext.classes/richtext.styles; richtext
blocks get the same treatment in BlockValidator. Store canonicalizes
envelope-marked values on save (byte-stable storage, empty docs -> null).
Value/RichText and the richtext block type render structured documents
through the Renderer (legacy format-less HTML keeps the sanitizer path);
uid references resolve through Owner::paths() (new UrlPaths service +
paths/activeByNodeUid.sql) and the asset catalog. The serializer's asset
map now includes link.asset references via Scanner::scanContent, and
NodeContentNormalizer passes the envelope through byte-identically.

The editor keeps ProseMirror behind the internal driver seam
(editor.ts + format.ts + schema.ts): components bind cosray richtext
documents, format.ts adapts them to and from the live PM doc, legacy
HTML parses through the editor schema once at mount and every save
submits the envelope (host.ts/FormPatch carry format/version through
the [json] channel; element.php exposes them in the SSR payload).
Schema: fontSize becomes the config-declared style mark, link learns
node/asset target kinds (rel dropped - render-time policy), and a new
inline image node references catalog assets by uid with a thumbnail
node view. Toolbar: paragraph classes and text styles come from
richtext.classes/richtext.styles (empty config = no control), image
insertion via a library/upload modal reusing LibraryBrowser and the
pool-first upload endpoint. Richtext blocks convert and stamp the
envelope the same way; their init is structured from the start.

The headless converter (panel/src/tools/richtext-convert.ts, bundled
to panel/build/tools/richtext-convert.mjs) parses legacy HTML through
the panel's own editor schema under jsdom, so migrated content is
byte-identical to what the editor itself would produce.
Migration 000000-000020 streams every legacy richtext value (fields,
blocks, entries subfields; nodes + drafts) through the converter as
NDJSON, resolves reference carriers (internal hrefs -> link.node via
url_paths, uid-form asset URLs -> link.asset, legacy <img src> tagged
with data-uid so the parse keeps them as image nodes), drops no-op
cms-text-base style marks, canonicalizes through the Normalizer
(empty -> null) and stamps the envelope. Idempotent: entries carrying
a format key are skipped. A report at richtext-migration-report.json
lists undeclared classes/styles, unresolved links, and dropped
images; the seed migration records 020 for fresh installs.

Migration 020 matches internal links with trailing slashes (legacy
content links pages that way) and resolves pre-catalog owner-scoped
asset URLs through asset-legacy-map.json where present. The
serializer's read payload gains a 'paths' map (node uid -> locale
path map) for every link.node reference, mirroring the assets map so
headless consumers can resolve internal richtext links.

read() merges the node's own 'paths' (locale to URL path of the
serialized node) over the base array, which silently clobbered the
new reference map.

ernstla added 1 commit 2026年07月06日 13:34:56 +02:00

All databases convert through migration 020 in the same deploy, so
format-less values (unmigrated legacy HTML) now render empty instead
of falling back to sanitized passthrough. Value/RichText loses
clean() and the sanitizer dependency (excerpt() works on the rendered
output); the richtext block type renders structured documents only.
The html format marker remains solely as the migration's input state.

ernstla merged commit 2977bd58b5 into main 2026年07月06日 13:38:21 +02:00

ernstla deleted branch rewrite/richtext-format 2026年07月06日 13:38:21 +02:00

Sign in to join this conversation.

Clear labels

bug

Something isn't working

dependencies

Pull requests that update a dependency file

documentation

Improvements or additions to documentation

duplicate

This issue or pull request already exists

enhancement

New feature or request

good first issue

Good for newcomers

help wanted

Extra attention is needed

invalid

This doesn't seem right

javascript

Pull requests that update javascript code

question

Further information is requested

wontfix

This will not be worked on

Milestone

No milestone

Subscribe

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".