@openleaf-editor/core 0.1.0-beta.2 → 0.1.0-beta.4

package/dist/preserve.js

@@ -27,7 +27,8 @@
  * and is preserved intact.
  */
 import { isFullyModelledStyle, safeLang } from './css.js';
-import { URL_ATTRIBUTES, isEventHandlerAttribute, isSafeUrl } from './url.js';
+import { DROP_WITH_CONTENT } from './elements.js';
+import { URL_ATTRIBUTES, isEventHandlerAttribute, isNeverCarriedAttribute, isSafeUrl, } from './url.js';
 /**
  * Elements that are never preserved, and whose contents are discarded with them.
  *
@@ -45,27 +46,36 @@ import { URL_ATTRIBUTES, isEventHandlerAttribute, isSafeUrl } from './url.js';
  * `ignore` rather than `skip`: the element AND its contents go. Skipping would
  * unwrap `<script>alert(1)</script>` into the literal text "alert(1)" appearing
  * in the document, which is a different kind of wrong.
+ *
+ * Spread from `@openleaf-editor/content-policy` rather than written out here.
+ * This list and the sanitize policy's `dropWithContent` are the same decision
+ * made in two packages, they were maintained by hand, and they drifted: `<svg>`
+ * and `<math>` were on the sanitizer's list and missing from this one, so the
+ * editor stored exactly the markup the server was configured to delete. Sharing
+ * the constant is what makes that drift impossible rather than merely unlikely.
+ *
+ * Exported so the divergence can be asserted in a test as well as prevented by
+ * construction. Not re-exported from the package index: it is an internal
+ * invariant, and the shared list itself is public in content-policy.
  */
-const NEVER_PRESERVE = [
-    'script',
-    'style',
-    'frame',
-    'frameset',
-    'object',
-    'embed',
-    'applet',
-    'form',
-    'input',
-    'button',
-    'select',
-    'textarea',
-    'option',
-    'link',
-    'meta',
-    'base',
-    'noscript',
-    'template',
-];
+export const NEVER_PRESERVE = [...DROP_WITH_CONTENT];
+/**
+ * Tags that must not survive *inside* a preserved subtree either.
+ *
+ * `iframe` is here and not in `NEVER_PRESERVE` because the two answer different
+ * questions. At the top level an iframe is claimed by the modelled embed node
+ * when its `src` is an allowlisted player, and ignored otherwise -- a
+ * priority-100 drop rule would outrank the embed node and delete legitimate
+ * players. Inside preserved markup there is no such question: the subtree is
+ * stored as an opaque string and re-emitted verbatim, so nothing re-checks the
+ * frame on the way out.
+ *
+ * That gap was the bypass. `<iframe src="https://evil.example/">` on its own was
+ * dropped; wrapped in a `<div class="c">` it round-tripped byte-identical, with
+ * `allow="camera; microphone; geolocation"` intact. One attribute on a wrapper
+ * defeated both the host allowlist and the permissions filter.
+ */
+const NEVER_INSIDE_PRESERVED = new Set([...NEVER_PRESERVE, 'iframe']);
 /** Parse rules that drop dangerous elements before any other rule sees them. */
 const dropRules = NEVER_PRESERVE.map((tag) => ({ tag, ignore: true, priority: 100 }));
 /**
@@ -140,7 +150,7 @@ export function scrub(el) {
     const clone = el.cloneNode(true);
     const visit = (node) => {
         for (const child of Array.from(node.children)) {
-            if (NEVER_PRESERVE.includes(child.nodeName.toLowerCase())) {
+            if (NEVER_INSIDE_PRESERVED.has(child.nodeName.toLowerCase())) {
                 child.remove();
                 continue;
             }
@@ -151,6 +161,13 @@ export function scrub(el) {
                 node.removeAttribute(attr.name);
                 continue;
             }
+            // Before the URL question, not as a case of it: `srcdoc` is a document
+            // rather than a URL, and asking a scheme checker about a document gets
+            // "no scheme, therefore relative, therefore safe".
+            if (isNeverCarriedAttribute(attr.name)) {
+                node.removeAttribute(attr.name);
+                continue;
+            }
             if (URL_ATTRIBUTES.has(attr.name.toLowerCase()) && !isSafeUrl(attr.value)) {
                 node.removeAttribute(attr.name);
             }
@@ -231,13 +248,63 @@ export function isLosslesslyUnwrappable(el) {
     }
     return true;
 }
+/**
+ * Parsed preserved markup, per Document, keyed on the exact stored string.
+ *
+ * `toDOM` runs on every render of every preserved atom, and an HTML parse is the
+ * most expensive thing in it: a document that is half preserved markup spent
+ * roughly 4x as long serializing as the same document in plain paragraphs.
+ *
+ * Caching one is only sound because the key is the whole input. `html` is a node
+ * attribute, ProseMirror nodes are immutable, and the string was produced by
+ * `scrub` before it was ever stored -- so two calls with the same string must
+ * produce the same element, and there is no edit that can invalidate an entry
+ * without changing the key.
+ *
+ * Three properties this must keep, none of them optional:
+ *
+ *   A CLONE every time. Handing out the cached node would let one caller's
+ *   mutation -- ProseMirror appending it, a decoration setting an attribute --
+ *   poison every later render of that markup.
+ *
+ *   PER DOCUMENT. Serialization can target a Document that is not the global
+ *   one, and a node has to be owned by the document it is going into. Scoping by
+ *   Document also means the entries die with it, so a throwaway serialization
+ *   document is still collectable.
+ *
+ *   BOUNDED. The keys are document content, so an unbounded map is a leak that
+ *   grows with what the user pastes. A cap with clear-on-overflow keeps it flat;
+ *   very large strings are not cached at all, because they are both the least
+ *   likely to repeat and the most expensive to hold.
+ */
+const PARSE_CACHE_ENTRIES = 256;
+const PARSE_CACHE_MAX_LENGTH = 4096;
+const parseCaches = new WeakMap();
 /** Rebuild a DOM element from stored markup. `<template>` is used because
  *  its parsing context permits otherwise-illegal fragments such as a bare
  *  `<tr>`, which a `<div>` container would silently discard. */
 function elementFromHtml(html, doc) {
+    let cache = parseCaches.get(doc);
+    const hit = cache?.get(html);
+    if (hit)
+        return hit.cloneNode(true);
     const tpl = doc.createElement('template');
     tpl.innerHTML = html;
-    return tpl.content.firstElementChild;
+    const parsed = tpl.content.firstElementChild;
+    if (!parsed)
+        return null;
+    if (html.length > PARSE_CACHE_MAX_LENGTH)
+        return parsed;
+    if (!cache) {
+        cache = new Map();
+        parseCaches.set(doc, cache);
+    }
+    if (cache.size >= PARSE_CACHE_ENTRIES)
+        cache.clear();
+    // The master copy is the clone, so the element handed back is the freshly
+    // parsed one and the cache holds something no caller has ever touched.
+    cache.set(html, parsed.cloneNode(true));
+    return parsed;
 }
 /**
  * ProseMirror's `toDOM` does not receive the `document` passed to
@@ -258,25 +325,6 @@ export function withSerializationDocument(doc, fn) {
         serializationDocument = previous;
     }
 }
-/**
- * True while `serializeHtml` is running.
- *
- * A node whose `toDOM` needs to render differently for the editor than for the
- * saved HTML has no other way to tell which one it is building. The table spec
- * needs exactly that: a preserved `<caption>` must be `contenteditable="false"`
- * on screen, because it sits inside the editable area but outside the node's
- * `contentDOM`, and letting a caret into it means typing that ProseMirror will
- * silently revert. That attribute must NOT reach the saved HTML, where it would
- * be our editor scribbling on the author's markup.
- *
- * Stripping it afterwards was the other option and is worse: it cannot tell the
- * attribute it just added from the same attribute in somebody's document -- the
- * collision the preserved-element WeakSet above exists to avoid. Not emitting it
- * at all has no such failure mode.
- */
-export function isSerializing() {
-    return serializationDocument !== undefined;
-}
 /**
  * The Document that serialization should build into.
  *
@@ -377,6 +425,114 @@ export const unknownBlock = {
         return rebuildOrCarry(node.attrs['html'], 'div');
     },
 };
+/**
+ * Where the inline catch-all is allowed to fire.
+ *
+ * ProseMirror matches `context` against the parent node the content is being
+ * parsed INTO, and the naive list -- paragraph and heading -- described the
+ * wrong thing. It named the nodes that hold inline content, when what the rule
+ * needs to name is every node that can END UP holding it.
+ *
+ * `list_item` is `paragraph block*`, so its implicit paragraph does not exist
+ * until some inline content arrives to force it. An unknown element that is the
+ * FIRST child of an `<li>` therefore sees a context of `.../list_item/`, the
+ * inline rule declines, and `unknownBlock` claims a block-level atom inside a
+ * container whose content expression cannot hold one. ProseMirror's recovery is
+ * to close the list and emit the atom as a sibling, so
+ * `<ol><li><ins>a</ins></li><li>b</li></ol>` came back as an emptied `<ol>`, an
+ * escaped `<ins>`, and a `<ul>` -- the list's contents gone and its very type
+ * changed. The identical element with a single character of text before it
+ * round-tripped perfectly, which is what kept this invisible: position was the
+ * whole trigger.
+ *
+ * `summary` and `figcaption` fail the same way for the same reason, and
+ * `blockquote`, `table_cell` and `table_header` are listed so that a first-child
+ * unknown *inline* element is wrapped in a paragraph exactly as it is when text
+ * precedes it. Those three can also hold a block atom legally. Wrapping a
+ * p-closing element (`div`, `section`, …) in that paragraph is the other
+ * failure: the HTML parser closes the `<p>` on the next parse, leaves empty
+ * paragraphs on both sides, and the document grows by two blanks on every save.
+ * `unknownInline` declines those tags so `unknownBlock` claims them instead.
+ *
+ * Naming containers explicitly rather than dropping the context altogether:
+ * without it the inline rule would outrank `unknownBlock` everywhere, and a
+ * genuinely block-level unknown element at the top of the document would become
+ * an inline atom inside a paragraph the author never had.
+ */
+const INLINE_CONTEXT = [
+    'paragraph/',
+    'heading/',
+    'summary/',
+    'figcaption/',
+    'list_item/',
+    'table_cell/',
+    'table_header/',
+    'blockquote/',
+].join('|');
+/**
+ * Start tags that close an open `<p>` in the HTML "in body" insertion mode.
+ *
+ * Serializing one of these as an inline atom wraps it in `<p>…</p>`. The next
+ * parse cannot keep it there, so it splits into an empty paragraph, the
+ * element, and another empty paragraph -- and the following save wraps it
+ * again. The same unbounded growth `<plaintext>` caused by having no end tag.
+ *
+ * Shared with anything else that must not claim a p-closing element as inline
+ * (`figcaption` being inline is the other door into this). Custom elements are
+ * not on the list: they do not close a `<p>`, so `<drupal-media>` inside a
+ * blockquote is correctly an inline atom.
+ *
+ * Source: WHATWG HTML parsing, "in body" -- every start tag that runs
+ * "if the stack of open elements has a p element in button scope, then close
+ * a p element" before inserting.
+ */
+export const CLOSES_OPEN_P = new Set([
+    'address',
+    'article',
+    'aside',
+    'blockquote',
+    'center',
+    'details',
+    'dialog',
+    'dir',
+    'div',
+    'dl',
+    'dt',
+    'dd',
+    'fieldset',
+    'figcaption',
+    'figure',
+    'footer',
+    'header',
+    'hgroup',
+    'main',
+    'menu',
+    'nav',
+    'ol',
+    'p',
+    'search',
+    'section',
+    'summary',
+    'ul',
+    'h1',
+    'h2',
+    'h3',
+    'h4',
+    'h5',
+    'h6',
+    'listing',
+    'plaintext',
+    'pre',
+    'form',
+    'li',
+    'button',
+    'table',
+    'hr',
+    'xmp',
+    'applet',
+    'marquee',
+    'object',
+]);
 /**
  * Inline preserved content, for unrecognised markup appearing inside a
  * paragraph -- the `<o:p>` and `<w:sdt>` debris of a Word paste, custom
@@ -401,11 +557,26 @@ export const unknownInline = {
             // the block rule cannot claim inline debris and split the paragraph
             // that contained it.
             priority: 1,
-            context: 'paragraph/|heading/',
+            context: INLINE_CONTEXT,
             getAttrs(dom) {
                 const el = dom;
                 if (isLosslesslyUnwrappable(el))
                     return false;
+                // A block-level element inside a paragraph-holding container must not
+                // become an inline atom: emitting it inside `<p>` is markup the HTML
+                // parser will not re-parse as itself. Declining lets unknownBlock
+                // claim it, which blockquote (`block+`) and list_item (`paragraph
+                // block*`) can hold. `<ins>` and custom elements stay on this path.
+                // `figcaption` is on CLOSES_OPEN_P, but declining it in summary/heading
+                // sends it to unknownBlock, which those containers cannot hold.
+                const tag = el.nodeName.toLowerCase();
+                if (tag === 'figcaption') {
+                    if (el.parentElement?.nodeName.toLowerCase() === 'p')
+                        return false;
+                }
+                else if (CLOSES_OPEN_P.has(tag)) {
+                    return false;
+                }
                 return { html: scrub(el), tag: el.nodeName.toLowerCase() };
             },
         },

package/dist/preserve.js.map

@@ -1 +1 @@
-{"version":3,"file":"preserve.js","sourceRoot":"","sources":["../src/preserve.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAGH,OAAO,EAAE,oBAAoB,EAAE,QAAQ,EAAE,MAAM,UAAU,CAAA;AACzD,OAAO,EAAE,cAAc,EAAE,uBAAuB,EAAE,SAAS,EAAE,MAAM,UAAU,CAAA;AAE7E;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,cAAc,GAAsB;IACxC,QAAQ;IACR,OAAO;IACP,OAAO;IACP,UAAU;IACV,QAAQ;IACR,OAAO;IACP,QAAQ;IACR,MAAM;IACN,OAAO;IACP,QAAQ;IACR,QAAQ;IACR,UAAU;IACV,QAAQ;IACR,MAAM;IACN,MAAM;IACN,MAAM;IACN,UAAU;IACV,UAAU;CACX,CAAA;AAED,gFAAgF;AAChF,MAAM,SAAS,GAAG,cAAc,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,CAAC,CAAC,CAAA;AAErF;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAAwB,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAA;AAErF;;;;;;GAMG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAW;IAC1C,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,EAAE,CAAC;QAC9C,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;YACzB,IAAI,CAAC,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;gBAAE,OAAO,IAAI,CAAA;YACxE,SAAQ;QACV,CAAC;QACD,wEAAwE;QACxE,sDAAsD;QACtD,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE;YAAE,OAAO,IAAI,CAAA;IAClF,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,yBAAyB,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IACjE,GAAG;IACH,QAAQ,EAAE,EAAE;IACZ,QAAQ,CAAC,GAAgB;QACvB,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC;YAAE,OAAO,KAAK,CAAA;QACxC,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,GAAG,CAAC,QAAQ,CAAC,WAAW,EAAE,EAAE,CAAA;IAC9D,CAAC;CACF,CAAC,CAAC,CAAA;AAEH;;;;;GAKG;AACH,MAAM,iBAAiB,GAAG,CAAC,QAAQ,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IACnE,GAAG;IACH,MAAM,EAAE,IAAa;IACrB,QAAQ,EAAE,EAAE;CACb,CAAC,CAAC,CAAA;AAEH;;;;;;;;;;GAUG;AACH,MAAM,UAAU,KAAK,CAAC,EAAW;IAC/B,MAAM,KAAK,GAAG,EAAE,CAAC,SAAS,CAAC,IAAI,CAAY,CAAA;IAE3C,MAAM,KAAK,GAAG,CAAC,IAAa,EAAQ,EAAE;QACpC,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC9C,IAAI,cAAc,CAAC,QAAQ,CAAC,KAAK,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC;gBAC1D,KAAK,CAAC,MAAM,EAAE,CAAA;gBACd,SAAQ;YACV,CAAC;YACD,KAAK,CAAC,KAAK,CAAC,CAAA;QACd,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,EAAE,CAAC;YAC/C,IAAI,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;gBACvC,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;gBAC/B,SAAQ;YACV,CAAC;YACD,IAAI,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;gBAC1E,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;YACjC,CAAC;QACH,CAAC;IACH,CAAC,CAAA;IAED,KAAK,CAAC,KAAK,CAAC,CAAA;IACZ,OAAO,KAAK,CAAC,SAAS,CAAA;AACxB,CAAC;AAED;;;;;;;;;;GAUG;AACH,MAAM,sBAAsB,GAAwB,IAAI,GAAG,CAAC;IAC1D,KAAK;IACL,SAAS;IACT,SAAS;IACT,MAAM;IACN,OAAO;IACP,QAAQ;IACR,QAAQ;IACR,KAAK;IACL,MAAM;IACN,QAAQ;CACT,CAAC,CAAA;AAEF;;;;;;;;GAQG;AACH,MAAM,UAAU,uBAAuB,CAAC,EAAW;IACjD,MAAM,IAAI,GAAG,EAAE,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAA;IACtC,IAAI,CAAC,sBAAsB,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAA;IACnD,IAAI,EAAE,CAAC,UAAU,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAA;IAC3C,IAAI,IAAI,KAAK,MAAM;QAAE,OAAO,KAAK,CAAA;IAEjC;;;;;;;;;;;;;;OAcG;IACH,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,EAAE,CAAC;QAC7C,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;YACzB,IAAI,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,IAAI;gBAAE,OAAO,KAAK,CAAA;YAC/C,SAAQ;QACV,CAAC;QACD,IAAI,IAAI,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;YAC1B,IAAK,EAAkB,CAAC,KAAK,EAAE,MAAM,KAAK,CAAC;gBAAE,OAAO,KAAK,CAAA;YACzD,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,KAAK,CAAC;gBAAE,OAAO,KAAK,CAAA;YACnD,SAAQ;QACV,CAAC;QACD,OAAO,KAAK,CAAA;IACd,CAAC;IACD,OAAO,IAAI,CAAA;AACb,CAAC;AAED;;gEAEgE;AAChE,SAAS,eAAe,CAAC,IAAY,EAAE,GAAa;IAClD,MAAM,GAAG,GAAG,GAAG,CAAC,aAAa,CAAC,UAAU,CAAC,CAAA;IACzC,GAAG,CAAC,SAAS,GAAG,IAAI,CAAA;IACpB,OAAO,GAAG,CAAC,OAAO,CAAC,iBAAiB,CAAA;AACtC,CAAC;AAED;;;;;;GAMG;AACH,IAAI,qBAA2C,CAAA;AAE/C,0EAA0E;AAC1E,MAAM,UAAU,yBAAyB,CAAI,GAAa,EAAE,EAAW;IACrE,MAAM,QAAQ,GAAG,qBAAqB,CAAA;IACtC,qBAAqB,GAAG,GAAG,CAAA;IAC3B,IAAI,CAAC;QACH,OAAO,EAAE,EAAE,CAAA;IACb,CAAC;YAAS,CAAC;QACT,qBAAqB,GAAG,QAAQ,CAAA;IAClC,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,aAAa;IAC3B,OAAO,qBAAqB,KAAK,SAAS,CAAA;AAC5C,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,mBAAmB;IACjC,OAAO,aAAa,EAAE,CAAA;AACxB,CAAC;AAED,SAAS,aAAa;IACpB,IAAI,qBAAqB;QAAE,OAAO,qBAAqB,CAAA;IACvD,IAAI,OAAO,QAAQ,KAAK,WAAW,EAAE,CAAC;QACpC,MAAM,IAAI,KAAK,CACb,2EAA2E;YACzE,+DAA+D;YAC/D,sCAAsC,CACzC,CAAA;IACH,CAAC;IACD,OAAO,QAAQ,CAAA;AACjB,CAAC;AAGD;;;;;;;;GAQG;AACH;;;;;;;;;;;;;GAaG;AACH,MAAM,iBAAiB,GAAG,IAAI,OAAO,EAAW,CAAA;AAEhD,kFAAkF;AAClF,MAAM,UAAU,iBAAiB,CAAC,IAAoB;IACpD,KAAK,IAAI,OAAO,GAAG,IAAI,EAAE,OAAO,EAAE,OAAO,GAAG,OAAO,CAAC,aAAa,EAAE,CAAC;QAClE,IAAI,iBAAiB,CAAC,GAAG,CAAC,OAAO,CAAC;YAAE,OAAO,IAAI,CAAA;IACjD,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC;AAED,SAAS,cAAc,CAAC,IAAY,EAAE,WAA2B;IAC/D,MAAM,GAAG,GAAG,aAAa,EAAE,CAAA;IAC3B,MAAM,OAAO,GAAG,eAAe,CAAC,IAAI,EAAE,GAAG,CAAC,CAAA;IAC1C,IAAI,OAAO,EAAE,CAAC;QACZ,iBAAiB,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;QAC9B,OAAO,OAAO,CAAA;IAChB,CAAC;IACD,MAAM,OAAO,GAAG,GAAG,CAAC,aAAa,CAAC,WAAW,CAAC,CAAA;IAC9C,OAAO,CAAC,YAAY,CAAC,0BAA0B,EAAE,IAAI,CAAC,CAAA;IACtD,iBAAiB,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;IAC9B,OAAO,OAAO,CAAA;AAChB,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,MAAM,YAAY,GAAa;IACpC,KAAK,EAAE,OAAO;IACd,IAAI,EAAE,IAAI;IACV,UAAU,EAAE,IAAI;IAChB,SAAS,EAAE,IAAI;IACf,KAAK,EAAE;QACL,IAAI,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE;QACrB,GAAG,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE;KACxB;IACD,QAAQ,EAAE;QACR,GAAG,SAAS;QACZ,GAAG,yBAAyB;QAC5B,GAAG,iBAAiB;QACpB;YACE,GAAG,EAAE,GAAG;YACR,qEAAqE;YACrE,wDAAwD;YACxD,QAAQ,EAAE,CAAC;YACX,QAAQ,CAAC,GAAG;gBACV,MAAM,EAAE,GAAG,GAAc,CAAA;gBACzB,kEAAkE;gBAClE,kEAAkE;gBAClE,IAAI,uBAAuB,CAAC,EAAE,CAAC;oBAAE,OAAO,KAAK,CAAA;gBAC7C,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC,EAAE,GAAG,EAAE,EAAE,CAAC,QAAQ,CAAC,WAAW,EAAE,EAAE,CAAA;YAC5D,CAAC;SACF;KACF;IACD,KAAK,CAAC,IAAI;QACR,OAAO,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAW,EAAE,KAAK,CAAC,CAAA;IAC5D,CAAC;CACF,CAAA;AAED;;;;GAIG;AACH,MAAM,CAAC,MAAM,aAAa,GAAa;IACrC,KAAK,EAAE,QAAQ;IACf,MAAM,EAAE,IAAI;IACZ,IAAI,EAAE,IAAI;IACV,UAAU,EAAE,IAAI;IAChB,KAAK,EAAE;QACL,IAAI,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE;QACrB,GAAG,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE;KACzB;IACD,QAAQ,EAAE;QACR,GAAG,SAAS;QACZ,GAAG,yBAAyB;QAC5B,GAAG,iBAAiB;QACpB;YACE,GAAG,EAAE,GAAG;YACR,qEAAqE;YACrE,oEAAoE;YACpE,qBAAqB;YACrB,QAAQ,EAAE,CAAC;YACX,OAAO,EAAE,qBAAqB;YAC9B,QAAQ,CAAC,GAAG;gBACV,MAAM,EAAE,GAAG,GAAc,CAAA;gBACzB,IAAI,uBAAuB,CAAC,EAAE,CAAC;oBAAE,OAAO,KAAK,CAAA;gBAC7C,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC,EAAE,GAAG,EAAE,EAAE,CAAC,QAAQ,CAAC,WAAW,EAAE,EAAE,CAAA;YAC5D,CAAC;SACF;KACF;IACD,KAAK,CAAC,IAAI;QACR,OAAO,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAW,EAAE,MAAM,CAAC,CAAA;IAC7D,CAAC;CACF,CAAA"}
+{"version":3,"file":"preserve.js","sourceRoot":"","sources":["../src/preserve.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAGH,OAAO,EAAE,oBAAoB,EAAE,QAAQ,EAAE,MAAM,UAAU,CAAA;AACzD,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAA;AACjD,OAAO,EACL,cAAc,EACd,uBAAuB,EACvB,uBAAuB,EACvB,SAAS,GACV,MAAM,UAAU,CAAA;AAEjB;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AACH,MAAM,CAAC,MAAM,cAAc,GAAsB,CAAC,GAAG,iBAAiB,CAAC,CAAA;AAEvE;;;;;;;;;;;;;;;GAeG;AACH,MAAM,sBAAsB,GAAwB,IAAI,GAAG,CAAC,CAAC,GAAG,cAAc,EAAE,QAAQ,CAAC,CAAC,CAAA;AAE1F,gFAAgF;AAChF,MAAM,SAAS,GAAG,cAAc,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,CAAC,CAAC,CAAA;AAErF;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAAwB,IAAI,GAAG,CAAC,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAA;AAErF;;;;;;GAMG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAW;IAC1C,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,EAAE,CAAC;QAC9C,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;YACzB,IAAI,CAAC,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;gBAAE,OAAO,IAAI,CAAA;YACxE,SAAQ;QACV,CAAC;QACD,wEAAwE;QACxE,sDAAsD;QACtD,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,WAAW,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE;YAAE,OAAO,IAAI,CAAA;IAClF,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,yBAAyB,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IACjE,GAAG;IACH,QAAQ,EAAE,EAAE;IACZ,QAAQ,CAAC,GAAgB;QACvB,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC;YAAE,OAAO,KAAK,CAAA;QACxC,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,GAAG,EAAE,GAAG,CAAC,QAAQ,CAAC,WAAW,EAAE,EAAE,CAAA;IAC9D,CAAC;CACF,CAAC,CAAC,CAAA;AAEH;;;;;GAKG;AACH,MAAM,iBAAiB,GAAG,CAAC,QAAQ,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;IACnE,GAAG;IACH,MAAM,EAAE,IAAa;IACrB,QAAQ,EAAE,EAAE;CACb,CAAC,CAAC,CAAA;AAEH;;;;;;;;;;GAUG;AACH,MAAM,UAAU,KAAK,CAAC,EAAW;IAC/B,MAAM,KAAK,GAAG,EAAE,CAAC,SAAS,CAAC,IAAI,CAAY,CAAA;IAE3C,MAAM,KAAK,GAAG,CAAC,IAAa,EAAQ,EAAE;QACpC,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC9C,IAAI,sBAAsB,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC;gBAC7D,KAAK,CAAC,MAAM,EAAE,CAAA;gBACd,SAAQ;YACV,CAAC;YACD,KAAK,CAAC,KAAK,CAAC,CAAA;QACd,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,EAAE,CAAC;YAC/C,IAAI,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;gBACvC,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;gBAC/B,SAAQ;YACV,CAAC;YACD,uEAAuE;YACvE,uEAAuE;YACvE,mDAAmD;YACnD,IAAI,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;gBACvC,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;gBAC/B,SAAQ;YACV,CAAC;YACD,IAAI,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;gBAC1E,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;YACjC,CAAC;QACH,CAAC;IACH,CAAC,CAAA;IAED,KAAK,CAAC,KAAK,CAAC,CAAA;IACZ,OAAO,KAAK,CAAC,SAAS,CAAA;AACxB,CAAC;AAED;;;;;;;;;;GAUG;AACH,MAAM,sBAAsB,GAAwB,IAAI,GAAG,CAAC;IAC1D,KAAK;IACL,SAAS;IACT,SAAS;IACT,MAAM;IACN,OAAO;IACP,QAAQ;IACR,QAAQ;IACR,KAAK;IACL,MAAM;IACN,QAAQ;CACT,CAAC,CAAA;AAEF;;;;;;;;GAQG;AACH,MAAM,UAAU,uBAAuB,CAAC,EAAW;IACjD,MAAM,IAAI,GAAG,EAAE,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAA;IACtC,IAAI,CAAC,sBAAsB,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAA;IACnD,IAAI,EAAE,CAAC,UAAU,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAA;IAC3C,IAAI,IAAI,KAAK,MAAM;QAAE,OAAO,KAAK,CAAA;IAEjC;;;;;;;;;;;;;;OAcG;IACH,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,EAAE,CAAC;QAC7C,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;YACzB,IAAI,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,IAAI;gBAAE,OAAO,KAAK,CAAA;YAC/C,SAAQ;QACV,CAAC;QACD,IAAI,IAAI,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;YAC1B,IAAK,EAAkB,CAAC,KAAK,EAAE,MAAM,KAAK,CAAC;gBAAE,OAAO,KAAK,CAAA;YACzD,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,KAAK,CAAC;gBAAE,OAAO,KAAK,CAAA;YACnD,SAAQ;QACV,CAAC;QACD,OAAO,KAAK,CAAA;IACd,CAAC;IACD,OAAO,IAAI,CAAA;AACb,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AACH,MAAM,mBAAmB,GAAG,GAAG,CAAA;AAC/B,MAAM,sBAAsB,GAAG,IAAI,CAAA;AACnC,MAAM,WAAW,GAAG,IAAI,OAAO,EAAkC,CAAA;AAEjE;;gEAEgE;AAChE,SAAS,eAAe,CAAC,IAAY,EAAE,GAAa;IAClD,IAAI,KAAK,GAAG,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;IAChC,MAAM,GAAG,GAAG,KAAK,EAAE,GAAG,CAAC,IAAI,CAAC,CAAA;IAC5B,IAAI,GAAG;QAAE,OAAO,GAAG,CAAC,SAAS,CAAC,IAAI,CAAY,CAAA;IAE9C,MAAM,GAAG,GAAG,GAAG,CAAC,aAAa,CAAC,UAAU,CAAC,CAAA;IACzC,GAAG,CAAC,SAAS,GAAG,IAAI,CAAA;IACpB,MAAM,MAAM,GAAG,GAAG,CAAC,OAAO,CAAC,iBAAiB,CAAA;IAC5C,IAAI,CAAC,MAAM;QAAE,OAAO,IAAI,CAAA;IACxB,IAAI,IAAI,CAAC,MAAM,GAAG,sBAAsB;QAAE,OAAO,MAAM,CAAA;IAEvD,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,KAAK,GAAG,IAAI,GAAG,EAAE,CAAA;QACjB,WAAW,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAA;IAC7B,CAAC;IACD,IAAI,KAAK,CAAC,IAAI,IAAI,mBAAmB;QAAE,KAAK,CAAC,KAAK,EAAE,CAAA;IACpD,0EAA0E;IAC1E,uEAAuE;IACvE,KAAK,CAAC,GAAG,CAAC,IAAI,EAAE,MAAM,CAAC,SAAS,CAAC,IAAI,CAAY,CAAC,CAAA;IAClD,OAAO,MAAM,CAAA;AACf,CAAC;AAED;;;;;;GAMG;AACH,IAAI,qBAA2C,CAAA;AAE/C,0EAA0E;AAC1E,MAAM,UAAU,yBAAyB,CAAI,GAAa,EAAE,EAAW;IACrE,MAAM,QAAQ,GAAG,qBAAqB,CAAA;IACtC,qBAAqB,GAAG,GAAG,CAAA;IAC3B,IAAI,CAAC;QACH,OAAO,EAAE,EAAE,CAAA;IACb,CAAC;YAAS,CAAC;QACT,qBAAqB,GAAG,QAAQ,CAAA;IAClC,CAAC;AACH,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,mBAAmB;IACjC,OAAO,aAAa,EAAE,CAAA;AACxB,CAAC;AAED,SAAS,aAAa;IACpB,IAAI,qBAAqB;QAAE,OAAO,qBAAqB,CAAA;IACvD,IAAI,OAAO,QAAQ,KAAK,WAAW,EAAE,CAAC;QACpC,MAAM,IAAI,KAAK,CACb,2EAA2E;YACzE,+DAA+D;YAC/D,sCAAsC,CACzC,CAAA;IACH,CAAC;IACD,OAAO,QAAQ,CAAA;AACjB,CAAC;AAGD;;;;;;;;GAQG;AACH;;;;;;;;;;;;;GAaG;AACH,MAAM,iBAAiB,GAAG,IAAI,OAAO,EAAW,CAAA;AAEhD,kFAAkF;AAClF,MAAM,UAAU,iBAAiB,CAAC,IAAoB;IACpD,KAAK,IAAI,OAAO,GAAG,IAAI,EAAE,OAAO,EAAE,OAAO,GAAG,OAAO,CAAC,aAAa,EAAE,CAAC;QAClE,IAAI,iBAAiB,CAAC,GAAG,CAAC,OAAO,CAAC;YAAE,OAAO,IAAI,CAAA;IACjD,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC;AAED,SAAS,cAAc,CAAC,IAAY,EAAE,WAA2B;IAC/D,MAAM,GAAG,GAAG,aAAa,EAAE,CAAA;IAC3B,MAAM,OAAO,GAAG,eAAe,CAAC,IAAI,EAAE,GAAG,CAAC,CAAA;IAC1C,IAAI,OAAO,EAAE,CAAC;QACZ,iBAAiB,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;QAC9B,OAAO,OAAO,CAAA;IAChB,CAAC;IACD,MAAM,OAAO,GAAG,GAAG,CAAC,aAAa,CAAC,WAAW,CAAC,CAAA;IAC9C,OAAO,CAAC,YAAY,CAAC,0BAA0B,EAAE,IAAI,CAAC,CAAA;IACtD,iBAAiB,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;IAC9B,OAAO,OAAO,CAAA;AAChB,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,MAAM,YAAY,GAAa;IACpC,KAAK,EAAE,OAAO;IACd,IAAI,EAAE,IAAI;IACV,UAAU,EAAE,IAAI;IAChB,SAAS,EAAE,IAAI;IACf,KAAK,EAAE;QACL,IAAI,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE;QACrB,GAAG,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE;KACxB;IACD,QAAQ,EAAE;QACR,GAAG,SAAS;QACZ,GAAG,yBAAyB;QAC5B,GAAG,iBAAiB;QACpB;YACE,GAAG,EAAE,GAAG;YACR,qEAAqE;YACrE,wDAAwD;YACxD,QAAQ,EAAE,CAAC;YACX,QAAQ,CAAC,GAAG;gBACV,MAAM,EAAE,GAAG,GAAc,CAAA;gBACzB,kEAAkE;gBAClE,kEAAkE;gBAClE,IAAI,uBAAuB,CAAC,EAAE,CAAC;oBAAE,OAAO,KAAK,CAAA;gBAC7C,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC,EAAE,GAAG,EAAE,EAAE,CAAC,QAAQ,CAAC,WAAW,EAAE,EAAE,CAAA;YAC5D,CAAC;SACF;KACF;IACD,KAAK,CAAC,IAAI;QACR,OAAO,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAW,EAAE,KAAK,CAAC,CAAA;IAC5D,CAAC;CACF,CAAA;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AACH,MAAM,cAAc,GAAG;IACrB,YAAY;IACZ,UAAU;IACV,UAAU;IACV,aAAa;IACb,YAAY;IACZ,aAAa;IACb,eAAe;IACf,aAAa;CACd,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;AAEX;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,CAAC,MAAM,aAAa,GAAwB,IAAI,GAAG,CAAC;IACxD,SAAS;IACT,SAAS;IACT,OAAO;IACP,YAAY;IACZ,QAAQ;IACR,SAAS;IACT,QAAQ;IACR,KAAK;IACL,KAAK;IACL,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,UAAU;IACV,YAAY;IACZ,QAAQ;IACR,QAAQ;IACR,QAAQ;IACR,QAAQ;IACR,MAAM;IACN,MAAM;IACN,KAAK;IACL,IAAI;IACJ,GAAG;IACH,QAAQ;IACR,SAAS;IACT,SAAS;IACT,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,IAAI;IACJ,SAAS;IACT,WAAW;IACX,KAAK;IACL,MAAM;IACN,IAAI;IACJ,QAAQ;IACR,OAAO;IACP,IAAI;IACJ,KAAK;IACL,QAAQ;IACR,SAAS;IACT,QAAQ;CACT,CAAC,CAAA;AAEF;;;;GAIG;AACH,MAAM,CAAC,MAAM,aAAa,GAAa;IACrC,KAAK,EAAE,QAAQ;IACf,MAAM,EAAE,IAAI;IACZ,IAAI,EAAE,IAAI;IACV,UAAU,EAAE,IAAI;IAChB,KAAK,EAAE;QACL,IAAI,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE;QACrB,GAAG,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE;KACzB;IACD,QAAQ,EAAE;QACR,GAAG,SAAS;QACZ,GAAG,yBAAyB;QAC5B,GAAG,iBAAiB;QACpB;YACE,GAAG,EAAE,GAAG;YACR,qEAAqE;YACrE,oEAAoE;YACpE,qBAAqB;YACrB,QAAQ,EAAE,CAAC;YACX,OAAO,EAAE,cAAc;YACvB,QAAQ,CAAC,GAAG;gBACV,MAAM,EAAE,GAAG,GAAc,CAAA;gBACzB,IAAI,uBAAuB,CAAC,EAAE,CAAC;oBAAE,OAAO,KAAK,CAAA;gBAC7C,sEAAsE;gBACtE,qEAAqE;gBACrE,kEAAkE;gBAClE,kEAAkE;gBAClE,oEAAoE;gBACpE,wEAAwE;gBACxE,gEAAgE;gBAChE,MAAM,GAAG,GAAG,EAAE,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAA;gBACrC,IAAI,GAAG,KAAK,YAAY,EAAE,CAAC;oBACzB,IAAI,EAAE,CAAC,aAAa,EAAE,QAAQ,CAAC,WAAW,EAAE,KAAK,GAAG;wBAAE,OAAO,KAAK,CAAA;gBACpE,CAAC;qBAAM,IAAI,aAAa,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;oBAClC,OAAO,KAAK,CAAA;gBACd,CAAC;gBACD,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC,EAAE,GAAG,EAAE,EAAE,CAAC,QAAQ,CAAC,WAAW,EAAE,EAAE,CAAA;YAC5D,CAAC;SACF;KACF;IACD,KAAK,CAAC,IAAI;QACR,OAAO,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAW,EAAE,MAAM,CAAC,CAAA;IAC7D,CAAC;CACF,CAAA"}

package/dist/schema.d.ts

@@ -8,6 +8,21 @@
  * there rather than at the package boundary.
  */
 import { Schema, type MarkSpec, type NodeSpec } from 'prosemirror-model';
+/**
+ * The `start` of an ordered list, or null when the attribute is not a number.
+ *
+ * `Number('abc')` is `NaN`, and `NaN` is a perfectly good value to store on a
+ * node -- so `<ol start="abc">` was stored as `NaN` and serialized as the string
+ * `start="NaN"`, turning a value the browser ignores into one it still ignores
+ * but that is now in the customer's database with our fingerprints on it.
+ * Returning null instead lets extensions.ts carry the original spelling as
+ * residue, so the attribute survives without being interpreted.
+ *
+ * Exported so that the carry mechanism asks the same question this does. Two
+ * copies of "is this a valid start" is how they would disagree and emit the
+ * attribute twice, or neither time.
+ */
+export declare function listStart(value: string | null | undefined): number | null;
 /** The base node specs. Extensions are appended to these. */
 export declare const coreNodes: Record<string, NodeSpec>;
 /** The base mark specs. */

package/dist/schema.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../src/schema.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,MAAM,EAAkC,KAAK,QAAQ,EAAE,KAAK,QAAQ,EAAE,MAAM,mBAAmB,CAAA;AAiJxG,6DAA6D;AAC7D,eAAO,MAAM,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,QAAQ,CA4L9C,CAAA;AAED,2BAA2B;AAC3B,eAAO,MAAM,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,QAAQ,CAsN9C,CAAA;AAED;;;;;;GAMG;AACH,eAAO,MAAM,UAAU,wBAAqD,CAAA"}
+{"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../src/schema.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,MAAM,EAAkC,KAAK,QAAQ,EAAE,KAAK,QAAQ,EAAE,MAAM,mBAAmB,CAAA;AAqHxG;;;;;;;;;;;;;GAaG;AACH,wBAAgB,SAAS,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,GAAG,MAAM,GAAG,IAAI,CAIzE;AA8BD,6DAA6D;AAC7D,eAAO,MAAM,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,QAAQ,CA4L9C,CAAA;AAgDD,2BAA2B;AAC3B,eAAO,MAAM,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,QAAQ,CAwQ9C,CAAA;AAED;;;;;;GAMG;AACH,eAAO,MAAM,UAAU,wBAAqD,CAAA"}

package/dist/schema.js

@@ -8,9 +8,9 @@
  * there rather than at the package boundary.
  */
 import { Schema } from 'prosemirror-model';
-import { applyStyleAttribute, indentCss, indentLevels, isFullyModelledStyle, parseDeclarations, safeAlign, safeColor, safeDir, safeFontFamily, safeFontSize, safeLang, safeLineHeight, safeListStyle, serializeDeclarations } from './css.js';
-import { serializationTarget, unknownBlock, unknownInline } from './preserve.js';
-import { audio, details, figcaption, figure, iframe, imageDomAttrs, imageParseAttrs, named_anchor, page_break, summary, video, } from './structure.js';
+import { INLINE_STYLE_PROPERTIES, applyStyleAttribute, indentCss, indentLevels, isFullyModelledStyle, modelledValue, parseDeclarations, safeAlign, safeColor, safeDir, safeFontFamily, safeFontSize, safeLang, safeLineHeight, safeListStyle, serializeDeclarations } from './css.js';
+import { isLosslesslyUnwrappable, serializationTarget, unknownBlock, unknownInline } from './preserve.js';
+import { audio, details, figcaption, figure, iframe, imageDomAttrs, imageParseAttrs, isEmptyNamedAnchorElement, named_anchor, page_break, summary, video, } from './structure.js';
 import { table, table_cell, table_header, table_row } from './tables.js';
 import { safeId } from './tokens.js';
 import { isSafeUrl } from './url.js';
@@ -108,6 +108,26 @@ function textBlockToDOM(tag, attrs) {
         return elementWithStyle(tag, domAttrs);
     return Object.keys(domAttrs).length > 0 ? [tag, domAttrs, 0] : [tag, 0];
 }
+/**
+ * The `start` of an ordered list, or null when the attribute is not a number.
+ *
+ * `Number('abc')` is `NaN`, and `NaN` is a perfectly good value to store on a
+ * node -- so `<ol start="abc">` was stored as `NaN` and serialized as the string
+ * `start="NaN"`, turning a value the browser ignores into one it still ignores
+ * but that is now in the customer's database with our fingerprints on it.
+ * Returning null instead lets extensions.ts carry the original spelling as
+ * residue, so the attribute survives without being interpreted.
+ *
+ * Exported so that the carry mechanism asks the same question this does. Two
+ * copies of "is this a valid start" is how they would disagree and emit the
+ * attribute twice, or neither time.
+ */
+export function listStart(value) {
+    if (value === null || value === undefined)
+        return null;
+    const parsed = Number.parseInt(value.trim(), 10);
+    return Number.isFinite(parsed) ? parsed : null;
+}
 function listAttrs(el, ordered) {
     const declarations = parseDeclarations(el.getAttribute('style'));
     const fromStyle = safeListStyle(declarations.get('list-style-type') ?? null);
@@ -115,8 +135,7 @@ function listAttrs(el, ordered) {
     const listStyle = fromStyle ?? fromType;
     if (!ordered)
         return { listStyle };
-    const start = el.getAttribute('start');
-    return { start: start ? Number(start) : 1, listStyle };
+    return { start: listStart(el.getAttribute('start')) ?? 1, listStyle };
 }
 function listToDOM(tag, attrs) {
     const out = {};
@@ -316,6 +335,55 @@ export const coreNodes = {
     unknown_block: unknownBlock,
     unknown_inline: unknownInline,
 };
+/**
+ * True when a `<span>` carries nothing but formatting the mark set can express.
+ *
+ * The question a TAG rule on a mark has to ask, and the reason it is not
+ * `isFullyModelledStyle` alone: `background` is the shorthand Word writes, and
+ * the rule below reads it through `el.style.backgroundColor`, so the CSSOM has
+ * already expanded it and it is genuinely modelled even though the declaration
+ * name is not in the modelled list. Refusing it would turn every highlighted
+ * Word paste into an opaque atom.
+ *
+ * Any other attribute, or any declaration a mark cannot hold, and the answer is
+ * no -- because a mark that claims the element has nowhere to keep the rest.
+ */
+function isSpanOnlyStyling(el) {
+    for (const attr of Array.from(el.attributes)) {
+        if (attr.name !== 'style')
+            return false;
+    }
+    const declarations = parseDeclarations(el.getAttribute('style'));
+    if (declarations.size === 0)
+        return false;
+    for (const [name, value] of declarations) {
+        if (name === 'background')
+            continue;
+        if (!INLINE_STYLE_PROPERTIES.includes(name))
+            return false;
+        if (modelledValue(name, value) === null)
+            return false;
+    }
+    return true;
+}
+/**
+ * True when a colour/font mark must not fire on this element.
+ *
+ * Those marks used to be `style:` rules. A style rule does not consume the
+ * element, which is why `<p style="color:red">` and a `<div>` wrapper still
+ * yield a mark -- and why a `<span style="color:red" class="hl">` yielded
+ * BOTH the mark and a preserved atom, nested. The preservation catch-all
+ * keeps any span `isLosslesslyUnwrappable` refuses, so the mark has to
+ * refuse the same elements.
+ *
+ * The replacement is a `tag: '*'` rule with `consuming: false`: same
+ * "applies to whoever carries the declaration" behaviour, but `getAttrs`
+ * receives the element and can decline. Hosts the schema models are not
+ * spans the catch-all would keep, so they still produce the mark.
+ */
+function declineStyleMarkOnPreservedAtom(el) {
+    return el.nodeName.toLowerCase() === 'span' && !isLosslesslyUnwrappable(el);
+}
 /** The base mark specs. */
 export const coreMarks = {
     strong: {
@@ -351,26 +419,36 @@ export const coreMarks = {
      * would have the second command overwrite the first's value with a default.
      * Two marks compose the way the toolbar's two controls do.
      *
-     * Both are parsed through a `style` rule, not a tag rule, and that choice is
-     * load-bearing. ProseMirror applies style rules to whatever element carries
-     * the declaration, so `<span style="color:red">`, `<p style="color:red">` and
-     * a `<div>` wrapper all yield the same mark -- whereas a `span[style]` tag rule
-     * would claim the span, drop it, and quietly lose any other declaration on it.
+     * Parsed through a `tag: '*'` rule that does not consume the element, not a
+     * `span[style]` tag rule. The un-consuming match is load-bearing: it is how
+     * `<span style="color:red">`, `<p style="color:red">` and a `<div>` wrapper
+     * all yield the same mark -- whereas a `span[style]` rule would claim the
+     * span, drop it, and quietly lose any other declaration on it.
+     *
+     * It is not a `style:` rule either. Those also do not consume, but they
+     * never see the element, so they cannot decline when preservation is going
+     * to keep it. `declineStyleMarkOnPreservedAtom` is that decline.
      *
      * The value is normalized by `safeColor`, which folds `rgb(255, 0, 0)` back to
-     * `#ff0000`. That is not cosmetic: ProseMirror reads style rules through the
-     * CSSOM, which returns the functional form for an authored hex colour, so
-     * without the fold every hex colour in an archive would be rewritten longer on
-     * the first save.
+     * `#ff0000`. That is not cosmetic: the rule still reads through the CSSOM
+     * (`el.style.color`), which returns the functional form for an authored hex
+     * colour, so without the fold every hex colour in an archive would be
+     * rewritten longer on the first save.
      */
     text_color: {
         attrs: { color: {} },
         parseDOM: [
             {
-                style: 'color',
-                getAttrs(value) {
-                    const color = safeColor(value);
-                    return color ? { color } : false;
+                tag: '*',
+                consuming: false,
+                getAttrs(dom) {
+                    const el = dom;
+                    const color = safeColor(el.style.color);
+                    if (!color)
+                        return false;
+                    if (declineStyleMarkOnPreservedAtom(el))
+                        return false;
+                    return { color };
                 },
             },
             {
@@ -403,6 +481,17 @@ export const coreMarks = {
                 tag: 'span',
                 getAttrs(dom) {
                     const el = dom;
+                    // Only when the span is nothing but formatting this mark can hold.
+                    // A tag rule CONSUMES the element. Character marks now carry leftover
+                    // attributes the way nodes do, but this rule still declines a span
+                    // that is more than a highlight: claiming
+                    // `<span class="hl" style="background-color:#ffff00;letter-spacing:1px">`
+                    // would swallow declarations other style-marks and the preservation
+                    // layer are meant to see. Declining leaves the element whole. This is
+                    // the same bargain `font[color]` and `span[lang]` already strike a
+                    // few lines away.
+                    if (!isSpanOnlyStyling(el))
+                        return false;
                     const fromAttr = parseDeclarations(el.getAttribute('style')).get('background-color');
                     const color = safeColor(el.style.backgroundColor) ?? safeColor(fromAttr);
                     return color ? { color } : false;
@@ -415,10 +504,16 @@ export const coreMarks = {
         attrs: { family: {} },
         parseDOM: [
             {
-                style: 'font-family',
-                getAttrs(value) {
-                    const family = safeFontFamily(value);
-                    return family ? { family } : false;
+                tag: '*',
+                consuming: false,
+                getAttrs(dom) {
+                    const el = dom;
+                    const family = safeFontFamily(el.style.fontFamily);
+                    if (!family)
+                        return false;
+                    if (declineStyleMarkOnPreservedAtom(el))
+                        return false;
+                    return { family };
                 },
             },
             {
@@ -440,10 +535,16 @@ export const coreMarks = {
         attrs: { size: {} },
         parseDOM: [
             {
-                style: 'font-size',
-                getAttrs(value) {
-                    const size = safeFontSize(value);
-                    return size ? { size } : false;
+                tag: '*',
+                consuming: false,
+                getAttrs(dom) {
+                    const el = dom;
+                    const size = safeFontSize(el.style.fontSize);
+                    if (!size)
+                        return false;
+                    if (declineStyleMarkOnPreservedAtom(el))
+                        return false;
+                    return { size };
                 },
             },
         ],
@@ -460,6 +561,8 @@ export const coreMarks = {
         toDOM: () => ['sup', 0],
     },
     language: {
+        // Content, not decoration -- the same claim `dir` makes as a node
+        // attribute. `clearFormatting` keeps this mark for that reason.
         attrs: { lang: {} },
         inclusive: false,
         parseDOM: [
@@ -485,7 +588,9 @@ export const coreMarks = {
     },
     link: {
         attrs: {
-            href: {},
+            // Null when the mark is a wrapped named destination (`<a id>` with
+            // text, no href). Empty `<a id></a>` is the `named_anchor` atom.
+            href: { default: null },
             title: { default: null },
             target: { default: null },
             rel: { default: null },
@@ -511,10 +616,34 @@ export const coreMarks = {
                     };
                 },
             },
+            {
+                tag: 'a[id]',
+                getAttrs(dom) {
+                    const el = dom;
+                    if (el.hasAttribute('href'))
+                        return false;
+                    const id = safeId(el.getAttribute('id'));
+                    if (!id)
+                        return false;
+                    // Empty `<a id>` belongs to `named_anchor`. This rule is the
+                    // wrapped-text spelling, so the heading stays in the document.
+                    if (isEmptyNamedAnchorElement(el))
+                        return false;
+                    return {
+                        href: null,
+                        title: el.getAttribute('title'),
+                        target: el.getAttribute('target'),
+                        rel: el.getAttribute('rel'),
+                        id,
+                    };
+                },
+            },
         ],
         toDOM(node) {
             const { href, title, target, rel, id } = node.attrs;
-            const attrs = { href: href };
+            const attrs = {};
+            if (href !== null)
+                attrs['href'] = href;
             if (title !== null)
                 attrs['title'] = title;
             if (target !== null)