@readme/markdown 14.4.0 → 14.4.1

package/dist/main.js

@@ -11594,6 +11594,7 @@ __webpack_require__.r(__webpack_exports__);
 __webpack_require__.d(__webpack_exports__, {
   Components: () => (/* reexport */ components_namespaceObject),
   FLOW_TYPES: () => (/* reexport */ FLOW_TYPES),
+  INLINE_ONLY_PARENT_TYPES: () => (/* reexport */ INLINE_ONLY_PARENT_TYPES),
   NodeTypes: () => (/* reexport */ NodeTypes),
   Owlmoji: () => (/* reexport */ Owlmoji),
   compile: () => (/* reexport */ lib_compile),
@@ -13061,6 +13062,34 @@ const FLOW_TYPES = new Set([
     'table',
     'thematicBreak',
 ]);
+/**
+ * MDAST parent types whose children are restricted to inline content. A block-level
+ * node placed inside any of these violates the parent's content model and breaks
+ * downstream consumers
+ *
+ * Derived from the mdast spec. Seven of these are direct "phrasing content" parents:
+ *   - paragraph, heading, link, linkReference, emphasis, strong
+ *   - tableCell (phrasing minus Break)
+ *
+ * @link {https://github.com/syntax-tree/mdast}
+ *
+ * `delete` is GFM and has *transparent* content, see:
+ * @link {https://github.com/syntax-tree/mdast-util-gfm-strikethrough}
+ *
+ * its children must also be phrasing, so it belongs here too.
+ *
+ * NOTE: exported so downstream consumers can reuse and not drift away
+ */
+const INLINE_ONLY_PARENT_TYPES = new Set([
+    'paragraph',
+    'heading',
+    'tableCell',
+    'link',
+    'linkReference',
+    'emphasis',
+    'strong',
+    'delete',
+]);
 // HELPER CONSTANTS FOR MDXISH RENDERING
 /**
  * Inline-only custom components that appear as phrasing content within
@@ -75543,7 +75572,15 @@ const walkTags = (html, handlers) => {
     const tagEnd = (parser) => (parser.endIndex ?? parser.startIndex) + 1;
     const parser = new Parser_Parser({
         onopentag(name) {
-            handlers.onOpen?.({ name, start: parser.startIndex, end: tagEnd(parser) });
+            const start = parser.startIndex;
+            const end = tagEnd(parser);
+            handlers.onOpen?.({
+                name,
+                start,
+                end,
+                isSelfClosing: html[end - 2] === '/',
+                isStrayCloser: html[start + 1] === '/',
+            });
         },
         onclosetag(name, implicit) {
             handlers.onClose?.({ name, start: parser.startIndex, end: tagEnd(parser), implicit });
@@ -75912,13 +75949,6 @@ const isStandardHtmlTag = (name) => STANDARD_HTML_TAGS.has(name.toLowerCase());
 // (htmlparser2) handles attribute parsing; this scan only needs to locate
 // orphan closers, which can't appear inside an attribute value anyway.
 const HTML_TAG_TOKEN_RE = /<(\/)?([a-zA-Z][a-zA-Z0-9-]*)\b[^>]*>/g;
-/**
- * htmlparser2 silently drops closing tags that have no matching opener
- * (e.g. the trailing `</li>` in `<ul><li>x</ul></li>`), leaving them in the
- * source makes mdxjs choke on the dangling closer. Scan the masked
- * source for `</name>` tokens that don't pair with any prior unmatched
- * `<name>` and return their spans so the caller can drop them.
- */
 const findOrphanClosers = (html) => {
     const masked = maskNonTagRegions(html);
     const stack = [];
@@ -75927,18 +75957,30 @@ const findOrphanClosers = (html) => {
     let match;
     while ((match = HTML_TAG_TOKEN_RE.exec(masked)) !== null) {
         const name = match[2].toLowerCase();
-        // Non-HTML names and void elements are handled by the main walker.
         // eslint-disable-next-line no-continue
-        if (!isStandardHtmlTag(name) || HTML_VOID_ELEMENTS.has(name))
+        if (!isStandardHtmlTag(name))
+            continue;
+        // Special case for <br>: htmlparser2 will normalize orphan </br> to <br>
+        // so we don't need to handle it here.
+        // eslint-disable-next-line no-continue
+        if (name === 'br' && match[1] === '/')
             continue;
+        const isVoid = HTML_VOID_ELEMENTS.has(name);
         if (match[1] === '/') {
+            // Other void closers (`</hr>`, `</img>`, ...) have no HTML5 rewrite rule
+            // and must be stripped before the strict mdxjs parse runs.
+            if (isVoid) {
+                orphans.push({ offset: match.index, length: match[0].length });
+                // eslint-disable-next-line no-continue
+                continue;
+            }
             const idx = stack.lastIndexOf(name);
             if (idx === -1)
                 orphans.push({ offset: match.index, length: match[0].length });
             else
                 stack.length = idx;
         }
-        else if (!match[0].endsWith('/>')) {
+        else if (!isVoid && !match[0].endsWith('/>')) {
             stack.push(name);
         }
     }
@@ -75971,7 +76013,7 @@ const repairUnclosedTags = (html) => {
     const inserts = [];
     const openTags = [];
     walkTags(html, {
-        onOpen({ name, start, end }) {
+        onOpen({ name, start, end, isSelfClosing, isStrayCloser }) {
             // Escape non-HTML names (custom components, typos, `<arbitrary-tag>`)
             // so MDX treats them as literal text instead of expecting a closer
             if (!isStandardHtmlTag(name)) {
@@ -75979,12 +76021,19 @@ const repairUnclosedTags = (html) => {
                 return;
             }
             if (HTML_VOID_ELEMENTS.has(name.toLowerCase())) {
-                // MDX requires void elements to be self-closing (`<br/>`, not `<br>`).
-                // If the source open tag doesn't end with `/`, inject one before the
-                // `>` so it parses. `end` is one past `>`, so `end - 2` is the char
-                // immediately before `>`.
-                if (html[end - 2] !== '/')
+                // MDX requires void elements to be self-closing (`<br/>`, not `<br>`)
+                // so we need to rewrite them to self closing tags.
+                if (isStrayCloser) {
+                    // htmlparser2 may normalize some stray closers like `</br>` into opener
+                    // events <br> per the HTML5 spec. In this case remove the closing /
+                    // and fully rewrite the tag to self closing.
+                    inserts.push({ offset: start, text: `<${name}/>`, consumes: end - start });
+                    return;
+                }
+                else if (!isSelfClosing) {
+                    // Slots in the / right before the closing >
                     inserts.push({ offset: end - 1, text: '/' });
+                }
                 return;
             }
             openTags.push({ name, start, end });
@@ -103402,6 +103451,7 @@ const mdxishHtmlBlocks = () => tree => {
 
 
 
+
 function toImageAlign(value) {
     if (value === 'left' || value === 'center' || value === 'right') {
         return value;
@@ -103954,16 +104004,12 @@ const mdxishJsxToMdast = () => tree => {
             parent.children[index] = newNode;
         }
     });
-    // Transform magic block images (type: 'image') to image-block
-    // Images inside paragraphs are standard markdown — handled by imageTransformer, normalized below
+    // Promote magic-block images (type: 'image') to image-block, except inside inline-only
+    // parents where the image must stay inline (authors use `<Image caption="…" />` instead).
     visit(tree, 'image', (node, index, parent) => {
         if (!parent || index === undefined)
             return SKIP;
-        if (parent.type === 'paragraph')
-            return SKIP;
-        // `![](url)` in any tableCell stays inline. Authors who want a captioned figure use
-        // `<Image caption="…" />` JSX, which becomes `image-block` via `COMPONENT_MAP` above.
-        if (parent.type === 'tableCell')
+        if (INLINE_ONLY_PARENT_TYPES.has(parent.type))
             return SKIP;
         const newNode = transformMagicBlockImage(node);
         parent.children[index] = newNode;

package/dist/main.node.js

@@ -19278,6 +19278,7 @@ __webpack_require__.r(__webpack_exports__);
 __webpack_require__.d(__webpack_exports__, {
   Components: () => (/* reexport */ components_namespaceObject),
   FLOW_TYPES: () => (/* reexport */ FLOW_TYPES),
+  INLINE_ONLY_PARENT_TYPES: () => (/* reexport */ INLINE_ONLY_PARENT_TYPES),
   NodeTypes: () => (/* reexport */ NodeTypes),
   Owlmoji: () => (/* reexport */ Owlmoji),
   compile: () => (/* reexport */ lib_compile),
@@ -25634,6 +25635,34 @@ const FLOW_TYPES = new Set([
     'table',
     'thematicBreak',
 ]);
+/**
+ * MDAST parent types whose children are restricted to inline content. A block-level
+ * node placed inside any of these violates the parent's content model and breaks
+ * downstream consumers
+ *
+ * Derived from the mdast spec. Seven of these are direct "phrasing content" parents:
+ *   - paragraph, heading, link, linkReference, emphasis, strong
+ *   - tableCell (phrasing minus Break)
+ *
+ * @link {https://github.com/syntax-tree/mdast}
+ *
+ * `delete` is GFM and has *transparent* content, see:
+ * @link {https://github.com/syntax-tree/mdast-util-gfm-strikethrough}
+ *
+ * its children must also be phrasing, so it belongs here too.
+ *
+ * NOTE: exported so downstream consumers can reuse and not drift away
+ */
+const INLINE_ONLY_PARENT_TYPES = new Set([
+    'paragraph',
+    'heading',
+    'tableCell',
+    'link',
+    'linkReference',
+    'emphasis',
+    'strong',
+    'delete',
+]);
 // HELPER CONSTANTS FOR MDXISH RENDERING
 /**
  * Inline-only custom components that appear as phrasing content within
@@ -95767,7 +95796,15 @@ const walkTags = (html, handlers) => {
     const tagEnd = (parser) => (parser.endIndex ?? parser.startIndex) + 1;
     const parser = new Parser_Parser({
         onopentag(name) {
-            handlers.onOpen?.({ name, start: parser.startIndex, end: tagEnd(parser) });
+            const start = parser.startIndex;
+            const end = tagEnd(parser);
+            handlers.onOpen?.({
+                name,
+                start,
+                end,
+                isSelfClosing: html[end - 2] === '/',
+                isStrayCloser: html[start + 1] === '/',
+            });
         },
         onclosetag(name, implicit) {
             handlers.onClose?.({ name, start: parser.startIndex, end: tagEnd(parser), implicit });
@@ -96136,13 +96173,6 @@ const isStandardHtmlTag = (name) => STANDARD_HTML_TAGS.has(name.toLowerCase());
 // (htmlparser2) handles attribute parsing; this scan only needs to locate
 // orphan closers, which can't appear inside an attribute value anyway.
 const HTML_TAG_TOKEN_RE = /<(\/)?([a-zA-Z][a-zA-Z0-9-]*)\b[^>]*>/g;
-/**
- * htmlparser2 silently drops closing tags that have no matching opener
- * (e.g. the trailing `</li>` in `<ul><li>x</ul></li>`), leaving them in the
- * source makes mdxjs choke on the dangling closer. Scan the masked
- * source for `</name>` tokens that don't pair with any prior unmatched
- * `<name>` and return their spans so the caller can drop them.
- */
 const findOrphanClosers = (html) => {
     const masked = maskNonTagRegions(html);
     const stack = [];
@@ -96151,18 +96181,30 @@ const findOrphanClosers = (html) => {
     let match;
     while ((match = HTML_TAG_TOKEN_RE.exec(masked)) !== null) {
         const name = match[2].toLowerCase();
-        // Non-HTML names and void elements are handled by the main walker.
         // eslint-disable-next-line no-continue
-        if (!isStandardHtmlTag(name) || HTML_VOID_ELEMENTS.has(name))
+        if (!isStandardHtmlTag(name))
+            continue;
+        // Special case for <br>: htmlparser2 will normalize orphan </br> to <br>
+        // so we don't need to handle it here.
+        // eslint-disable-next-line no-continue
+        if (name === 'br' && match[1] === '/')
             continue;
+        const isVoid = HTML_VOID_ELEMENTS.has(name);
         if (match[1] === '/') {
+            // Other void closers (`</hr>`, `</img>`, ...) have no HTML5 rewrite rule
+            // and must be stripped before the strict mdxjs parse runs.
+            if (isVoid) {
+                orphans.push({ offset: match.index, length: match[0].length });
+                // eslint-disable-next-line no-continue
+                continue;
+            }
             const idx = stack.lastIndexOf(name);
             if (idx === -1)
                 orphans.push({ offset: match.index, length: match[0].length });
             else
                 stack.length = idx;
         }
-        else if (!match[0].endsWith('/>')) {
+        else if (!isVoid && !match[0].endsWith('/>')) {
             stack.push(name);
         }
     }
@@ -96195,7 +96237,7 @@ const repairUnclosedTags = (html) => {
     const inserts = [];
     const openTags = [];
     walkTags(html, {
-        onOpen({ name, start, end }) {
+        onOpen({ name, start, end, isSelfClosing, isStrayCloser }) {
             // Escape non-HTML names (custom components, typos, `<arbitrary-tag>`)
             // so MDX treats them as literal text instead of expecting a closer
             if (!isStandardHtmlTag(name)) {
@@ -96203,12 +96245,19 @@ const repairUnclosedTags = (html) => {
                 return;
             }
             if (HTML_VOID_ELEMENTS.has(name.toLowerCase())) {
-                // MDX requires void elements to be self-closing (`<br/>`, not `<br>`).
-                // If the source open tag doesn't end with `/`, inject one before the
-                // `>` so it parses. `end` is one past `>`, so `end - 2` is the char
-                // immediately before `>`.
-                if (html[end - 2] !== '/')
+                // MDX requires void elements to be self-closing (`<br/>`, not `<br>`)
+                // so we need to rewrite them to self closing tags.
+                if (isStrayCloser) {
+                    // htmlparser2 may normalize some stray closers like `</br>` into opener
+                    // events <br> per the HTML5 spec. In this case remove the closing /
+                    // and fully rewrite the tag to self closing.
+                    inserts.push({ offset: start, text: `<${name}/>`, consumes: end - start });
+                    return;
+                }
+                else if (!isSelfClosing) {
+                    // Slots in the / right before the closing >
                     inserts.push({ offset: end - 1, text: '/' });
+                }
                 return;
             }
             openTags.push({ name, start, end });
@@ -123626,6 +123675,7 @@ const mdxishHtmlBlocks = () => tree => {
 
 
 
+
 function toImageAlign(value) {
     if (value === 'left' || value === 'center' || value === 'right') {
         return value;
@@ -124178,16 +124228,12 @@ const mdxishJsxToMdast = () => tree => {
             parent.children[index] = newNode;
         }
     });
-    // Transform magic block images (type: 'image') to image-block
-    // Images inside paragraphs are standard markdown — handled by imageTransformer, normalized below
+    // Promote magic-block images (type: 'image') to image-block, except inside inline-only
+    // parents where the image must stay inline (authors use `<Image caption="…" />` instead).
     visit(tree, 'image', (node, index, parent) => {
         if (!parent || index === undefined)
             return SKIP;
-        if (parent.type === 'paragraph')
-            return SKIP;
-        // `![](url)` in any tableCell stays inline. Authors who want a captioned figure use
-        // `<Image caption="…" />` JSX, which becomes `image-block` via `COMPONENT_MAP` above.
-        if (parent.type === 'tableCell')
+        if (INLINE_ONLY_PARENT_TYPES.has(parent.type))
             return SKIP;
         const newNode = transformMagicBlockImage(node);
         parent.children[index] = newNode;