@readme/markdown 14.6.0 → 14.7.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/main.node.js CHANGED
@@ -24899,11 +24899,20 @@ const Image = (Props) => {
24899
24899
  lightboxOverlay));
24900
24900
  }
24901
24901
  if (children || caption) {
24902
- return (external_react_.createElement("figure", null,
24903
- closedLightbox(alt || 'Expand image', external_react_.createElement(external_react_.Fragment, null,
24904
- imgElement,
24905
- external_react_.createElement("figcaption", null, children || caption))),
24906
- lightboxOverlay));
24902
+ // Mirrors the framed pattern: left/right captioned figures float and shrink
24903
+ // to fit so a long caption doesn't widen the float past the image.
24904
+ const isFloating = (align === 'left' || align === 'right') && !noWrap;
24905
+ const figureClass = [
24906
+ (align === 'left' || align === 'right') && `img-figure-${align}`,
24907
+ noWrap && 'img-no-wrap',
24908
+ ]
24909
+ .filter(Boolean)
24910
+ .join(' ');
24911
+ const figureStyle = isFloating && typeof width === 'string' && width.endsWith('%') ? { width } : undefined;
24912
+ return (external_react_.createElement("figure", { className: figureClass || undefined, style: figureStyle },
24913
+ closedLightbox(alt || 'Expand image', imgElement),
24914
+ lightboxOverlay,
24915
+ external_react_.createElement("figcaption", null, children || caption)));
24907
24916
  }
24908
24917
  return (external_react_.createElement(external_react_.Fragment, null,
24909
24918
  closedLightbox('Expand image', imgElement),
@@ -25682,6 +25691,7 @@ const INLINE_COMPONENT_TAGS = new Set(['Anchor', 'Glossary']);
25682
25691
  /**
25683
25692
  * PascalCase tags that have their own dedicated tokenizer / transformer
25684
25693
  * and must not be claimed by the generic `mdxComponent` construct.
25694
+ * Subject to change as we add more dedicated tokenizers.
25685
25695
  */
25686
25696
  const DEDICATED_COMPONENT_TAGS = ['HTMLBlock', 'Table'];
25687
25697
  /**
@@ -25692,6 +25702,14 @@ const GENERIC_MDX_COMPONENT_EXCLUDED_TAGS = new Set([
25692
25702
  ...DEDICATED_COMPONENT_TAGS,
25693
25703
  ...INLINE_COMPONENT_TAGS,
25694
25704
  ]);
25705
+ /**
25706
+ * Tags the micromark `mdxComponent` tokenizer must not claim, which
25707
+ * are inline components and those that have their own dedicated tokenizer
25708
+ */
25709
+ const TOKENIZER_MDX_COMPONENT_EXCLUDED_TAGS = new Set([
25710
+ 'Table',
25711
+ ...INLINE_COMPONENT_TAGS,
25712
+ ]);
25695
25713
  /**
25696
25714
  * Lowercased variant of {@link INLINE_COMPONENT_TAGS} for consumers that
25697
25715
  * run after rehype (where hast `tagName` is normalized to lowercase).
@@ -74737,18 +74755,28 @@ const isMDXEsm = (node) => {
74737
74755
  * Takes an HTML string and formats it for display in the editor. Removes leading/trailing newlines
74738
74756
  * and unindents the HTML.
74739
74757
  *
74740
- * @param {string} html - HTML content from template literal
74758
+ * @param {string} html - cooked HTML payload (callers strip any template-literal backticks first)
74759
+ * @param {number} [openingTagIndent=0] - column the `<HTMLBlock>` opening tag sits at, used to
74760
+ * dedent each content line so its indentation reads relative to the tag, not the line start
74741
74761
  * @returns {string} processed HTML
74742
74762
  */
74743
- function formatHtmlForMdxish(html) {
74744
- // Remove leading/trailing backticks if present, since they're used to keep the HTML
74745
- // from being parsed prematurely
74746
- let processed = html;
74747
- if (processed.startsWith('`') && processed.endsWith('`')) {
74748
- processed = processed.slice(1, -1);
74749
- }
74763
+ function formatHtmlForMdxish(html, openingTagIndent = 0) {
74750
74764
  // Removes the leading/trailing newlines
74751
- let cleaned = processed.replace(/^\s*\n|\n\s*$/g, '');
74765
+ let cleaned = html.replace(/^\s*\n|\n\s*$/g, '');
74766
+ // Strip / deindent the lines in the HTML string so that the indents are relative
74767
+ // to the opening HTMLBlock tag, not the literal line start
74768
+ // Keep any deeper indent
74769
+ if (openingTagIndent > 0) {
74770
+ cleaned = cleaned
74771
+ .split('\n')
74772
+ .map(line => {
74773
+ let i = 0;
74774
+ while (i < openingTagIndent && (line[i] === ' ' || line[i] === '\t'))
74775
+ i += 1;
74776
+ return line.slice(i);
74777
+ })
74778
+ .join('\n');
74779
+ }
74752
74780
  // Convert literal \n sequences to actual newlines only inside <pre> and <code>.
74753
74781
  // Because <pre> needs to respect the newline visual and
74754
74782
  // escape characters should be processed in the <code> tag.
@@ -95905,11 +95933,15 @@ const symmetrizePair = (html, { openStart, openEnd, closeStart, closeEnd }) => {
95905
95933
  const postCloser = html.slice(closeEnd, closeLine.end);
95906
95934
  const openerHasExtras = preOpener.trim().length > 0 || postOpener.trim().length > 0;
95907
95935
  const closerHasExtras = preCloser.trim().length > 0 || postCloser.trim().length > 0;
95908
- // Both match (both bare or both attached) — mdxjs parses this fine.
95909
- if (openerHasExtras === closerHasExtras)
95936
+ // A blank line splits opener/closer into separate paragraphs.
95937
+ // If either tag is attached to surrounding text, mdxjs can fail to match.
95938
+ const spansBlankLine = /\n[^\S\n]*\n/.test(html.slice(openEnd, closeStart));
95939
+ // If both sides are already symmetric, keep as-is.
95940
+ // Exception: attached + blank-line-split pairs still need normalization.
95941
+ if (openerHasExtras === closerHasExtras && !(openerHasExtras && spansBlankLine))
95910
95942
  return [];
95911
- // Asymmetric. Push non-tag content on the offending side to its own line,
95912
- // visually aligning the inserted line with the existing whitespace prefix.
95943
+ // For asymmetric (or attached-but-split) pairs, move side content to its own
95944
+ // line so opener/closer become line-symmetric.
95913
95945
  const indentFor = (linePrefix) => linePrefix.match(/^\s*/)?.[0] ?? '';
95914
95946
  const inserts = [];
95915
95947
  if (openerHasExtras) {
@@ -96355,6 +96387,7 @@ const repairUnclosedTags = (html) => {
96355
96387
 
96356
96388
 
96357
96389
 
96390
+
96358
96391
 
96359
96392
 
96360
96393
  const isTableCell = (node) => isMDXElement(node) && ['th', 'td'].includes(node.name);
@@ -96455,6 +96488,15 @@ const processTableNode = (node, index, parent, documentPosition) => {
96455
96488
  const { align: alignAttr } = getAttrs(node);
96456
96489
  const align = Array.isArray(alignAttr) ? alignAttr : null;
96457
96490
  let tableHasFlowContent = false;
96491
+ // An `<HTMLBlock>` (still a JSX element here; converted to `html-block` by
96492
+ // `mdxishHtmlBlocks` after this transformer) is block-level content that a
96493
+ // markdown table cell can't represent, so keep the table as a JSX `<Table>`.
96494
+ visit(node, candidate => candidate.type === NodeTypes.htmlBlock ||
96495
+ ((candidate.type === 'mdxJsxFlowElement' || candidate.type === 'mdxJsxTextElement') &&
96496
+ candidate.name === 'HTMLBlock'), () => {
96497
+ tableHasFlowContent = true;
96498
+ return EXIT;
96499
+ });
96458
96500
  // Re-parse text-only cells through markdown and detect flow content
96459
96501
  visit(node, isTableCell, (cell) => {
96460
96502
  if (!isTextOnly(cell.children))
@@ -96537,13 +96579,32 @@ const processTableNode = (node, index, parent, documentPosition) => {
96537
96579
  // remarkMdx wraps inline `<tr>`s in a paragraph; unwrap one level so the
96538
96580
  // hasRow check below sees them.
96539
96581
  const flattenSectionChildren = (nodes) => nodes.flatMap(n => n.type === 'paragraph' && 'children' in n && Array.isArray(n.children) ? n.children : [n]);
96540
- visit(node, isMDXElement, (child) => {
96541
- if (child.name !== 'thead' && child.name !== 'tbody')
96582
+ // Iterate the table's direct children in document order. Rows may live
96583
+ // inside a `<thead>`/`<tbody>` section, or sit bare directly under the table
96584
+ // with no section wrapper — both are collected here so the first row becomes
96585
+ // the markdown table header.
96586
+ const tableChildren = flattenSectionChildren(node.children);
96587
+ tableChildren.forEach(child => {
96588
+ if (!isMDXElement(child))
96589
+ return;
96590
+ const childElement = child;
96591
+ // Path for a `<tr>` directly under the table, without a `<thead>`/`<tbody>` wrapper.
96592
+ if (childElement.name === 'tr') {
96593
+ children.push({
96594
+ type: 'tableRow',
96595
+ children: collectCells(childElement),
96596
+ position: childElement.position,
96597
+ });
96598
+ return;
96599
+ }
96600
+ // Path for when the rows are wrapped in a `<thead>`/`<tbody>`
96601
+ // We visit & collect the entire rows under them directly here
96602
+ if (childElement.name !== 'thead' && childElement.name !== 'tbody')
96542
96603
  return;
96543
- const sectionChildren = flattenSectionChildren(child.children);
96604
+ const sectionChildren = flattenSectionChildren(childElement.children);
96544
96605
  const hasRow = sectionChildren.some(c => isMDXElement(c) && c.name === 'tr');
96545
96606
  if (hasRow) {
96546
- visit(child, isMDXElement, (row) => {
96607
+ visit(childElement, isMDXElement, (row) => {
96547
96608
  if (row.name !== 'tr')
96548
96609
  return;
96549
96610
  children.push({
@@ -96557,7 +96618,7 @@ const processTableNode = (node, index, parent, documentPosition) => {
96557
96618
  // No `<tr>`, chunk bare cells into rows using the prior row's column
96558
96619
  // count (e.g. from `<thead>`), so 4 bare `<td>`s under a 2-col header
96559
96620
  // become 2 rows of 2.
96560
- const cells = collectCells(child);
96621
+ const cells = collectCells(childElement);
96561
96622
  if (cells.length === 0)
96562
96623
  return;
96563
96624
  const cols = children[0]?.children?.length || cells.length;
@@ -96565,11 +96626,12 @@ const processTableNode = (node, index, parent, documentPosition) => {
96565
96626
  children.push({
96566
96627
  type: 'tableRow',
96567
96628
  children: cells.slice(i, i + cols),
96568
- position: child.position,
96629
+ position: childElement.position,
96569
96630
  });
96570
96631
  }
96571
96632
  }
96572
96633
  });
96634
+ // Output the markdown table node
96573
96635
  const firstRow = children[0];
96574
96636
  const columnCount = firstRow?.children?.length || 0;
96575
96637
  const alignArray = align && columnCount > 0
@@ -121095,7 +121157,7 @@ function createTokenize(mode) {
121095
121157
  return tagNameRest;
121096
121158
  }
121097
121159
  // Tag name complete — check exclusions
121098
- if (GENERIC_MDX_COMPONENT_EXCLUDED_TAGS.has(tagName)) {
121160
+ if (TOKENIZER_MDX_COMPONENT_EXCLUDED_TAGS.has(tagName)) {
121099
121161
  return nok(code);
121100
121162
  }
121101
121163
  depth = 1;
@@ -123186,557 +123248,158 @@ const magicBlockTransformer = (options = {}) => tree => {
123186
123248
  };
123187
123249
  /* harmony default export */ const magic_block_transformer = (magicBlockTransformer);
123188
123250
 
123189
- ;// ./lib/micromark/jsx-comment/pattern.ts
123190
- /**
123191
- * Matches a JSX comment: `{/*`, content, `*\/}` — no whitespace tolerated
123192
- * between the braces and the comment markers.
123193
- *
123194
- * This grammar is mirrored by the flow tokenizer in ./syntax.ts. Any change
123195
- * here needs a mirror change in the state machine; the parity test in
123196
- * __tests__/lib/micromark/jsx-comment-pattern-parity.test.ts locks the two
123197
- * together so they can't silently drift.
123198
- */
123199
- const JSX_COMMENT_REGEX = /\{\/\*[^*]*(?:\*(?!\/)[^*]*)*\*\/\}/g;
123251
+ ;// ./processor/transform/mdxish/mdxish-html-blocks.ts
123200
123252
 
123201
- ;// ./processor/transform/mdxish/preprocess-jsx-expressions.ts
123202
123253
 
123203
123254
 
123204
- // Base64 encode (Node.js + browser compatible)
123205
- function base64Encode(str) {
123206
- if (typeof Buffer !== 'undefined') {
123207
- return Buffer.from(str, 'utf-8').toString('base64');
123208
- }
123209
- return btoa(unescape(encodeURIComponent(str)));
123210
- }
123211
- // Base64 decode (Node.js + browser compatible)
123212
- function base64Decode(str) {
123213
- if (typeof Buffer !== 'undefined') {
123214
- return Buffer.from(str, 'base64').toString('utf-8');
123215
- }
123216
- return decodeURIComponent(escape(atob(str)));
123217
- }
123218
- // Markers for protected HTMLBlock content (HTML comments avoid markdown parsing issues)
123219
- const HTML_BLOCK_CONTENT_START = '<!--RDMX_HTMLBLOCK:';
123220
- const HTML_BLOCK_CONTENT_END = ':RDMX_HTMLBLOCK-->';
123255
+ // `<HTMLBlock …>{`…`}</HTMLBlock>` embedded inside a raw HTML block (e.g. a
123256
+ // single-line `<div>…</div>`). CommonMark slurps the whole div as one `html`
123257
+ // node, so the tokenizer never sees the HTMLBlock — we recover it here.
123258
+ const RAW_HTML_BLOCK_RE = /<HTMLBlock\b([^>]*)>\s*\{\s*`((?:[^`\\]|\\.)*)`\s*\}\s*<\/HTMLBlock>/g;
123259
+ // Opening `<HTMLBlock …>` as its own `html` node — produced inside a paragraph
123260
+ // when an HTMLBlock appears inline alongside text.
123261
+ const HTML_BLOCK_OPEN_RE = /^<HTMLBlock\b([^>]*)>$/;
123221
123262
  /**
123222
- * Base64 encodes HTMLBlock template literal content to prevent markdown parser from consuming <script>/<style> tags.
123223
- *
123224
- * @param content
123225
- * @returns Content with HTMLBlock template literals base64 encoded in HTML comments
123226
- * @example
123227
- * ```typescript
123228
- * const input = '<HTMLBlock>{`<script>alert("xss")</script>`}</HTMLBlock>';
123229
- * protectHTMLBlockContent(input)
123230
- * // Returns: '<HTMLBlock><!--RDMX_HTMLBLOCK:PHNjcmlwdD5hbGVydCgieHNzIik8L3NjcmlwdD4=:RDMX_HTMLBLOCK--></HTMLBlock>'
123231
- * ```
123263
+ * Builds the canonical `html-block` MDAST node the renderer expects.
123232
123264
  */
123233
- function protectHTMLBlockContent(content) {
123234
- return content.replace(/(<HTMLBlock[^>]*>)\{\s*`((?:[^`\\]|\\.)*)`\s*\}(<\/HTMLBlock>)/g, (_match, openTag, templateContent, closeTag) => {
123235
- const encoded = base64Encode(templateContent);
123236
- return `${openTag}${HTML_BLOCK_CONTENT_START}${encoded}${HTML_BLOCK_CONTENT_END}${closeTag}`;
123237
- });
123238
- }
123265
+ const createHtmlBlockNode = (html, position, runScripts, safeMode) => ({
123266
+ position,
123267
+ children: [{ type: 'text', value: html }],
123268
+ type: NodeTypes.htmlBlock,
123269
+ data: {
123270
+ hName: 'html-block',
123271
+ hProperties: {
123272
+ html,
123273
+ ...(runScripts !== undefined && { runScripts }),
123274
+ ...(safeMode !== undefined && { safeMode }),
123275
+ },
123276
+ },
123277
+ });
123239
123278
  /**
123240
- * Removes JSX-style comments (e.g., { /* comment *\/ }) from content.
123241
- *
123242
- * @param content
123243
- * @returns Content with JSX comments removed
123244
- * @example
123245
- * ```typescript
123246
- * removeJSXComments('Text { /* comment *\/ } more text')
123247
- * // Returns: 'Text more text'
123248
- * ```
123279
+ * Reads the cooked string out of a brace expression wrapping a single template
123280
+ * literal (`` `<p>n</p>` `` → `<p>n</p>`).
123249
123281
  */
123250
- function removeJSXComments(content) {
123251
- return content.replace(JSX_COMMENT_REGEX, '');
123252
- }
123253
- const HTML_ELEM_PLACEHOLDER_PREFIX = '___MDXISH_HTML_ELEM_';
123254
- const HTML_ELEM_PLACEHOLDER = new RegExp(`${HTML_ELEM_PLACEHOLDER_PREFIX}(\\d+)___`, 'g');
123255
- // Matches an HTML element that starts at a line boundary and ends at a line boundary.
123256
- // Allows optional leading indentation and lazily matches until the same closing tag.
123257
- const BLOCK_HTML_RE = /(?<=^|\n)[ \t]*<([a-z][a-zA-Z0-9]*)(?:\s[^>]*)?>[\s\S]*?<\/\1>[ \t]*(?=\n|$)/g;
123282
+ const extractTemplateLiteral = (value) => {
123283
+ if (!value)
123284
+ return '';
123285
+ const match = value.trim().match(/^`([\s\S]*)`$/);
123286
+ // Non-template-literal bodies (e.g. `{someVar}`) are malformed mdxish input;
123287
+ // returning '' beats shipping JS identifier source as an HTML payload.
123288
+ return match ? match[1] : '';
123289
+ };
123290
+ const toRunScripts = (raw) => raw === 'true' ? true : raw === 'false' ? false : raw;
123291
+ /** Reads an attribute from a raw `<HTMLBlock …>` attribute string. */
123292
+ const rawAttr = (attrs, name) => {
123293
+ const quoted = attrs.match(new RegExp(`\\b${name}\\s*=\\s*"([^"]*)"`));
123294
+ if (quoted)
123295
+ return quoted[1];
123296
+ const expr = attrs.match(new RegExp(`\\b${name}\\s*=\\s*\\{(true|false)\\}`));
123297
+ if (expr)
123298
+ return expr[1];
123299
+ return new RegExp(`\\b${name}\\b`).test(attrs) ? 'true' : undefined;
123300
+ };
123301
+ /** Reads an attribute from a parsed `<HTMLBlock>` JSX element. */
123302
+ const jsxAttr = (element, name) => {
123303
+ const attr = element.attributes.find(a => a.type === 'mdxJsxAttribute' && a.name === name);
123304
+ if (!attr || attr.type !== 'mdxJsxAttribute')
123305
+ return undefined;
123306
+ if (typeof attr.value === 'string')
123307
+ return attr.value;
123308
+ if (attr.value && typeof attr.value === 'object' && 'value' in attr.value)
123309
+ return attr.value.value;
123310
+ return 'true'; // bare boolean attribute, e.g. <HTMLBlock runScripts />
123311
+ };
123312
+ /** Builds an `html-block` from a raw attribute string and (unparsed) body. */
123313
+ const htmlBlockFromRaw = (attrs, html, position, openingTagIndent = 0) => createHtmlBlockNode(formatHtmlForMdxish(html, openingTagIndent), position, toRunScripts(rawAttr(attrs, 'runScripts')), rawAttr(attrs, 'safeMode'));
123258
123314
  /**
123259
- * Hides line-anchored HTML elements from the brace-escaping pass so we don't leak `\{`
123260
- * into rendered output (rehypeRaw renders the `\` literally, e.g. `<div>{foo</div>`).
123315
+ * Splits a raw `html` node that embeds one or more `<HTMLBlock>`s into
123316
+ * `[html before, html-block, html after, …]`. Returns null when there is none.
123261
123317
  *
123262
- * One carve-out: if an interior line at column 0 has bare text containing `{`, mdxish
123263
- * parses that line as a paragraph and the mdxExpression step would throw without an
123264
- * escape — so we leave that case to the brace balancer.
123318
+ * `String.split` on a regex with capture groups interleaves the captures into
123319
+ * the result, so segments arrive as `[text, attrs, body, text, attrs, body, …]`.
123265
123320
  */
123266
- function protectHTMLElements(content) {
123267
- const htmlElements = [];
123268
- const protectedContent = content.replace(BLOCK_HTML_RE, match => {
123269
- // Look at the lines between the open and close tags. If any of them starts
123270
- // at column 0 with bare text (not whitespace, not another tag) and contains
123271
- // `{`, mdxish will parse that line as a paragraph and the brace as an MDX
123272
- // expression, which would throw an error. So we let the brace balancer escape it.
123273
- // Otherwise, we need to extract the sequence to protect it from the brace escaping.
123274
- const interior = match.split('\n').slice(1, -1);
123275
- const hazard = interior.some(line => line.length > 0 && line[0] !== ' ' && line[0] !== '\t' && line[0] !== '<' && line.includes('{'));
123276
- if (hazard)
123277
- return match;
123278
- htmlElements.push(match);
123279
- return `${HTML_ELEM_PLACEHOLDER_PREFIX}${htmlElements.length - 1}___`;
123280
- });
123281
- return { htmlElements, protectedContent };
123282
- }
123283
- function restoreHTMLElements(content, htmlElements) {
123284
- if (htmlElements.length === 0)
123285
- return content;
123286
- return content.replace(HTML_ELEM_PLACEHOLDER, (_m, idx) => htmlElements[parseInt(idx, 10)]);
123287
- }
123288
- /**
123289
- * Escapes unbalanced and paragraph-spanning braces so MDX doesn't trip on them.
123290
- */
123291
- function escapeProblematicBraces(content) {
123292
- const { htmlElements, protectedContent } = protectHTMLElements(content);
123293
- let strDelim = null;
123294
- let strEscaped = false;
123295
- // Track position of last newline (outside strings) to detect blank lines
123296
- // -2 means no recent newline
123297
- let lastNewlinePos = -2;
123298
- // Character state machine trackers
123299
- const toEscape = new Set();
123300
- // Convert to array of Unicode code points so that emojis and multi-byte characters are correctly tracked
123301
- const chars = Array.from(protectedContent);
123302
- const openStack = [];
123303
- for (let i = 0; i < chars.length; i += 1) {
123304
- const ch = chars[i];
123305
- // Track string delimiters inside expressions to ignore braces within them
123306
- if (openStack.length > 0) {
123307
- if (strDelim) {
123308
- if (strEscaped)
123309
- strEscaped = false;
123310
- else if (ch === '\\')
123311
- strEscaped = true;
123312
- else if (ch === strDelim)
123313
- strDelim = null;
123314
- // eslint-disable-next-line no-continue
123315
- continue;
123316
- }
123317
- if (ch === '"' || ch === "'" || ch === '`') {
123318
- strDelim = ch;
123319
- // eslint-disable-next-line no-continue
123320
- continue;
123321
- }
123322
- if (ch === '\n') {
123323
- if (lastNewlinePos >= 0) {
123324
- const between = chars.slice(lastNewlinePos + 1, i).join('');
123325
- if (/^[ \t]*$/.test(between)) {
123326
- openStack.forEach(entry => { entry.hasBlankLine = true; });
123327
- }
123328
- }
123329
- lastNewlinePos = i;
123330
- }
123331
- }
123332
- // Skip already-escaped braces (odd run of preceding backslashes).
123333
- if (ch === '{' || ch === '}') {
123334
- let bs = 0;
123335
- for (let j = i - 1; j >= 0 && chars[j] === '\\'; j -= 1)
123336
- bs += 1;
123337
- if (bs % 2 === 1) {
123338
- // eslint-disable-next-line no-continue
123339
- continue;
123340
- }
123341
- }
123342
- if (ch === '{') {
123343
- // `=` (after whitespace) before `{` ⇒ JSX attribute expression. The
123344
- // mdxComponent tokenizer captures the whole component, so blank lines
123345
- // inside attribute values are harmless. Nested `{` inherits the flag.
123346
- let isAttrExpr = false;
123347
- for (let j = i - 1; j >= 0; j -= 1) {
123348
- const pc = chars[j];
123349
- if (pc === '=') {
123350
- isAttrExpr = true;
123351
- break;
123352
- }
123353
- if (pc !== ' ' && pc !== '\t')
123354
- break;
123355
- }
123356
- // Nested `{ ... }` inside an attribute value (e.g. `data={[{ ... }]}` or
123357
- // `data={{ a: { b: 1 } }}`) must inherit the same exemption; only the
123358
- // outer `{` is directly after `=`.
123359
- if (!isAttrExpr && openStack.length > 0 && openStack[openStack.length - 1].isAttrExpr) {
123360
- isAttrExpr = true;
123361
- }
123362
- openStack.push({ pos: i, hasBlankLine: false, isAttrExpr });
123363
- lastNewlinePos = -2;
123364
- }
123365
- else if (ch === '}') {
123366
- if (openStack.length > 0) {
123367
- const entry = openStack.pop();
123368
- // Pure `{/* ... */}` comments are handled downstream by the jsxComment
123369
- // tokenizer — escaping their braces would prevent it from running.
123370
- const isPureJsxComment = chars[entry.pos + 1] === '/' &&
123371
- chars[entry.pos + 2] === '*' &&
123372
- chars[i - 1] === '/' &&
123373
- chars[i - 2] === '*';
123374
- if (entry.hasBlankLine && !isPureJsxComment && !entry.isAttrExpr) {
123375
- toEscape.add(entry.pos);
123376
- toEscape.add(i);
123377
- }
123378
- }
123379
- else {
123380
- toEscape.add(i);
123381
- }
123321
+ const splitRawHtmlBlocks = (node) => {
123322
+ const segments = node.value.split(RAW_HTML_BLOCK_RE);
123323
+ if (segments.length === 1)
123324
+ return null; // no <HTMLBlock> present
123325
+ const parts = [];
123326
+ for (let i = 0; i < segments.length; i += 3) {
123327
+ const [text, attrs, body] = segments.slice(i, i + 3);
123328
+ if (text)
123329
+ parts.push({ type: 'html', value: text });
123330
+ if (body !== undefined) {
123331
+ // The opening tag's column equals the length of the line it starts on
123332
+ // (the text run since the previous newline preceding the match).
123333
+ const openingTagIndent = text.slice(text.lastIndexOf('\n') + 1).length;
123334
+ parts.push(htmlBlockFromRaw(attrs, body, node.position, openingTagIndent));
123382
123335
  }
123383
123336
  }
123384
- // Anything still open is unbalanced.
123385
- openStack.forEach(entry => toEscape.add(entry.pos));
123386
- // Reconstruct the content with the escaped braces.
123387
- const escapedContent = toEscape.size === 0 ? protectedContent : chars.map((ch, i) => (toEscape.has(i) ? `\\${ch}` : ch)).join('');
123388
- return restoreHTMLElements(escapedContent, htmlElements);
123389
- }
123337
+ return parts;
123338
+ };
123390
123339
  /**
123391
- * Preprocesses JSX-like markdown content before parsing.
123340
+ * Converts every `<HTMLBlock>` shape that survives parsing into the canonical
123341
+ * `html-block` MDAST node, reading the body from the tokenizer's template-literal
123342
+ * expression. Three shapes occur:
123392
123343
  *
123393
- * JSX attribute expressions (`href={baseUrl}`) are no longer rewritten here —
123394
- * they flow through the tokenizer as `mdxJsxAttributeValueExpression` nodes
123395
- * and are evaluated at the hast handler step.
123344
+ * 1. JSX element (`mdxJsxFlowElement`/`mdxJsxTextElement`) multiline/block
123345
+ * context and table cells (after their remarkMdx re-parse).
123346
+ * 2. Raw `html` blob (`splitRawHtmlBlocks`) single-line top-level, or nested
123347
+ * in raw HTML like an inline `<div>`.
123348
+ * 3. Inline-in-paragraph — split into `html` + expression + `html` siblings.
123396
123349
  *
123397
- * @param content
123398
- * @returns Preprocessed content ready for markdown parsing
123399
- */
123400
- function preprocessJSXExpressions(content) {
123401
- let processed = protectHTMLBlockContent(content);
123402
- const { protectedCode, protectedContent } = protectCodeBlocks(processed);
123403
- processed = escapeProblematicBraces(protectedContent);
123404
- processed = restoreCodeBlocks(processed, protectedCode);
123405
- return processed;
123406
- }
123407
-
123408
- ;// ./processor/transform/mdxish/mdxish-html-blocks.ts
123409
-
123410
-
123411
-
123412
-
123413
- /**
123414
- * Decodes HTMLBlock content that was protected during preprocessing.
123415
- * Content is wrapped in <!--RDMX_HTMLBLOCK:base64:RDMX_HTMLBLOCK-->
123416
- */
123417
- function decodeProtectedContent(content) {
123418
- // Escape special regex characters in the markers
123419
- const startEscaped = HTML_BLOCK_CONTENT_START.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
123420
- const endEscaped = HTML_BLOCK_CONTENT_END.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
123421
- const markerRegex = new RegExp(`${startEscaped}([A-Za-z0-9+/=]+)${endEscaped}`, 'g');
123422
- return content.replace(markerRegex, (_match, encoded) => {
123423
- try {
123424
- return base64Decode(encoded);
123425
- }
123426
- catch {
123427
- return encoded;
123428
- }
123429
- });
123430
- }
123431
- /**
123432
- * Collects text content from a node and its children recursively
123433
- */
123434
- function collectTextContent(node) {
123435
- const parts = [];
123436
- if (node.type === 'text' && node.value) {
123437
- parts.push(node.value);
123438
- }
123439
- else if (node.type === 'html' && node.value) {
123440
- parts.push(node.value);
123441
- }
123442
- else if (node.type === 'inlineCode' && node.value) {
123443
- parts.push(node.value);
123444
- }
123445
- else if (node.type === 'code' && node.value) {
123446
- // Reconstruct code fence syntax (markdown parser consumes opening ```)
123447
- const lang = node.lang || '';
123448
- const fence = `\`\`\`${lang ? `${lang}\n` : ''}`;
123449
- parts.push(fence);
123450
- parts.push(node.value);
123451
- // Add newline before closing fence if missing
123452
- const closingFence = node.value.endsWith('\n') ? '```' : '\n```';
123453
- parts.push(closingFence);
123454
- }
123455
- else if (node.children && Array.isArray(node.children)) {
123456
- node.children.forEach(child => {
123457
- if (typeof child === 'object' && child !== null) {
123458
- parts.push(collectTextContent(child));
123459
- }
123460
- });
123461
- }
123462
- return parts.join('');
123463
- }
123464
- /**
123465
- * Extracts boolean attribute from HTML tag. Handles JSX (safeMode={true}) and string (safeMode="true") syntax.
123466
- * Returns "true"/"false" string to survive rehypeRaw serialization.
123467
- */
123468
- function extractBooleanAttr(attrs, name) {
123469
- // Try JSX syntax: name={true|false}
123470
- const jsxMatch = attrs.match(new RegExp(`${name}=\\{(true|false)\\}`));
123471
- if (jsxMatch) {
123472
- return jsxMatch[1];
123473
- }
123474
- // Try string syntax: name="true"|true
123475
- const stringMatch = attrs.match(new RegExp(`${name}="?(true|false)"?`));
123476
- if (stringMatch) {
123477
- return stringMatch[1];
123478
- }
123479
- return undefined;
123480
- }
123481
- /**
123482
- * Extracts runScripts attribute from HTML tag. Returns boolean for "true"/"false", string for other values, or undefined if not found.
123483
- */
123484
- function extractRunScriptsAttr(attrs) {
123485
- const runScriptsMatch = attrs.match(/runScripts="?([^">\s]+)"?/);
123486
- if (!runScriptsMatch) {
123487
- return undefined;
123488
- }
123489
- const value = runScriptsMatch[1];
123490
- if (value === 'true') {
123491
- return true;
123492
- }
123493
- if (value === 'false') {
123494
- return false;
123495
- }
123496
- return value;
123497
- }
123498
- /**
123499
- * Creates an HTMLBlock node from HTML string and optional attributes
123500
- */
123501
- function createHTMLBlockNode(htmlString, position, runScripts, safeMode) {
123502
- return {
123503
- position,
123504
- children: [{ type: 'text', value: htmlString }],
123505
- type: NodeTypes.htmlBlock,
123506
- data: {
123507
- hName: 'html-block',
123508
- hProperties: {
123509
- html: htmlString,
123510
- ...(runScripts !== undefined && { runScripts }),
123511
- ...(safeMode !== undefined && { safeMode }),
123512
- },
123513
- },
123514
- };
123515
- }
123516
- /**
123517
- * Checks for opening tag only (for split detection)
123518
- */
123519
- function hasOpeningTagOnly(node) {
123520
- let hasOpening = false;
123521
- let hasClosed = false;
123522
- let attrs = '';
123523
- const check = (n) => {
123524
- if (n.type === 'html' && n.value) {
123525
- if (n.value === '<HTMLBlock>') {
123526
- hasOpening = true;
123527
- }
123528
- else {
123529
- const match = n.value.match(/^<HTMLBlock(\s[^>]*)?>$/);
123530
- if (match) {
123531
- hasOpening = true;
123532
- attrs = match[1] || '';
123533
- }
123534
- }
123535
- if (n.value === '</HTMLBlock>' || n.value.includes('</HTMLBlock>')) {
123536
- hasClosed = true;
123537
- }
123538
- }
123539
- if (n.children && Array.isArray(n.children)) {
123540
- n.children.forEach(child => {
123541
- check(child);
123542
- });
123543
- }
123544
- };
123545
- check(node);
123546
- // Return true only if opening without closing (split case)
123547
- return { attrs, found: hasOpening && !hasClosed };
123548
- }
123549
- /**
123550
- * Checks if a node contains an HTMLBlock closing tag
123551
- */
123552
- function hasClosingTag(node) {
123553
- if (node.type === 'html' && node.value) {
123554
- if (node.value === '</HTMLBlock>' || node.value.includes('</HTMLBlock>'))
123555
- return true;
123556
- }
123557
- if (node.children && Array.isArray(node.children)) {
123558
- return node.children.some(child => hasClosingTag(child));
123559
- }
123560
- return false;
123561
- }
123562
- /**
123563
- * Transforms HTMLBlock MDX JSX to html-block nodes. Handles <HTMLBlock>{`...`}</HTMLBlock> syntax.
123350
+ * Runs *after* `mdxishTables` so table cells are re-parsed first;
123351
+ * `mdxishTables` recognizes the still-JSX `<HTMLBlock>` element when deciding to
123352
+ * keep a table as a JSX `<Table>`. This replaces the old base64-comment marker
123353
+ * machinery — the #1455 tokenizer hands the body over already parsed.
123564
123354
  */
123565
123355
  const mdxishHtmlBlocks = () => tree => {
123566
- // Handle HTMLBlock split across root children (caused by newlines)
123567
- visit(tree, 'root', (root) => {
123568
- const children = root.children;
123569
- let i = 0;
123570
- while (i < children.length) {
123571
- const child = children[i];
123572
- const { attrs, found: hasOpening } = hasOpeningTagOnly(child);
123573
- if (hasOpening) {
123574
- // Find closing tag in subsequent siblings
123575
- let closingIdx = -1;
123576
- for (let j = i + 1; j < children.length; j += 1) {
123577
- if (hasClosingTag(children[j])) {
123578
- closingIdx = j;
123579
- break;
123580
- }
123581
- }
123582
- if (closingIdx !== -1) {
123583
- // Collect inner content between tags
123584
- const contentParts = [];
123585
- for (let j = i; j <= closingIdx; j += 1) {
123586
- const node = children[j];
123587
- contentParts.push(collectTextContent(node));
123588
- }
123589
- // Remove the opening/closing tags and template literal syntax from content
123590
- let content = contentParts.join('');
123591
- content = content.replace(/^<HTMLBlock[^>]*>\s*\{?\s*`?/, '').replace(/`?\s*\}?\s*<\/HTMLBlock>$/, '');
123592
- // Decode protected content that was base64 encoded during preprocessing
123593
- content = decodeProtectedContent(content);
123594
- const htmlString = formatHtmlForMdxish(content);
123595
- const runScripts = extractRunScriptsAttr(attrs);
123596
- const safeMode = extractBooleanAttr(attrs, 'safeMode');
123597
- // Replace range with single HTMLBlock node
123598
- const mdNode = createHTMLBlockNode(htmlString, children[i].position, runScripts, safeMode);
123599
- root.children.splice(i, closingIdx - i + 1, mdNode);
123600
- }
123601
- }
123602
- i += 1;
123603
- }
123356
+ // Shape 1: tokenized JSX element.
123357
+ visit(tree, node => node.type === 'mdxJsxFlowElement' || node.type === 'mdxJsxTextElement', (node, index, parent) => {
123358
+ const element = node;
123359
+ if (element.name !== 'HTMLBlock' || !parent || index === undefined)
123360
+ return;
123361
+ const exprChild = element.children.find(child => child.type === 'mdxFlowExpression' || child.type === 'mdxTextExpression');
123362
+ const openingTagIndent = (element.position?.start.column ?? 1) - 1;
123363
+ parent.children[index] = createHtmlBlockNode(formatHtmlForMdxish(extractTemplateLiteral(exprChild?.value), openingTagIndent), element.position, toRunScripts(jsxAttr(element, 'runScripts')), jsxAttr(element, 'safeMode'));
123604
123364
  });
123605
- // Handle HTMLBlock parsed as HTML elements (when template literal contains block-level HTML tags)
123365
+ // Shape 2: raw HTML blob.
123606
123366
  visit(tree, 'html', (node, index, parent) => {
123607
123367
  if (!parent || index === undefined)
123608
123368
  return;
123609
- const value = node.value;
123610
- if (!value)
123611
- return;
123612
- // Case 1: Full HTMLBlock in single node
123613
- const fullMatch = value.match(/^<HTMLBlock(\s[^>]*)?>([\s\S]*)<\/HTMLBlock>$/);
123614
- if (fullMatch) {
123615
- const attrs = fullMatch[1] || '';
123616
- let content = fullMatch[2] || '';
123617
- // Remove template literal syntax if present: {`...`}
123618
- content = content.replace(/^\s*\{\s*`/, '').replace(/`\s*\}\s*$/, '');
123619
- // Decode protected content that was base64 encoded during preprocessing
123620
- content = decodeProtectedContent(content);
123621
- const htmlString = formatHtmlForMdxish(content);
123622
- const runScripts = extractRunScriptsAttr(attrs);
123623
- const safeMode = extractBooleanAttr(attrs, 'safeMode');
123624
- parent.children[index] = createHTMLBlockNode(htmlString, node.position, runScripts, safeMode);
123625
- return;
123626
- }
123627
- // Case 2: Opening tag only (split by blank lines)
123628
- if (value === '<HTMLBlock>' || value.match(/^<HTMLBlock\s[^>]*>$/)) {
123629
- const siblings = parent.children;
123630
- let closingIdx = -1;
123631
- // Find closing tag in siblings
123632
- for (let i = index + 1; i < siblings.length; i += 1) {
123633
- const sibling = siblings[i];
123634
- if (sibling.type === 'html') {
123635
- const sibVal = sibling.value;
123636
- if (sibVal === '</HTMLBlock>' || sibVal?.includes('</HTMLBlock>')) {
123637
- closingIdx = i;
123638
- break;
123639
- }
123640
- }
123641
- }
123642
- if (closingIdx === -1)
123643
- return;
123644
- // Collect content between tags, skipping template literal delimiters
123645
- const contentParts = [];
123646
- for (let i = index + 1; i < closingIdx; i += 1) {
123647
- const sibling = siblings[i];
123648
- // Skip template literal delimiters
123649
- if (sibling.type === 'text') {
123650
- const textVal = sibling.value;
123651
- if (textVal === '{' || textVal === '}' || textVal === '{`' || textVal === '`}') {
123652
- // eslint-disable-next-line no-continue
123653
- continue;
123654
- }
123655
- }
123656
- contentParts.push(collectTextContent(sibling));
123657
- }
123658
- // Decode protected content that was base64 encoded during preprocessing
123659
- const decodedContent = decodeProtectedContent(contentParts.join(''));
123660
- const htmlString = formatHtmlForMdxish(decodedContent);
123661
- const runScripts = extractRunScriptsAttr(value);
123662
- const safeMode = extractBooleanAttr(value, 'safeMode');
123663
- // Replace opening tag with HTMLBlock node, remove consumed siblings
123664
- parent.children[index] = createHTMLBlockNode(htmlString, node.position, runScripts, safeMode);
123665
- parent.children.splice(index + 1, closingIdx - index);
123666
- }
123369
+ const replacement = splitRawHtmlBlocks(node);
123370
+ if (replacement)
123371
+ parent.children.splice(index, 1, ...replacement);
123667
123372
  });
123668
- // Handle HTMLBlock inside paragraphs (parsed as inline elements)
123669
- visit(tree, 'paragraph', (node, index, parent) => {
123670
- if (!parent || index === undefined)
123671
- return;
123672
- const children = node.children || [];
123673
- let htmlBlockStartIdx = -1;
123674
- let htmlBlockEndIdx = -1;
123675
- let templateLiteralStartIdx = -1;
123676
- let templateLiteralEndIdx = -1;
123373
+ // Shape 3: inline within a paragraph `<HTMLBlock>` open/close arrive as
123374
+ // separate `html` siblings with the template-literal expression between them.
123375
+ visit(tree, 'paragraph', (paragraph) => {
123376
+ // An html-block is block content, so it isn't a valid PhrasingContent child;
123377
+ // widen to RootContent (which HTMLBlock belongs to) for the in-place splice.
123378
+ const children = paragraph.children;
123677
123379
  for (let i = 0; i < children.length; i += 1) {
123678
- const child = children[i];
123679
- if (child.type === 'html' && typeof child.value === 'string') {
123680
- const value = child.value;
123681
- if (value === '<HTMLBlock>' || value.match(/^<HTMLBlock\s[^>]*>$/)) {
123682
- htmlBlockStartIdx = i;
123683
- }
123684
- else if (value === '</HTMLBlock>') {
123685
- htmlBlockEndIdx = i;
123686
- }
123687
- }
123688
- // Find opening brace after HTMLBlock start
123689
- if (htmlBlockStartIdx !== -1 && templateLiteralStartIdx === -1 && child.type === 'text') {
123690
- const value = child.value;
123691
- if (value === '{') {
123692
- templateLiteralStartIdx = i;
123380
+ const open = children[i];
123381
+ const openMatch = open.type === 'html' ? open.value.match(HTML_BLOCK_OPEN_RE) : null;
123382
+ if (!openMatch)
123383
+ continue; // eslint-disable-line no-continue
123384
+ const closeIdx = children.findIndex((child, j) => j > i && child.type === 'html' && child.value === '</HTMLBlock>');
123385
+ if (closeIdx === -1)
123386
+ continue; // eslint-disable-line no-continue
123387
+ const body = children
123388
+ .slice(i + 1, closeIdx)
123389
+ .map(child => {
123390
+ if (child.type === 'mdxTextExpression' || child.type === 'mdxFlowExpression') {
123391
+ return extractTemplateLiteral(child.value);
123693
123392
  }
123694
- }
123695
- // Find closing brace before HTMLBlock end
123696
- if (htmlBlockStartIdx !== -1 && htmlBlockEndIdx === -1 && child.type === 'text') {
123697
- const value = child.value;
123698
- if (value === '}') {
123699
- templateLiteralEndIdx = i;
123700
- }
123701
- }
123702
- }
123703
- if (htmlBlockStartIdx !== -1 &&
123704
- htmlBlockEndIdx !== -1 &&
123705
- templateLiteralStartIdx !== -1 &&
123706
- templateLiteralEndIdx !== -1 &&
123707
- templateLiteralStartIdx < templateLiteralEndIdx) {
123708
- const openingTag = children[htmlBlockStartIdx];
123709
- // Collect content between braces (handles code blocks)
123710
- const templateContent = [];
123711
- for (let i = templateLiteralStartIdx + 1; i < templateLiteralEndIdx; i += 1) {
123712
- const child = children[i];
123713
- templateContent.push(collectTextContent(child));
123714
- }
123715
- // Decode protected content that was base64 encoded during preprocessing
123716
- const decodedContent = decodeProtectedContent(templateContent.join(''));
123717
- const htmlString = formatHtmlForMdxish(decodedContent);
123718
- const runScripts = openingTag.value ? extractRunScriptsAttr(openingTag.value) : undefined;
123719
- const safeMode = openingTag.value ? extractBooleanAttr(openingTag.value, 'safeMode') : undefined;
123720
- const mdNode = createHTMLBlockNode(htmlString, node.position, runScripts, safeMode);
123721
- parent.children[index] = mdNode;
123722
- }
123723
- });
123724
- // Ensure html-block nodes have HTML in children as text node
123725
- visit(tree, 'html-block', (node) => {
123726
- const html = node.data?.hProperties?.html;
123727
- if (html &&
123728
- (!node.children ||
123729
- node.children.length === 0 ||
123730
- (node.children.length === 1 && node.children[0].type === 'text' && node.children[0].value !== html))) {
123731
- node.children = [
123732
- {
123733
- type: 'text',
123734
- value: html,
123735
- },
123736
- ];
123393
+ // Preserve raw text from any other phrasing sibling (e.g. stray
123394
+ // whitespace or content the tokenizer didn't claim) so it isn't
123395
+ // silently dropped from the html payload.
123396
+ return 'value' in child && typeof child.value === 'string' ? child.value : '';
123397
+ })
123398
+ .join('');
123399
+ const openingTagIndent = (open.position?.start.column ?? 1) - 1;
123400
+ children.splice(i, closeIdx - i + 1, htmlBlockFromRaw(openMatch[1], body, open.position, openingTagIndent));
123737
123401
  }
123738
123402
  });
123739
- return tree;
123740
123403
  };
123741
123404
  /* harmony default export */ const mdxish_html_blocks = (mdxishHtmlBlocks);
123742
123405
 
@@ -124517,6 +124180,189 @@ const normalizeMdxJsxNodes = () => tree => {
124517
124180
  };
124518
124181
  /* harmony default export */ const normalize_mdx_jsx_nodes = (normalizeMdxJsxNodes);
124519
124182
 
124183
+ ;// ./lib/micromark/jsx-comment/pattern.ts
124184
+ /**
124185
+ * Matches a JSX comment: `{/*`, content, `*\/}` — no whitespace tolerated
124186
+ * between the braces and the comment markers.
124187
+ *
124188
+ * This grammar is mirrored by the flow tokenizer in ./syntax.ts. Any change
124189
+ * here needs a mirror change in the state machine; the parity test in
124190
+ * __tests__/lib/micromark/jsx-comment-pattern-parity.test.ts locks the two
124191
+ * together so they can't silently drift.
124192
+ */
124193
+ const JSX_COMMENT_REGEX = /\{\/\*[^*]*(?:\*(?!\/)[^*]*)*\*\/\}/g;
124194
+
124195
+ ;// ./processor/transform/mdxish/preprocess-jsx-expressions.ts
124196
+
124197
+
124198
+ /**
124199
+ * Removes JSX-style comments (e.g., { /* comment *\/ }) from content.
124200
+ *
124201
+ * @param content
124202
+ * @returns Content with JSX comments removed
124203
+ * @example
124204
+ * ```typescript
124205
+ * removeJSXComments('Text { /* comment *\/ } more text')
124206
+ * // Returns: 'Text more text'
124207
+ * ```
124208
+ */
124209
+ function removeJSXComments(content) {
124210
+ return content.replace(JSX_COMMENT_REGEX, '');
124211
+ }
124212
+ const HTML_ELEM_PLACEHOLDER_PREFIX = '___MDXISH_HTML_ELEM_';
124213
+ const HTML_ELEM_PLACEHOLDER = new RegExp(`${HTML_ELEM_PLACEHOLDER_PREFIX}(\\d+)___`, 'g');
124214
+ // Matches an HTML element that starts at a line boundary and ends at a line boundary.
124215
+ // Allows optional leading indentation and lazily matches until the same closing tag.
124216
+ const BLOCK_HTML_RE = /(?<=^|\n)[ \t]*<([a-z][a-zA-Z0-9]*)(?:\s[^>]*)?>[\s\S]*?<\/\1>[ \t]*(?=\n|$)/g;
124217
+ /**
124218
+ * Hides line-anchored HTML elements from the brace-escaping pass so we don't leak `\{`
124219
+ * into rendered output (rehypeRaw renders the `\` literally, e.g. `<div>{foo</div>`).
124220
+ *
124221
+ * One carve-out: if an interior line at column 0 has bare text containing `{`, mdxish
124222
+ * parses that line as a paragraph and the mdxExpression step would throw without an
124223
+ * escape — so we leave that case to the brace balancer.
124224
+ */
124225
+ function protectHTMLElements(content) {
124226
+ const htmlElements = [];
124227
+ const protectedContent = content.replace(BLOCK_HTML_RE, match => {
124228
+ // Look at the lines between the open and close tags. If any of them starts
124229
+ // at column 0 with bare text (not whitespace, not another tag) and contains
124230
+ // `{`, mdxish will parse that line as a paragraph and the brace as an MDX
124231
+ // expression, which would throw an error. So we let the brace balancer escape it.
124232
+ // Otherwise, we need to extract the sequence to protect it from the brace escaping.
124233
+ const interior = match.split('\n').slice(1, -1);
124234
+ const hazard = interior.some(line => line.length > 0 && line[0] !== ' ' && line[0] !== '\t' && line[0] !== '<' && line.includes('{'));
124235
+ if (hazard)
124236
+ return match;
124237
+ htmlElements.push(match);
124238
+ return `${HTML_ELEM_PLACEHOLDER_PREFIX}${htmlElements.length - 1}___`;
124239
+ });
124240
+ return { htmlElements, protectedContent };
124241
+ }
124242
+ function restoreHTMLElements(content, htmlElements) {
124243
+ if (htmlElements.length === 0)
124244
+ return content;
124245
+ return content.replace(HTML_ELEM_PLACEHOLDER, (_m, idx) => htmlElements[parseInt(idx, 10)]);
124246
+ }
124247
+ /**
124248
+ * Escapes unbalanced and paragraph-spanning braces so MDX doesn't trip on them.
124249
+ */
124250
+ function escapeProblematicBraces(content) {
124251
+ const { htmlElements, protectedContent } = protectHTMLElements(content);
124252
+ let strDelim = null;
124253
+ let strEscaped = false;
124254
+ // Track position of last newline (outside strings) to detect blank lines
124255
+ // -2 means no recent newline
124256
+ let lastNewlinePos = -2;
124257
+ // Character state machine trackers
124258
+ const toEscape = new Set();
124259
+ // Convert to array of Unicode code points so that emojis and multi-byte characters are correctly tracked
124260
+ const chars = Array.from(protectedContent);
124261
+ const openStack = [];
124262
+ for (let i = 0; i < chars.length; i += 1) {
124263
+ const ch = chars[i];
124264
+ // Track string delimiters inside expressions to ignore braces within them
124265
+ if (openStack.length > 0) {
124266
+ if (strDelim) {
124267
+ if (strEscaped)
124268
+ strEscaped = false;
124269
+ else if (ch === '\\')
124270
+ strEscaped = true;
124271
+ else if (ch === strDelim)
124272
+ strDelim = null;
124273
+ // eslint-disable-next-line no-continue
124274
+ continue;
124275
+ }
124276
+ if (ch === '"' || ch === "'" || ch === '`') {
124277
+ strDelim = ch;
124278
+ // eslint-disable-next-line no-continue
124279
+ continue;
124280
+ }
124281
+ if (ch === '\n') {
124282
+ if (lastNewlinePos >= 0) {
124283
+ const between = chars.slice(lastNewlinePos + 1, i).join('');
124284
+ if (/^[ \t]*$/.test(between)) {
124285
+ openStack.forEach(entry => { entry.hasBlankLine = true; });
124286
+ }
124287
+ }
124288
+ lastNewlinePos = i;
124289
+ }
124290
+ }
124291
+ // Skip already-escaped braces (odd run of preceding backslashes).
124292
+ if (ch === '{' || ch === '}') {
124293
+ let bs = 0;
124294
+ for (let j = i - 1; j >= 0 && chars[j] === '\\'; j -= 1)
124295
+ bs += 1;
124296
+ if (bs % 2 === 1) {
124297
+ // eslint-disable-next-line no-continue
124298
+ continue;
124299
+ }
124300
+ }
124301
+ if (ch === '{') {
124302
+ // `=` (after whitespace) before `{` ⇒ JSX attribute expression. The
124303
+ // mdxComponent tokenizer captures the whole component, so blank lines
124304
+ // inside attribute values are harmless. Nested `{` inherits the flag.
124305
+ let isAttrExpr = false;
124306
+ for (let j = i - 1; j >= 0; j -= 1) {
124307
+ const pc = chars[j];
124308
+ if (pc === '=') {
124309
+ isAttrExpr = true;
124310
+ break;
124311
+ }
124312
+ if (pc !== ' ' && pc !== '\t')
124313
+ break;
124314
+ }
124315
+ // Nested `{ ... }` inside an attribute value (e.g. `data={[{ ... }]}` or
124316
+ // `data={{ a: { b: 1 } }}`) must inherit the same exemption; only the
124317
+ // outer `{` is directly after `=`.
124318
+ if (!isAttrExpr && openStack.length > 0 && openStack[openStack.length - 1].isAttrExpr) {
124319
+ isAttrExpr = true;
124320
+ }
124321
+ openStack.push({ pos: i, hasBlankLine: false, isAttrExpr });
124322
+ lastNewlinePos = -2;
124323
+ }
124324
+ else if (ch === '}') {
124325
+ if (openStack.length > 0) {
124326
+ const entry = openStack.pop();
124327
+ // Pure `{/* ... */}` comments are handled downstream by the jsxComment
124328
+ // tokenizer — escaping their braces would prevent it from running.
124329
+ const isPureJsxComment = chars[entry.pos + 1] === '/' &&
124330
+ chars[entry.pos + 2] === '*' &&
124331
+ chars[i - 1] === '/' &&
124332
+ chars[i - 2] === '*';
124333
+ if (entry.hasBlankLine && !isPureJsxComment && !entry.isAttrExpr) {
124334
+ toEscape.add(entry.pos);
124335
+ toEscape.add(i);
124336
+ }
124337
+ }
124338
+ else {
124339
+ toEscape.add(i);
124340
+ }
124341
+ }
124342
+ }
124343
+ // Anything still open is unbalanced.
124344
+ openStack.forEach(entry => toEscape.add(entry.pos));
124345
+ // Reconstruct the content with the escaped braces.
124346
+ const escapedContent = toEscape.size === 0 ? protectedContent : chars.map((ch, i) => (toEscape.has(i) ? `\\${ch}` : ch)).join('');
124347
+ return restoreHTMLElements(escapedContent, htmlElements);
124348
+ }
124349
+ /**
124350
+ * Preprocesses JSX-like markdown content before parsing.
124351
+ *
124352
+ * JSX attribute expressions (`href={baseUrl}`) are no longer rewritten here —
124353
+ * they flow through the tokenizer as `mdxJsxAttributeValueExpression` nodes
124354
+ * and are evaluated at the hast handler step.
124355
+ *
124356
+ * @param content
124357
+ * @returns Preprocessed content ready for markdown parsing
124358
+ */
124359
+ function preprocessJSXExpressions(content) {
124360
+ const { protectedCode, protectedContent } = protectCodeBlocks(content);
124361
+ let processed = escapeProblematicBraces(protectedContent);
124362
+ processed = restoreCodeBlocks(processed, protectedCode);
124363
+ return processed;
124364
+ }
124365
+
124520
124366
  ;// ./processor/transform/mdxish/restore-snake-case-component-name.ts
124521
124367
 
124522
124368
 
@@ -125633,7 +125479,7 @@ function mdxishAstProcessor(mdContent, opts = {}) {
125633
125479
  .use(inline_html, { safeMode })
125634
125480
  .use(restore_snake_case_component_name, { mapping: snakeCaseMapping })
125635
125481
  .use(mdxish_tables)
125636
- .use(mdxish_html_blocks)
125482
+ .use(mdxish_html_blocks) // Convert every <HTMLBlock> shape → html-block
125637
125483
  // The next few transformers must appear after mdxishMdxComponentBlocks
125638
125484
  // so nodes produced by the inline re-parse of component bodies
125639
125485
  // (e.g. code/image/embed inside <Tabs>) get visited too