@rohal12/spindle 0.51.4 → 0.52.0

package/src/markup/code-attributes.ts

@@ -0,0 +1,64 @@
+import { createScanMemo, scanBalancedBrace } from './tokenizer';
+
+/**
+ * Attributes whose value is code with braces of its own, not text: event
+ * handlers (`on…`, JavaScript), `pattern` (a regular expression, as in
+ * `\p{L}{2,3}`) and `srcdoc` (an HTML document, with its own scripts and
+ * styles). Their braces are code, so they are not read as markup: only
+ * sigil references (`{$x}`, `{_x}`, `{@x}`, `{%x}` and expressions starting
+ * with one) are resolved, and any other `{` is literal.
+ */
+export function isCodeAttribute(name: string): boolean {
+  const lower = name.toLowerCase();
+  return lower.startsWith('on') || lower === 'pattern' || lower === 'srcdoc';
+}
+
+/** Literal text, the source of one `{…}` reference (without braces), or a
+ * `{…}` block that is kept exactly as written. */
+export type SigilPart =
+  | { text: string }
+  | { expr: string }
+  | { verbatim: string };
+
+/**
+ * Split a code attribute value into literal text and its `{…}` blocks that
+ * start with a sigil. Braces inside strings don't close a block; an
+ * unclosed block, any other brace and backslashes are text. A block whose
+ * sigil is not followed by a word character (`{$(…)}`) is code, kept
+ * verbatim: not resolved, and not decoded like text.
+ */
+export function splitSigilTemplate(template: string): SigilPart[] {
+  const parts: SigilPart[] = [];
+  const memo = createScanMemo();
+  let text = '';
+  let i = 0;
+
+  while (i < template.length) {
+    const brace = template.indexOf('{', i);
+    if (brace === -1) {
+      text += template.slice(i);
+      break;
+    }
+    text += template.slice(i, brace);
+    i = brace + 1;
+    const sigil = template[i];
+    const end =
+      sigil === '$' || sigil === '_' || sigil === '@' || sigil === '%'
+        ? scanBalancedBrace(template, i, memo)
+        : -1;
+    if (end === -1) {
+      text += '{';
+      continue;
+    }
+    if (text) parts.push({ text });
+    text = '';
+    const block = template.slice(i, end);
+    parts.push(
+      /^.\w/.test(block) ? { expr: block } : { verbatim: `{${block}}` },
+    );
+    i = end + 1;
+  }
+
+  if (text) parts.push({ text });
+  return parts;
+}

package/src/markup/markdown.ts

@@ -1,4 +1,11 @@
 import { micromark } from 'micromark';
+import type {
+  Code,
+  Construct,
+  Extension,
+  Options,
+  Resolver,
+} from 'micromark-util-types';
 import { gfmTable, gfmTableHtml } from 'micromark-extension-gfm-table';
 import {
   gfmStrikethrough,
@@ -6,17 +13,15 @@ import {
 } from 'micromark-extension-gfm-strikethrough';
 
 /**
- * Parse a text string as CommonMark markdown and return an HTML string.
- * Includes GFM table and strikethrough extensions.
+ * micromark options for spindle's markdown: CommonMark with GFM tables and
+ * strikethrough, raw HTML passed through, and no indented code.
  *
  * When `inline` is true, block-level constructs (lists, headings, blockquotes,
- * thematic breaks) are disabled.  This is used when rendering content inside
- * inline HTML elements like `<span>` where block-level output is invalid.
+ * thematic breaks, fenced code, tables) are disabled.  This is used when
+ * rendering content inside inline HTML elements like `<span>` where
+ * block-level output is invalid. Fences then read as code spans.
  */
-export function markdownToHtml(
-  text: string,
-  options?: { inline?: boolean },
-): string {
+export function markdownOptions(options?: { inline?: boolean }): Options {
   const disabled: string[] = ['codeIndented'];
   if (options?.inline) {
     disabled.push(
@@ -25,9 +30,11 @@ export function markdownToHtml(
       'setextUnderline',
       'thematicBreak',
       'blockQuote',
+      'codeFenced',
+      'table',
     );
   }
-  return micromark(text, {
+  return {
     allowDangerousHtml: true,
     extensions: [
       gfmTable(),
@@ -35,5 +42,177 @@ export function markdownToHtml(
       { disable: { null: disabled } },
     ],
     htmlExtensions: [gfmTableHtml(), gfmStrikethroughHtml()],
+  };
+}
+
+/**
+ * Parse a text string as CommonMark markdown and return an HTML string, as
+ * micromark does with `markdownOptions`, but in about linear time (see
+ * `unclosedHtml` and `textData`).
+ */
+export function markdownToHtml(
+  text: string,
+  options?: { inline?: boolean },
+): string {
+  const base = markdownOptions(options);
+  return micromark(text, {
+    ...base,
+    extensions: [...(base.extensions ?? []), unclosedHtml(text), textData],
   });
 }
+
+/** Join a data token just read to a data token right before it. */
+const joinData: Resolver = (events) => {
+  const at = events.length - 2;
+  const before = events[at - 1];
+  if (before && before[0] === 'exit' && before[1].type === 'data') {
+    before[1].end = events[at]![1].end;
+    events.length = at;
+  }
+  return events;
+};
+
+/**
+ * Text that no construct takes, read as data joined to the data before it.
+ *
+ * Where every text construct fails at a character (a `!` without `[`, a `]`
+ * without a label, a `\` before a letter, an unclosed `<!--`), micromark
+ * reads it, on to the next character a construct may start at, as a new data
+ * token. Runs of data tokens are joined only after the paragraph is read,
+ * each run by splicing the paragraph's whole list of events: time quadratic
+ * in the number of runs where other tokens part them (`\]a!` repeated: an
+ * escape, data `a`, data `!`), and erratically so, as the cost of moving
+ * the list's elements depends on where the garbage collector keeps it.
+ *
+ * Tried where micromark's text tries constructs, after all of them (as a
+ * construct of every character), this reads that data as micromark would
+ * and joins it to the data before it at once, at the end of the list.
+ */
+const textData: Extension = {
+  text: {
+    null: {
+      name: 'textData',
+      tokenize(effects, ok) {
+        const self = this;
+        const constructs = this.parser.constructs.text;
+
+        /** Whether a construct may start at `code`, as micromark's text asks. */
+        function atBreak(code: Code): boolean {
+          if (code === null) return true;
+          const list = constructs[code] ?? [];
+          return (Array.isArray(list) ? list : [list]).some(
+            (item) => !item.previous || item.previous.call(self, self.previous),
+          );
+        }
+
+        function data(code: Code) {
+          if (atBreak(code)) {
+            effects.exit('data');
+            return ok(code);
+          }
+          effects.consume(code);
+          return data;
+        }
+
+        return (code) => {
+          effects.enter('data');
+          effects.consume(code);
+          return data;
+        };
+      },
+      resolveTo: joinData,
+    },
+  },
+};
+
+/** Characters of an email autolink's local part (micromark's `asciiAtext`). */
+const ATEXT = /[#-'*+\--9=?A-Z^-~]/;
+const ALPHANUMERIC = /[0-9A-Za-z]/;
+const ALPHA = /[A-Za-z]/;
+
+/**
+ * Inline raw HTML that can't be closed in `text`, read as the text it is.
+ *
+ * At a `<` opening a comment (`<!--`), a processing instruction (`<?`), a
+ * CDATA section (`<![CDATA[`) or a declaration (`<!` and a letter), micromark
+ * reads on for the closer (`-->`, `?>`, `]]>`, `>`) to the end of the
+ * paragraph, and without one the `<` is text. Many such openers took time
+ * quadratic in their number. Where no closer comes later in the whole text,
+ * this construct, tried at each `<` before micromark's own, takes the `<`
+ * as text at once, as micromark would end up doing.
+ *
+ * It is tried exactly where micromark tries raw HTML: not in code, link
+ * destinations or HTML blocks (which start at a line start and run to their
+ * end condition or the end of the document), whose output stays micromark's.
+ * A `<?` may begin an email autolink (`<?a@b.c>`, tried before raw HTML),
+ * which it then leaves to micromark.
+ */
+function unclosedHtml(text: string): Extension {
+  // micromark drops a byte order mark, which offsets don't count
+  const shift = text.charCodeAt(0) === 0xfeff ? 1 : 0;
+  const comment = text.lastIndexOf('-->');
+  const instruction = text.lastIndexOf('?>');
+  const cdata = text.lastIndexOf(']]>');
+  const declaration = text.lastIndexOf('>');
+
+  /** Whether the raw HTML opening at `p` can't be closed. */
+  function unclosed(p: number): boolean {
+    if (text[p] !== '<') return false;
+    // A comment closes with a `-->` from just past `<!`: `<!-->` is one
+    if (text.startsWith('<!--', p)) return comment < p + 2;
+    if (text.startsWith('<![CDATA[', p)) return cdata < p + 9;
+    if (text[p + 1] === '!' && ALPHA.test(text[p + 2] ?? '')) {
+      return declaration < p + 3;
+    }
+    if (text[p + 1] === '?') return instruction < p + 2 && !emailAutolink(p);
+    return false;
+  }
+
+  /** Whether an email autolink opens at `p`, as micromark reads one. */
+  function emailAutolink(p: number): boolean {
+    let i = p + 1;
+    while (ATEXT.test(text[i] ?? '')) i++;
+    if (text[i] !== '@') return false;
+    i++;
+    let size = 0;
+    let state: 'atSignOrDot' | 'label' | 'value' = 'atSignOrDot';
+    for (;;) {
+      const c = text[i] ?? '';
+      if (state === 'atSignOrDot') {
+        if (!ALPHANUMERIC.test(c)) return false;
+        state = 'label';
+      } else if (state === 'label') {
+        if (c === '>') return true;
+        if (c === '.') {
+          i++;
+          size = 0;
+          state = 'atSignOrDot';
+        } else {
+          state = 'value';
+        }
+      } else {
+        if (!((c === '-' || ALPHANUMERIC.test(c)) && size++ < 63)) {
+          return false;
+        }
+        i++;
+        state = c === '-' ? 'value' : 'label';
+      }
+    }
+  }
+
+  const construct: Construct = {
+    name: 'unclosedHtml',
+    resolveTo: joinData,
+    tokenize(effects, ok, nok) {
+      const at = this.now().offset + shift;
+      return (code) => {
+        if (code !== 60 || !unclosed(at)) return nok(code);
+        effects.enter('data');
+        effects.consume(code);
+        effects.exit('data');
+        return ok;
+      };
+    },
+  };
+  return { text: { 60: construct } };
+}