cssgrep 1.1.0 → 1.3.0

package/completions/_cssgrep

@@ -5,6 +5,15 @@
 
 _cssgrep() {
   _arguments -s -S \
+    '*'{-e,--selector}'[add a labeled selector (repeatable)]:selector:' \
+    '*--add-class[rewrite: add a class to matched elements]:class:' \
+    '*--remove-class[rewrite: remove a class]:class:' \
+    '*--set-attr[rewrite: set attribute name=value]:name=value:' \
+    '*--remove-attr[rewrite: remove an attribute]:attribute:' \
+    '--rename-tag[rewrite: rename the element]:tag:' \
+    '--watch[re-run the search whenever a watched file changes]' \
+    '--no-clear[with --watch on a TTY: append instead of clearing]' \
+    '--diff[emit a unified diff instead of the document]' \
     '(-r --recursive)'{-r,--recursive}'[recurse into directory arguments]' \
     '--max-depth[limit -r recursion depth (1 = given dir only)]:depth:' \
     '--ext[extensions to scan with -r]:extensions (comma-separated):' \
@@ -12,6 +21,7 @@ _cssgrep() {
     '*'{-i,--ignore}'[skip files/dirs matching glob while recursing]:glob:' \
     '*--exclude[alias for --ignore]:glob:' \
     '--ignore-file[read ignore globs from a file]:file:_files' \
+    '(-S --follow)'{-S,--follow}'[follow symbolic links while recursing]' \
     '(-n --line-number)'{-n,--line-number}'[prefix each match with line:col]' \
     '(-p --print)'{-p,--print}'[pretty-print the matched node HTML]' \
     '--attr[print the value of an attribute]:attribute name:' \

package/completions/cssgrep.bash

@@ -6,7 +6,7 @@ _cssgrep() {
     cur="${COMP_WORDS[COMP_CWORD]}"
     prev="${COMP_WORDS[COMP_CWORD-1]}"
 
-    opts="-r --recursive --max-depth --ext --include -i --ignore --exclude --ignore-file -n --line-number \
+    opts="-e --selector --watch --no-clear --add-class --remove-class --set-attr --remove-attr --rename-tag --diff -r --recursive --max-depth --ext --include -i --ignore --exclude --ignore-file -S --follow -n --line-number \
 -p --print --attr --text --json --parent -w --max-width \
 -A --after-context -B --before-context -C --context \
 -m --max-count -M --max-total -c --count \
@@ -23,7 +23,7 @@ _cssgrep() {
             COMPREPLY=( $(compgen -f -- "$cur") )
             return 0
             ;;
-        --max-depth|--ext|--include|-i|--ignore|--exclude|--attr|--parent|-w|--max-width|-A|--after-context|\
+        -e|--selector|--add-class|--remove-class|--set-attr|--remove-attr|--rename-tag|--max-depth|--ext|--include|-i|--ignore|--exclude|--attr|--parent|-w|--max-width|-A|--after-context|\
 -B|--before-context|-C|--context|-m|--max-count|-M|--max-total)
             # value-taking flag with no enumerable completion; offer nothing
             return 0

package/completions/cssgrep.fish

@@ -1,6 +1,15 @@
 # fish completion for cssgrep. Keep in sync with the USAGE string in index.js.
 # Install: place this file in ~/.config/fish/completions/ (named cssgrep.fish).
 
+complete -c cssgrep -s e -l selector -x -d 'Add a [label=]selector (repeatable)'
+complete -c cssgrep -l add-class -x -d 'Rewrite: add a class to matched elements'
+complete -c cssgrep -l remove-class -x -d 'Rewrite: remove a class (attr dropped when empty)'
+complete -c cssgrep -l set-attr -x -d 'Rewrite: set attribute (name=value)'
+complete -c cssgrep -l remove-attr -x -d 'Rewrite: remove an attribute'
+complete -c cssgrep -l rename-tag -x -d 'Rewrite: rename the element (and closing tag)'
+complete -c cssgrep -l diff -d 'Emit a unified diff instead of the document'
+complete -c cssgrep -l watch -d 'Re-run the search whenever a watched file changes'
+complete -c cssgrep -l no-clear -d 'With --watch on a TTY: append instead of clearing'
 complete -c cssgrep -s r -l recursive -d 'Recurse into directory arguments'
 complete -c cssgrep -l max-depth -x -d 'Limit -r recursion depth (1 = given dir only)'
 complete -c cssgrep -l ext -x -d 'Extensions to scan with -r (comma-separated)'
@@ -8,6 +17,7 @@ complete -c cssgrep -l include -x -d 'Only search files matching glob (replaces
 complete -c cssgrep -s i -l ignore -x -d 'Skip files/dirs matching glob while recursing'
 complete -c cssgrep -l exclude -x -d 'Alias for --ignore'
 complete -c cssgrep -l ignore-file -r -F -d 'Read ignore globs from a file'
+complete -c cssgrep -s S -l follow -d 'Follow symbolic links while recursing'
 complete -c cssgrep -s n -l line-number -d 'Prefix each match with line:col'
 complete -c cssgrep -s p -l print -d 'Pretty-print the matched node HTML'
 complete -c cssgrep -l attr -x -d 'Print the value of an attribute'

package/index.d.ts

@@ -0,0 +1,93 @@
+// Type definitions for cssgrep's library API (lib.js).
+//
+// `parse`, the document's `html`/`search`, and the Match fields below are the
+// stable public surface. `node` is an advanced escape hatch: the raw
+// htmlparser2 element, whose shape belongs to htmlparser2 and may change with
+// it. Other properties reachable on the document object at runtime are
+// internal and unstable.
+
+export interface SearchOptions {
+  /**
+   * Re-target each match to its n-th element ancestor (clamped at the
+   * document root), deduplicated by identity — the CLI's --parent.
+   */
+  parent?: number;
+}
+
+export interface Match {
+  /** 0-based offset of the node's opening `<` in the input string. */
+  start: number;
+  /** 0-based exclusive end offset: `doc.html.slice(start, end)` is the node. */
+  end: number;
+  /** 1-based line of the opening `<` (computed lazily on first read). */
+  line: number;
+  /** 1-based *byte* column of the opening `<` (what vim's grepformat %c expects). */
+  col: number;
+  /** Lowercased tag name. */
+  tag: string;
+  /** The node's attributes (names lowercased by the parser). */
+  attribs: Record<string, string>;
+  /** The exact source slice of the node (computed lazily on first read). */
+  html: string;
+  /** Text content, whitespace collapsed (computed lazily on first read). */
+  text: string;
+  /** The raw htmlparser2 element (advanced/unstable escape hatch). */
+  node: unknown;
+}
+
+export interface TransformOps {
+  /** Rename each matched element (its closing tag too, when one exists). */
+  renameTag?: string;
+  /** Attribute name(s) to remove (all source occurrences). */
+  removeAttr?: string | string[];
+  /** Attributes to set: value replaced or attribute added; values escaped. */
+  setAttr?: Record<string, string>;
+  /** Class token(s) to remove; the attribute is dropped when emptied. */
+  removeClass?: string | string[];
+  /** Class token(s) to add (deduplicated; attribute created if missing). */
+  addClass?: string | string[];
+  /** Re-target each match to its n-th element ancestor first (--parent). */
+  parent?: number;
+}
+
+export interface Edit {
+  /** 0-based offset in the original html where this splice starts. */
+  start: number;
+  /** 0-based exclusive end offset in the original html. */
+  end: number;
+  /** The replaced source text (`doc.html.slice(start, end)`). */
+  before: string;
+  /** The replacement text. */
+  after: string;
+}
+
+export interface TransformResult {
+  /** The rewritten document; bytes outside the edits are untouched. */
+  html: string;
+  /** One record per splice, in document order. Empty when nothing matched. */
+  edits: Edit[];
+}
+
+export interface Document {
+  /** The source string this document was parsed from. */
+  html: string;
+  /**
+   * Run a CSS selector against the parsed tree; callable any number of times
+   * per document. Returns one Match per hit in DOM order. Throws on a
+   * selector css-select cannot parse.
+   */
+  search(selector: string, opts?: SearchOptions): Match[];
+  /**
+   * Rewrite the matched elements. Ops compose in a fixed pipeline order
+   * (rename → remove-attr → set-attr → remove-class → add-class) regardless
+   * of key order; matching runs once against the original tree. Throws on an
+   * invalid selector, invalid op names, or an empty ops object.
+   */
+  transform(selector: string, ops: TransformOps): TransformResult;
+}
+
+/**
+ * Parse an HTML string once (tracking source positions) and return a document
+ * handle to query with `search`. Throws a TypeError on non-string input.
+ */
+export function parse(html: string): Document;

package/lib.js

@@ -0,0 +1,403 @@
+'use strict';
+
+// cssgrep's engine as a library. The public API is `parse(html)`: parse an
+// HTML string once (with source positions) and get back a document handle
+// whose `search(selector, opts)` can run any number of selectors against the
+// same tree — parse once, query many. The CLI (cli.js) is a consumer of this
+// API; everything here is string-in/data-out — file discovery, output
+// formatting and process concerns live in the CLI.
+//
+// Stability contract: `parse`, the document's `html`/`search`, and the match
+// fields are the public surface. Each match also carries `node`, the raw
+// htmlparser2 element, as an advanced escape hatch — its shape belongs to
+// htmlparser2, not to cssgrep. The document's `dom`/`lineStarts`/`position`
+// and the other module exports are internals shared with the CLI.
+
+const { parseDocument } = require('htmlparser2');
+const { selectAll } = require('css-select');
+
+// Precompute the offset at which each line starts, so offset->line/col is a
+// binary search rather than a re-scan per match.
+function lineIndex(src) {
+  const starts = [0];
+  for (let i = 0; i < src.length; i++) {
+    if (src.charCodeAt(i) === 10 /* \n */) starts.push(i + 1);
+  }
+  return starts;
+}
+
+function offsetToPosition(starts, src, offset) {
+  // binary search for the greatest line start <= offset
+  let lo = 0, hi = starts.length - 1;
+  while (lo < hi) {
+    const mid = (lo + hi + 1) >> 1;
+    if (starts[mid] <= offset) lo = mid;
+    else hi = mid - 1;
+  }
+  const lineStart = starts[lo];
+  let lineEnd = src.indexOf('\n', lineStart);
+  if (lineEnd === -1) lineEnd = src.length;
+  // strip a trailing \r so CRLF files render cleanly
+  let text = src.slice(lineStart, lineEnd);
+  if (text.endsWith('\r')) text = text.slice(0, -1);
+  return {
+    line: lo + 1,            // 1-based
+    // col in UTF-16 code units: a JS string index into `text`, used by the
+    // highlight math. bcol in bytes: what gets printed — vim's grepformat %c
+    // and terminals count bytes, so non-ASCII text before the match would
+    // otherwise land the cursor short.
+    col: offset - lineStart + 1, // 1-based
+    bcol: Buffer.byteLength(src.slice(lineStart, offset), 'utf8') + 1, // 1-based
+    text,
+  };
+}
+
+// Text of a 1-based line number, with the trailing \r stripped (CRLF), plus the
+// line's byte start (so a node's offset maps to a column within it).
+function lineTextAt(starts, src, lineNo) {
+  const lineStart = starts[lineNo - 1];
+  let lineEnd = src.indexOf('\n', lineStart);
+  if (lineEnd === -1) lineEnd = src.length;
+  let text = src.slice(lineStart, lineEnd);
+  if (text.endsWith('\r')) text = text.slice(0, -1);
+  return { lineStart, text };
+}
+
+// Concatenate the text of a node and all its descendants (dependency-free,
+// rather than pulling in domutils).
+function textOf(node) {
+  if (node.type === 'text') return node.data || '';
+  if (!node.children) return '';
+  let s = '';
+  for (const child of node.children) s += textOf(child);
+  return s;
+}
+
+const collapseWs = s => s.replace(/\s+/g, ' ').trim();
+
+const isElement = n => n && (n.type === 'tag' || n.type === 'script' || n.type === 'style');
+
+// Climb n element levels from el, clamping at the document root.
+function ancestor(el, n) {
+  let node = el;
+  for (let k = 0; k < n; k++) {
+    if (!isElement(node.parent)) break;
+    node = node.parent;
+  }
+  return node;
+}
+
+// `parent` re-points each match to its n-th ancestor; dedup by identity so a
+// shared container is reported once, preserving first-seen order.
+function retarget(nodes, opts) {
+  if (!opts.parent) return nodes;
+  const seen = new Set();
+  const result = [];
+  for (const el of nodes) {
+    const a = ancestor(el, opts.parent);
+    if (!seen.has(a)) { seen.add(a); result.push(a); }
+  }
+  return result;
+}
+
+// --- rewrite machinery --------------------------------------------------------
+
+// Lex one opening tag starting at `start` (which must point at its `<`).
+// htmlparser2 records node offsets but not attribute offsets, so this tiny
+// lexer recovers the spans transform() needs. Quoted attribute values may
+// contain `>`; `/` acts as whitespace (self-closing slash, sloppy `a/b`).
+// Returns { nameStart, nameEnd, attrs, end } — offsets into `src`, `end` just
+// past the closing `>`; each attr is { name, start, end, nameEnd, vStart,
+// vEnd } (vStart -1 for valueless attributes; value offsets exclude quotes).
+function lexOpenTag(src, start) {
+  let i = start + 1;
+  const nameStart = i;
+  while (i < src.length && !/[\s/>]/.test(src[i])) i++;
+  const nameEnd = i;
+  const attrs = [];
+  for (;;) {
+    while (i < src.length && /[\s/]/.test(src[i])) i++;
+    if (i >= src.length || src[i] === '>') break;
+    const aStart = i;
+    while (i < src.length && !/[\s=/>]/.test(src[i])) i++;
+    const aNameEnd = i;
+    let vStart = -1, vEnd = -1;
+    let j = i;
+    while (j < src.length && /\s/.test(src[j])) j++;
+    if (src[j] === '=') {
+      j++;
+      while (j < src.length && /\s/.test(src[j])) j++;
+      if (src[j] === '"' || src[j] === "'") {
+        const quote = src[j];
+        vStart = ++j;
+        while (j < src.length && src[j] !== quote) j++;
+        vEnd = j;
+        if (j < src.length) j++;            // past the closing quote
+      } else {
+        vStart = j;
+        while (j < src.length && !/[\s>]/.test(src[j])) j++;
+        vEnd = j;
+      }
+      i = j;
+    }
+    attrs.push({
+      name: src.slice(aStart, aNameEnd),
+      start: aStart,
+      end: vStart === -1 ? aNameEnd : i,
+      nameEnd: aNameEnd,
+      vStart,
+      vEnd,
+    });
+  }
+  return { nameStart, nameEnd, attrs, end: i + 1 };
+}
+
+const escAttr = v => v.replace(/&/g, '&amp;').replace(/"/g, '&quot;');
+
+// Each helper takes the opening tag's source text and returns it edited (or
+// unchanged). They re-lex per call — the string is one tag, so that's cheap —
+// which lets ops compose sequentially without span bookkeeping.
+
+function removeAttrFromTag(t, name) {
+  const lc = name.toLowerCase();
+  for (;;) {                                // all occurrences (source may dupe)
+    const a = lexOpenTag(t, 0).attrs.find(a => a.name.toLowerCase() === lc);
+    if (!a) return t;
+    let ws = a.start;
+    while (ws > 0 && /\s/.test(t[ws - 1])) ws--;
+    t = t.slice(0, ws) + t.slice(a.end);
+  }
+}
+
+function setAttrInTag(t, name, value) {
+  const lc = name.toLowerCase();
+  const a = lexOpenTag(t, 0).attrs.find(a => a.name.toLowerCase() === lc);
+  const rendered = `${name}="${escAttr(value)}"`;
+  if (a) return t.slice(0, a.start) + rendered + t.slice(a.end);
+  const tail = /\s*\/?>$/.exec(t);
+  return t.slice(0, tail.index) + ' ' + rendered + t.slice(tail.index);
+}
+
+// Class edits rewrite the whole class attribute (original name case kept,
+// value re-quoted with `"`) — the one attribute being edited is normalized,
+// every other byte of the tag stays put.
+function classEdit(t, mutate) {
+  const a = lexOpenTag(t, 0).attrs.find(a => a.name.toLowerCase() === 'class');
+  const tokens = a && a.vStart >= 0
+    ? t.slice(a.vStart, a.vEnd).split(/\s+/).filter(Boolean)
+    : [];
+  const next = mutate(tokens);
+  if (next === null) return t;              // no change
+  if (!a) return setAttrInTag(t, 'class', next.join(' '));
+  if (next.length === 0) return removeAttrFromTag(t, t.slice(a.start, a.nameEnd));
+  return t.slice(0, a.start)
+    + `${t.slice(a.start, a.nameEnd)}="${next.join(' ')}"`
+    + t.slice(a.end);
+}
+
+const addClassToTag = (t, cls) =>
+  classEdit(t, tokens => (tokens.includes(cls) ? null : [...tokens, cls]));
+
+const removeClassFromTag = (t, cls) =>
+  classEdit(t, tokens => (tokens.includes(cls) ? tokens.filter(c => c !== cls) : null));
+
+// Validate and normalize a transform ops object. Names must not be able to
+// break out of the tag or its quoting; values are escaped at render time.
+const OP_KEYS = ['renameTag', 'removeAttr', 'setAttr', 'removeClass', 'addClass', 'parent'];
+const TAG_NAME = /^[a-zA-Z][^\s/>'"=]*$/;
+const ATTR_NAME = /^[^\s/>'"=]+$/;
+const CLASS_TOKEN = /^[^\s'"&<>]+$/;
+
+function normalizeOps(ops) {
+  for (const k of Object.keys(ops)) {
+    if (!OP_KEYS.includes(k)) throw new TypeError(`unknown transform op: ${k}`);
+  }
+  const toArr = v => (v == null ? [] : Array.isArray(v) ? v : [v]).map(String);
+  const norm = {
+    renameTag: ops.renameTag != null ? String(ops.renameTag) : null,
+    removeAttr: toArr(ops.removeAttr),
+    setAttr: {},
+    removeClass: toArr(ops.removeClass),
+    addClass: toArr(ops.addClass),
+    parent: ops.parent || 0,
+  };
+  if (ops.parent != null && (!Number.isInteger(ops.parent) || ops.parent < 0)) {
+    throw new TypeError('parent must be a non-negative integer');
+  }
+  if (norm.renameTag !== null && !TAG_NAME.test(norm.renameTag)) {
+    throw new TypeError(`invalid tag name: ${norm.renameTag}`);
+  }
+  for (const k of norm.removeAttr) {
+    if (!ATTR_NAME.test(k)) throw new TypeError(`invalid attribute name: ${k}`);
+  }
+  if (ops.setAttr != null) {
+    if (typeof ops.setAttr !== 'object' || Array.isArray(ops.setAttr)) {
+      throw new TypeError('setAttr must be an object of name: value');
+    }
+    for (const [k, v] of Object.entries(ops.setAttr)) {
+      if (!ATTR_NAME.test(k)) throw new TypeError(`invalid attribute name: ${k}`);
+      norm.setAttr[k] = String(v);
+    }
+  }
+  for (const c of [...norm.removeClass, ...norm.addClass]) {
+    if (!CLASS_TOKEN.test(c)) throw new TypeError(`invalid class name: ${c}`);
+  }
+  const count = (norm.renameTag ? 1 : 0) + norm.removeAttr.length
+    + Object.keys(norm.setAttr).length + norm.removeClass.length + norm.addClass.length;
+  if (count === 0) throw new TypeError('no transform operations given');
+  return norm;
+}
+
+// Ops compose in a fixed documented pipeline order — deterministic no matter
+// how the caller (or argv) ordered them: rename → remove-attr → set-attr →
+// remove-class → add-class.
+function applyOpsToOpenTag(t, ops) {
+  if (ops.renameTag) {
+    const lx = lexOpenTag(t, 0);
+    t = t.slice(0, lx.nameStart) + ops.renameTag + t.slice(lx.nameEnd);
+  }
+  for (const k of ops.removeAttr) t = removeAttrFromTag(t, k);
+  for (const [k, v] of Object.entries(ops.setAttr)) t = setAttrInTag(t, k, v);
+  for (const c of ops.removeClass) t = removeClassFromTag(t, c);
+  for (const c of ops.addClass) t = addClassToTag(t, c);
+  return t;
+}
+
+// Parse an HTML string once and return a document handle. `search(selector,
+// opts)` can then run any number of selectors against the same tree — the
+// parse and the line index are paid once per document, not per query.
+//
+// Each search returns one match object per hit, in DOM order:
+//
+//   { start, end, line, col, tag, attribs, html, text, node }
+//
+// - start/end: 0-based offsets of the node in `html` (end exclusive), so
+//   `html.slice(start, end)` is the exact source of the match.
+// - line/col: 1-based position of the opening `<`; col counts *bytes* (what
+//   vim's grepformat %c expects), matching the CLI's -n output.
+// - tag: lowercased tag name; attribs: a fresh plain object copy.
+// - html: the exact source slice; text: collapsed text content.
+// - node: the raw htmlparser2 element (advanced/unstable escape hatch).
+//
+// line/col/html/text are lazy getters — a match costs nothing for fields
+// never read (aggregate-style consumers skip the position math entirely),
+// while JSON.stringify still serializes complete records.
+//
+// opts.parent (n >= 1) re-targets each match to its n-th element ancestor,
+// deduplicated by identity — the CLI's --parent.
+//
+// parse throws on non-string input; search throws on selectors css-select
+// cannot parse.
+function parse(html) {
+  if (typeof html !== 'string') throw new TypeError('html must be a string');
+  const dom = parseDocument(html, {
+    withStartIndices: true,
+    withEndIndices: true,
+  });
+  let starts = null;
+  const lineStarts = () => starts || (starts = lineIndex(html));
+  const position = offset => offsetToPosition(lineStarts(), html, offset);
+  return {
+    html,
+    // Internal (unstable): the raw htmlparser2 document and the cached
+    // line-index helpers, shared with the CLI.
+    dom,
+    lineStarts,
+    position,
+    search(selector, opts = {}) {
+      if (typeof selector !== 'string' || selector.trim() === '') {
+        throw new TypeError('selector must be a non-empty string');
+      }
+      if (opts.parent != null && (!Number.isInteger(opts.parent) || opts.parent < 0)) {
+        throw new TypeError('opts.parent must be a non-negative integer');
+      }
+      const targets = retarget(selectAll(selector, dom), { parent: opts.parent || 0 });
+      return targets.map(el => {
+        const start = el.startIndex == null ? 0 : el.startIndex;
+        const end = (el.endIndex == null ? start : el.endIndex) + 1; // exclusive
+        const match = {
+          start,
+          end,
+          tag: el.name,
+          attribs: el.attribs ? { ...el.attribs } : {},
+          get line() { return position(start).line; },
+          get col() { return position(start).bcol; },
+          get html() { return html.slice(start, end); },
+          get text() { return collapseWs(textOf(el)); },
+        };
+        // The escape hatch is a reference, not data: htmlparser2 nodes link
+        // parent/children circularly, so keep `node` non-enumerable to let
+        // JSON.stringify serialize matches cleanly.
+        Object.defineProperty(match, 'node', { value: el });
+        return match;
+      });
+    },
+    // Rewrite the elements matched by `selector` and return { html, edits }:
+    // the edited document plus one splice record per changed span
+    // ({ start, end, before, after }, offsets into the ORIGINAL html). Only
+    // the matched tags' bytes change — everything else passes through
+    // untouched (the fidelity contract; see ROADMAP Phase 9).
+    //
+    // ops: { renameTag, removeAttr, setAttr, removeClass, addClass, parent }.
+    // Scalars or arrays where it makes sense; ops compose in the fixed
+    // pipeline order rename → remove-attr → set-attr → remove-class →
+    // add-class regardless of key order. The selector runs once against the
+    // parsed tree; edits never re-match. Throws on invalid selectors, op
+    // names, or an empty ops object.
+    transform(selector, ops = {}) {
+      if (typeof selector !== 'string' || selector.trim() === '') {
+        throw new TypeError('selector must be a non-empty string');
+      }
+      const norm = normalizeOps(ops);
+      const targets = retarget(selectAll(selector, dom), { parent: norm.parent });
+      const edits = [];
+      for (const el of targets) {
+        if (el.startIndex == null) continue;
+        const start = el.startIndex;
+        const open = lexOpenTag(html, start);
+        const before = html.slice(start, open.end);
+        const after = applyOpsToOpenTag(before, norm);
+        if (after !== before) edits.push({ start, end: open.end, before, after });
+        if (norm.renameTag) {
+          // Rename the closing tag too — but only when one explicitly exists:
+          // voids (<img>), self-closing (<x/>) and parser-implied closes
+          // (<li> without </li>) end without their own </name>.
+          const nodeEnd = (el.endIndex == null ? start : el.endIndex) + 1;
+          const m = /<\/([^\s>]+)(\s*)>$/.exec(html.slice(start, nodeEnd));
+          if (m && m[1].toLowerCase() === el.name) {
+            edits.push({
+              start: start + m.index,
+              end: nodeEnd,
+              before: m[0],
+              after: `</${norm.renameTag}${m[2]}>`,
+            });
+          }
+        }
+      }
+      edits.sort((a, b) => a.start - b.start);
+      for (let i = 1; i < edits.length; i++) {
+        if (edits[i].start < edits[i - 1].end) {
+          throw new Error('internal error: overlapping edits');
+        }
+      }
+      let out = html;
+      for (let i = edits.length - 1; i >= 0; i--) {
+        out = out.slice(0, edits[i].start) + edits[i].after + out.slice(edits[i].end);
+      }
+      return { html: out, edits };
+    },
+  };
+}
+
+module.exports = {
+  parse,
+  // Internal helpers, shared with cli.js; not part of the stable API.
+  lineIndex,
+  offsetToPosition,
+  lineTextAt,
+  textOf,
+  collapseWs,
+  isElement,
+  ancestor,
+  retarget,
+};