@knowvah/dot-engine 2.6.10 → 2.6.12

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@knowvah/dot-engine",
-  "version": "2.6.10",
+  "version": "2.6.12",
   "description": "Faithful TypeScript port of Graphviz (dot, neato, fdp, sfdp, circo, twopi, osage, patchwork) — no C: no native binary, no WASM. Oracle-verified against the C implementation, runs in the browser.",
   "keywords": [
     "graphviz",

package/src/async/render-async.ts

@@ -18,6 +18,7 @@ import type { EngineName } from '../gvc/context.js';
 import { createDefaultContext } from '../gvc/default-context.js';
 import { checkRenderArgs } from '../render/public.js';
 import { render as deviceRender } from '../gvc/device.js';
+import { resetRenderState } from '../common/render-state.js';
 import type { ImageSizer } from '../common/htmltable-types.js';
 import type { ImageResolver } from '../gvc/image-resolver.js';
 import type { OutputFormat, RenderOptions } from '../render/public.js';
@@ -148,6 +149,11 @@ export async function renderAsync(
   const engine: EngineName = opts?.engine ?? 'dot';
   try {
     const { fontIssues, sizes, bytes } = await prefetchAll(g, opts);
+    // AD-4: reset after the last await, so the synchronous layout + render
+    // below starts from C's initial process state even when other renders ran
+    // while the prefetch was pending. renderSvgAsync and renderSvgInto reach
+    // this through renderAsync.
+    resetRenderState();
     // Created after fonts settle so the measurer sees the loaded faces.
     const ctx = createDefaultContext();
     if (opts?.imageSizer !== undefined) {

package/src/common/agerr.ts

@@ -0,0 +1,96 @@
+// SPDX-License-Identifier: EPL-2.0
+
+/**
+ * The one stderr writer for messages C prints through cgraph's agerr family
+ * (`agerr`, `agerrorf`, `agwarningf`, `AGPREV` continuations). Sites whose C
+ * counterpart is `fprintf(stderr)`/`fputs`/`printf` do not use it.
+ */
+
+import { registerRenderState } from './render-state.js';
+
+/** `agerrlevel_t` as the port uses it: AGERR, AGWARN, AGPREV. */
+export type AgerrLevel = 'error' | 'warning' | 'prev';
+
+/** The console channel a line goes to. */
+export type AgerrChannel = 'warn' | 'error';
+
+const PREFIX_ERROR = 'Error';
+const PREFIX_WARNING = 'Warning';
+const PREFIX_SEP = ': ';
+const DEL = 0x7f;
+const LAST_CNTRL = 0x1f;
+const WHITESPACE = new Set([0x09, 0x0a, 0x0b, 0x0c, 0x0d]);
+const OCTAL_WIDTH = 3;
+const OCTAL_RADIX = 8;
+
+/**
+ * The level of the last non-AGPREV message, which AGPREV continues: C's
+ * `agerrno`, zero-initialised, i.e. AGWARN. Reset per render (AD-4).
+ * @see lib/cgraph/agerror.c:agerrno
+ */
+let lastLevel: 'error' | 'warning' = 'warning';
+
+registerRenderState(() => {
+  lastLevel = 'warning';
+});
+
+/**
+ * `default_usererrf`: every character that `gv_iscntrl(c) && !gv_isspace(c)`
+ * (0x00-0x1F except `\t\n\v\f\r`, and 0x7F) is written as `\ooo`.
+ * @see lib/cgraph/agerror.c:default_usererrf
+ */
+export function escapeControl(message: string): string {
+  let out = '';
+  for (const ch of message) {
+    const c = ch.codePointAt(0)!;
+    if ((c <= LAST_CNTRL || c === DEL) && !WHITESPACE.has(c)) {
+      out += `\\${c.toString(OCTAL_RADIX).padStart(OCTAL_WIDTH, '0')}`;
+    } else out += ch;
+  }
+  return out;
+}
+
+/**
+ * Write one already-assembled stderr line (prefix included) on a console
+ * channel, escaped as `default_usererrf` would.
+ */
+export function agerrLineOn(channel: AgerrChannel, line: string): void {
+  if (channel === 'warn') console.warn(escapeControl(line));
+  else console.error(escapeControl(line));
+}
+
+/**
+ * The part of C's `agerr(level, ...)` that is not formatting: the `Error: `/
+ * `Warning: ` prefix (none for AGPREV), escaping, and the console channel
+ * (`console.warn` for warnings, `console.error` otherwise; AGPREV continues
+ * the previous level's channel).
+ * @see lib/cgraph/agerror.c:out
+ */
+export function agerr(level: AgerrLevel, msg: string, channel?: AgerrChannel): void {
+  if (level === 'prev') {
+    agerrLineOn(channel ?? (lastLevel === 'warning' ? 'warn' : 'error'), msg);
+    return;
+  }
+  lastLevel = level;
+  const prefix = level === 'error' ? PREFIX_ERROR : PREFIX_WARNING;
+  agerrLineOn(channel ?? (level === 'warning' ? 'warn' : 'error'), `${prefix}${PREFIX_SEP}${msg}`);
+}
+
+/** @see lib/cgraph/agerror.c:agerrorf */
+export const agerrorf = (msg: string, channel?: AgerrChannel): void => agerr('error', msg, channel);
+
+/**
+ * @see lib/cgraph/agerror.c:agwarningf
+ * `channel` overrides the console channel for sites that have always written
+ * their warning on `console.error`.
+ */
+export const agwarningf = (msg: string, channel?: AgerrChannel): void =>
+  agerr('warning', msg, channel);
+
+/**
+ * A finished line whose prefix is already part of the text. The channel
+ * follows the prefix (`Warning` -> `console.warn`) unless `channel` is given.
+ */
+export function agerrFinished(line: string, channel?: AgerrChannel): void {
+  agerrLineOn(channel ?? (line.startsWith(PREFIX_WARNING) ? 'warn' : 'error'), line);
+}

package/src/common/arrows-shapes-poly.ts

@@ -11,6 +11,7 @@
 import type { Point } from '../model/geom.js';
 import { ccwrotatepf, cwrotatepf } from '../model/geom.js';
 import { bezier } from './bezier.js';
+import { fma } from './fma.js';
 import type { GenResult } from './arrows-shapes-util.js';
 import {
   vsub, vadd, miterShape, negUnit, backwardDelta, axialProjection,
@@ -206,14 +207,27 @@ function cubicHalf(c: Point[], rightHalf: boolean): Point[] {
   return half;
 }
 
-/** Build the four Bézier control points for the curve arms. */
+/**
+ * Build the four Bézier control points for the curve arms.
+ *
+ * The oracle (libgvc 16.1 `_arrow_type_curve`) contracts only the
+ * `p.x +/- 0.95 * v.x` term into `fmadd(v.x, +/-0.95, p.x)`; every other add
+ * is a plain fadd, and `w.x * 4.0 / 3.0` is `(w.x * 4) / 3` (fmul, fdiv).
+ * Only the x arms carry the 0.95 term, so AF[1].y / AF[2].y stay unfused.
+ * @see lib/common/arrows.c:arrow_type_curve
+ */
 function curveControlPoints(p: Point, v: Point, w: Point, flag: number): Point[] {
   const AF0: Point = { x: p.x + v.x + w.x, y: p.y + v.y + w.y };
   const AF3: Point = { x: p.x - v.x + w.x, y: p.y - v.y + w.y };
-  const s = flag & ARR_MOD_INV ? 1 : -1; // inv bulges the curve the other way
-  const AF1: Point = { x: p.x + 0.95 * v.x + w.x + s * w.x * (4.0 / 3.0), y: AF0.y + s * w.y * (4.0 / 3.0) };
-  const AF2: Point = { x: p.x - 0.95 * v.x + w.x + s * w.x * (4.0 / 3.0), y: AF3.y + s * w.y * (4.0 / 3.0) };
-  return [AF0, AF1, AF2, AF3];
+  const bulgeX = (w.x * 4.0) / 3.0;
+  const bulgeY = (w.y * 4.0) / 3.0;
+  const x1 = fma(v.x, 0.95, p.x) + w.x;
+  const x2 = fma(v.x, -0.95, p.x) + w.x;
+  if (flag & ARR_MOD_INV) { /* ----(-| */
+    return [AF0, { x: x1 + bulgeX, y: AF0.y + bulgeY }, { x: x2 + bulgeX, y: AF3.y + bulgeY }, AF3];
+  }
+  /* ----)-| */
+  return [AF0, { x: x1 - bulgeX, y: AF0.y - bulgeY }, { x: x2 - bulgeX, y: AF3.y - bulgeY }, AF3];
 }
 
 /** @see lib/common/arrows.c:arrow_type_curve */

package/src/common/color.ts

@@ -7,6 +7,7 @@
 
 import { isExactlyEqual } from '../util/math.js';
 import { COLOR_LIB } from './colorData.js';
+import { registerRenderState } from './render-state.js';
 
 // ---------------------------------------------------------------------------
 // Public types
@@ -37,6 +38,7 @@ interface HsvaTuple { H: number; S: number; V: number; A: number }
 // ---------------------------------------------------------------------------
 
 let colorScheme: string = '';
+registerRenderState(() => { colorScheme = ''; });
 
 /** @see lib/common/colxlate.c:setColorScheme */
 export function setColorScheme(scheme: string): void {
@@ -256,17 +258,43 @@ function setBlackOpaque(color: GVColor, targetType: ColorType): void {
 // Hex and HSV string parsers
 // ---------------------------------------------------------------------------
 
-/** Parse "#rrggbb", "#rrggbbaa", or "#rgb". Returns null on failure. */
+/** C isspace in the "C" locale. */
+const SSCANF_SPACE = new Set([' ', '\t', '\n', '\v', '\f', '\r']);
+
+/**
+ * `sscanf(p, "#" + "%<width>x" * max)`: the values converted before the first
+ * failure. Each `%x` skips leading white space then takes up to `width` hex
+ * digits, so a short or odd-length run still converts ("#12345" gives 0x12,
+ * 0x34, 0x5). Sign and `0x` prefixes are not modelled.
+ * @see lib/common/colxlate.c:colorxlate
+ */
+function scanHexGroups(p: string, width: number, max: number): number[] {
+  const out: number[] = [];
+  if (p.charAt(0) !== '#') return out;
+  let i = 1;
+  while (out.length < max) {
+    while (SSCANF_SPACE.has(p.charAt(i))) i++;
+    let j = i;
+    while (j < p.length && j - i < width && /[0-9a-fA-F]/.test(p.charAt(j))) j++;
+    if (j === i) break;
+    out.push(parseInt(p.slice(i, j), 16));
+    i = j;
+  }
+  return out;
+}
+
+/**
+ * Parse "#rrggbb", "#rrggbbaa" (any run of at least three two-digit groups),
+ * or the 4-character "#rgb" form, as colorxlate's sscanf calls do. Returns
+ * null on failure.
+ */
 function parseHexColor(p: string): RgbaTuple | null {
-  if (p.charAt(0) !== '#') return null;
-  const m8 = /^#([0-9a-fA-F]{2})([0-9a-fA-F]{2})([0-9a-fA-F]{2})([0-9a-fA-F]{2})$/.exec(p);
-  if (m8) return { r: parseInt(m8[1], 16), g: parseInt(m8[2], 16), b: parseInt(m8[3], 16), a: parseInt(m8[4], 16) };
-  const m6 = /^#([0-9a-fA-F]{2})([0-9a-fA-F]{2})([0-9a-fA-F]{2})$/.exec(p);
-  if (m6) return { r: parseInt(m6[1], 16), g: parseInt(m6[2], 16), b: parseInt(m6[3], 16), a: 255 };
-  const m3 = /^#([0-9a-fA-F])([0-9a-fA-F])([0-9a-fA-F])$/.exec(p);
-  if (!m3) return null;
-  const r = parseInt(m3[1], 16); const g = parseInt(m3[2], 16); const b = parseInt(m3[3], 16);
-  return { r: r | (r << 4), g: g | (g << 4), b: b | (b << 4), a: 255 };
+  const g = scanHexGroups(p, 2, 4);
+  if (g.length >= 3) return { r: g[0]!, g: g[1]!, b: g[2]!, a: g[3] ?? 255 };
+  const short = p.length === 4 ? scanHexGroups(p, 1, 3) : [];
+  if (short.length !== 3) return null;
+  const [r, gg, b] = short as [number, number, number];
+  return { r: r | (r << 4), g: gg | (gg << 4), b: b | (b << 4), a: 255 };
 }
 
 /** Clamp a number to [0, 1]. */

package/src/common/crand.ts

@@ -21,11 +21,17 @@
  * @see lib/util/random.c:gv_random / gv_permutation (15.0.0)
  */
 
+import { registerRenderState } from './render-state.js';
+
 /** RAND_MAX on the reference platform. */
 export const C_RAND_MAX = 2147483647;
 
-/** Current stream state; unseeded C rand() behaves as srand(1). */
-let state = 1;
+/** Unseeded C rand() behaves as srand(1). */
+const UNSEEDED_STATE = 1;
+
+/** Current stream state. */
+let state = UNSEEDED_STATE;
+registerRenderState(() => { state = UNSEEDED_STATE; });
 
 /** srand(3). */
 export function csrand(seed: number): void {

package/src/common/emit-once.ts

@@ -0,0 +1,27 @@
+// SPDX-License-Identifier: EPL-2.0
+
+/**
+ * Once-per-process message gate. C keeps the seen strings in a file static
+ * dictionary that is cleared only by `emit_once_reset` in `gvFreeContext`; one
+ * native run is one process, so the port clears it per render (AD-4).
+ *
+ * @see lib/common/emit.c:emit_once
+ */
+
+import { registerRenderState } from './render-state.js';
+
+const seen = new Set<string>();
+
+registerRenderState(() => {
+  seen.clear();
+});
+
+/**
+ * True the first time `str` is offered since the last reset, false after.
+ * @see lib/common/emit.c:emit_once
+ */
+export function emitOnce(str: string): boolean {
+  if (seen.has(str)) return false;
+  seen.add(str);
+  return true;
+}

package/src/common/graph-init.ts

@@ -29,6 +29,7 @@ import type { RatioKind } from '../model/layoutParams.js';
 import { doGraphLabel } from '../layout/dot/graph-label.js';
 import { mapbool, setClType, LOCAL, GLOBAL, NOCLUST } from '../layout/dot/rank.js';
 import { cround } from './arith.js';
+import { agwarningf } from './agerr.js';
 
 // ---------------------------------------------------------------------------
 // Constants — @see lib/common/const.h
@@ -88,7 +89,7 @@ export function findCharset(g: Graph): number {
   }
   if (p === 'big-5' || p === 'big5') return CHAR_BIG5;
   if (p === 'utf-8' || p === 'utf8') return CHAR_UTF8;
-  console.error(`Warning: Unsupported charset "${raw}" - assuming utf-8`);
+  agwarningf(`Unsupported charset "${raw}" - assuming utf-8`, 'error');
   return CHAR_UTF8;
 }
 

package/src/common/html-entities.ts

@@ -260,40 +260,92 @@ export const HTML_ENTITIES: Readonly<Record<string, number>> = {
   "zwnj": 8204,
 };
 
-/** The five basic XML/HTML entities → code point. */
-const BASIC_ENTITIES: Readonly<Record<string, number>> = {
-  amp: 0x26, lt: 0x3c, gt: 0x3e, quot: 0x22, apos: 0x27,
-};
+/**
+ * Longest name htmlEntity looks up: its scan for the `;` stops after this
+ * many characters.
+ * @see lib/common/entities.h:ENTITY_NAME_LENGTH_MAX
+ */
+const ENTITY_NAME_LENGTH_MAX = 8;
 
-/** Decode a numeric entity body (`#123`, `#x1F`) to its code point, or null. */
-function numericEntityValue(body: string): number | null {
-  const hex = body[1] === 'x' || body[1] === 'X';
-  const v = hex ? parseInt(body.slice(2), 16) : parseInt(body.slice(1), 10);
-  return Number.isFinite(v) && v > 0 && v <= 0x10ffff ? v : null;
+/** htmlEntity's result: the value (0 for none) and where scanning resumes. */
+interface EntityScan {
+  value: number;
+  next: number;
+}
+
+/**
+ * The `#` branch of htmlEntity, `at` on the `#`. It reads up to six decimal
+ * or five hex digits (`x` or `X`); the `;` must come within the first eight
+ * characters, else nothing is consumed. A `;` right after the digits, even
+ * none, is consumed with value `n`, so `&#;` keeps its `&` and loses `#;`.
+ * @see lib/common/utils.c:htmlEntity
+ */
+function numericEntity(s: string, at: number): EntityScan {
+  const hex = s[at + 1] === 'x' || s[at + 1] === 'X';
+  const digit = hex ? /^[0-9A-Fa-f]$/ : /^[0-9]$/;
+  let n = 0;
+  let i = hex ? 2 : 1;
+  for (; i < 8; i++) {
+    const c = s[at + i] ?? '';
+    if (!digit.test(c)) break;
+    n = n * (hex ? 16 : 10) + parseInt(c, 16);
+  }
+  if (i < 8 && s[at + i] === ';') return { value: n, next: at + i + 1 };
+  return { value: 0, next: at };
+}
+
+/**
+ * The name branch of htmlEntity, `at` on the first name character: a `;`
+ * within ENTITY_NAME_LENGTH_MAX characters ends a name looked up in
+ * entities.h (case-sensitive; no `apos`).
+ * @see lib/common/utils.c:htmlEntity
+ */
+function namedEntity(s: string, at: number): EntityScan {
+  for (let i = 0; i < ENTITY_NAME_LENGTH_MAX; i++) {
+    const c = s[at + i];
+    if (c === undefined || c === '\0') break;
+    if (c === ';') {
+      const key = s.slice(at, at + i);
+      if (Object.hasOwn(HTML_ENTITIES, key)) return { value: HTML_ENTITIES[key]!, next: at + i + 1 };
+      break;
+    }
+  }
+  return { value: 0, next: at };
 }
 
 /**
- * Decode one entity body (`amp`, `alpha`, `#123`, `#x1F`) to its code point, or
- * null for an unknown name. Named entities cover the five basic XML entities
- * plus the full HTML 4 table (Symbol etc.). @see lib/common/utils.c:htmlEntity
+ * Check for an HTML entity at `at`, the character after an `&`.
+ * @see lib/common/utils.c:htmlEntity
  */
-function entityValue(body: string): number | null {
-  if (body[0] === '#') return numericEntityValue(body);
-  return BASIC_ENTITIES[body] ?? HTML_ENTITIES[body] ?? null;
+function htmlEntity(s: string, at: number): EntityScan {
+  return s[at] === '#' ? numericEntity(s, at) : namedEntity(s, at);
 }
 
 /**
- * Decode XML/numeric (&#NNN; / &#xHH;) and HTML 4 named (`&alpha;`) entities to
- * their characters, leaving unknown names literal — C's make_label UTF-8 branch.
- * Latin-1 inputs are decoded at the byte->string boundary by the caller (the
- * survey's render-one), so this UTF-8 branch covers them too.
- * @see lib/common/utils.c:htmlEntityUTF8 / htmlEntity
+ * Substitute entities like `&#123;` and `&amp;` with their characters: C's
+ * make_label UTF-8 branch. An `&` whose entity has no value stays (scanning
+ * resumes where htmlEntity left it). Latin-1 inputs are decoded at the
+ * byte->string boundary by the caller (the survey's render-one), so this
+ * covers them too, as does latin1ToUTF8's identical entity handling.
+ *
+ * C appends a value's bytes itself: below 0x7F one byte, below 0x7FF two,
+ * else three, truncated to `char`. For every value whose bytes are valid
+ * UTF-8 that is the character itself. 0x7F, 0x7FF, surrogates and values
+ * above 0xFFFF come out as invalid UTF-8, which a JS string cannot hold; the
+ * port keeps the character there (documented divergence).
+ * @see lib/common/utils.c:htmlEntityUTF8
+ * @see lib/common/utils.c:latin1ToUTF8
  * @see lib/common/labels.c:make_label (CHAR_LATIN1 vs default)
  */
 export function htmlEntityUTF8(s: string): string {
-  if (s.indexOf('&') < 0) return s;
-  return s.replace(/&(#x[0-9a-fA-F]+|#[0-9]+|[A-Za-z]+);/g, (m, body: string) => {
-    const v = entityValue(body);
-    return v === null ? m : String.fromCodePoint(v);
-  });
+  let amp = s.indexOf('&');
+  if (amp < 0) return s;
+  let out = '';
+  let from = 0;
+  for (; amp >= 0; amp = s.indexOf('&', from)) {
+    const e = htmlEntity(s, amp + 1);
+    out += s.slice(from, amp) + (e.value ? String.fromCodePoint(e.value) : '&');
+    from = e.next;
+  }
+  return out + s.slice(from);
 }

package/src/common/htmltable-attr-warn.ts

@@ -11,6 +11,8 @@
  * @see lib/common/htmllex.c:doAttrs
  */
 
+import { agerrFinished } from './agerr.js';
+
 /** An attribute as startElement gets it: name in source case, value decoded by expat. */
 export type AttrEntry = readonly [name: string, value: string];
 
@@ -272,10 +274,7 @@ export class StderrLines {
 }
 
 /** A finished stderr line on the console channel the port uses for it. */
-export const consoleLine = (line: string): void => {
-  if (line.startsWith('Warning')) console.warn(line);
-  else console.error(line);
-};
+export const consoleLine = (line: string): void => agerrFinished(line);
 
 /** One label's stderr: what make_html_label and make_label print, in order. */
 export const labelStderr = (): StderrLines => new StderrLines(consoleLine);

package/src/common/htmltable-emit-fill.ts

@@ -18,6 +18,7 @@ import { FillType, PenType } from '../gvc/context.js';
 import { transformPoint } from '../gvc/device.js';
 import { resolveRenderColor } from '../render/color-resolve.js';
 import { emitRoundedBezier } from './poly-shapes.js';
+import type { PlacedHtml, PlacedCell } from './htmltable-pos.js';
 import { htmlStyle } from './htmltable-style.js';
 import {
   BORDER_LEFT,
@@ -26,6 +27,7 @@ import {
   BORDER_BOTTOM,
   BORDER_MASK,
 } from './htmltable-types.js';
+import { registerRenderState } from './render-state.js';
 
 // ---------------------------------------------------------------------------
 // Gradient spec parser
@@ -136,6 +138,7 @@ export function htmlFillPenWidth(): number {
 export function resetHtmlFillPenWidth(): void {
   htmlFillPen = 1.0;
 }
+registerRenderState(resetHtmlFillPenWidth);
 
 /** Map an HTML STYLE to the C pen type: dashed wins over dotted. @see lib/common/htmltable.c:doBorder */
 function penTypeOf(style: string | undefined): PenType {
@@ -366,3 +369,48 @@ function emitSidedPolylines(
     renderer.polyline(seq.map((i) => corners[i]!), job);
   }
 }
+
+// ---------------------------------------------------------------------------
+// htmlLabelColors — the colour names C resolves while emitting an HTML label
+// ---------------------------------------------------------------------------
+
+/** @see lib/common/const.h DEFAULT_COLOR */
+const HTML_DEFAULT_COLOR = 'black';
+
+/** setFill's names: the fill (and a gradient's stop). @see htmltable.c:setFill */
+function bgNames(bgcolor: string): string[] {
+  const spec = parseGradientSpec(bgcolor);
+  return spec === null ? [bgcolor] : [spec[0], spec[1]];
+}
+
+/** One cell in emit_html_cell order: fill, border, then the child. */
+function cellNames(cell: PlacedCell): string[] {
+  if (htmlStyle(cell.style).invisible) return [];
+  const out = cell.bgcolor !== undefined ? bgNames(cell.bgcolor) : [];
+  if (cell.border > 0) out.push(cell.color ?? HTML_DEFAULT_COLOR);
+  if (cell.nested !== undefined) out.push(...tableNames(cell.nested));
+  else for (const line of cell.lines) if (line.fontColor !== null) out.push(line.fontColor);
+  return out;
+}
+
+/** One table in emit_html_tbl order: fill, rules, border, cells. */
+function tableNames(tbl: PlacedHtml): string[] {
+  if (htmlStyle(tbl.style).invisible) return [];
+  const out = tbl.bgcolor !== undefined ? bgNames(tbl.bgcolor) : [];
+  const color = tbl.color ?? HTML_DEFAULT_COLOR;
+  if (tbl.cells.some((c) => c.vruled === true || c.hruled === true)) out.push(color);
+  if (tbl.border > 0) out.push(color);
+  for (const cell of tbl.cells) out.push(...cellNames(cell));
+  return out;
+}
+
+/**
+ * The names an HTML label hands to gvrender_resolve_color, in C's order: the
+ * root table's pen first (emit_html_label), then emit_html_tbl's walk.
+ * @see lib/common/htmltable.c:emit_html_label
+ * @see lib/common/htmltable.c:emit_html_tbl
+ * @see lib/common/htmltable.c:emit_html_cell
+ */
+export function htmlLabelColors(label: PlacedHtml): string[] {
+  return [label.color ?? HTML_DEFAULT_COLOR, ...tableNames(label)];
+}

package/src/common/htmltable-emit-rules.ts

@@ -14,6 +14,7 @@ import type { RenderJob } from '../gvc/job.js';
 import type { RendererPlugin } from '../gvc/context.js';
 import type { PlacedCell, PlacedHtml } from './htmltable-pos.js';
 import { doSide, withHtmlPaint } from './htmltable-emit-fill.js';
+import { registerRenderState } from './render-state.js';
 
 // ---------------------------------------------------------------------------
 // Rule context
@@ -217,6 +218,7 @@ export function resetHtmlAnchorIds(): void {
   anchorEnv.objLabel = null;
   anchorEnv.imgscale = 'false';
 }
+registerRenderState(resetHtmlAnchorIds);
 
 /**
  * Open an anchor when the data carries href/title/target.

package/src/common/htmltable-lex-chardata.ts

@@ -12,7 +12,8 @@
  * @see lib/common/htmllex.c:characterData
  */
 
-import { htmlEntityUTF8 } from './html-entities.js';
+import { HTML_ENTITIES } from './html-entities.js';
+import { charRefNumber } from './htmltable-lex-attvalue.js';
 import type { TagChunk } from './htmltable-lex-expat.js';
 import type { Token } from './htmltable-types.js';
 
@@ -77,15 +78,48 @@ function cleanText(decoded: string) {
   return decoded.replace(/[\x00-\x1f]/g, '');
 }
 
+/** expat's predefined entities: no DTD declares others. */
+const XML_PREDEFINED: Readonly<Record<string, string>> = { lt: '<', gt: '>', amp: '&', quot: '"', apos: "'" };
+
+/** Shortest and longest name scanEntity looks up. @see lib/common/utils.c:scanEntity */
+const SCAN_ENTITY_MIN = 2;
+const SCAN_ENTITY_MAX = 8;
+
+const TEXT_REF = /&(#x[0-9A-Fa-f]+|#[0-9]+|[^&;]+);/g;
+
+/** One reference of a text run, as scanEntity then expat leave it. */
+function textRefValue(ref: string, body: string): string {
+  if (body[0] === '#') {
+    const v = charRefNumber(ref);
+    return v < 0 ? ref : String.fromCodePoint(v);
+  }
+  const n = body.length;
+  if (n >= SCAN_ENTITY_MIN && n <= SCAN_ENTITY_MAX && Object.hasOwn(HTML_ENTITIES, body)) {
+    return String.fromCodePoint(HTML_ENTITIES[body]!);
+  }
+  return Object.hasOwn(XML_PREDEFINED, body) ? XML_PREDEFINED[body]! : ref;
+}
+
+/**
+ * The characters expat hands characterData for a text run: scanEntity first
+ * rewrites each entities.h name to `&#N;` (its `&` only when no `#` follows),
+ * then expat decodes character references, of any length, and its five
+ * predefined entities. Runs expat rejects never get here (textRefError).
+ * @see lib/common/htmllex.c:findNext
+ * @see lib/common/utils.c:scanEntity
+ */
+export function decodeTextRefs(raw: string): string {
+  return raw.indexOf('&') < 0 ? raw : raw.replace(TEXT_REF, textRefValue);
+}
+
 /** The text token of a text run, if anything is left after cleaning. */
 export function textTokens(raw: string): Token[] {
-  // Decode XML/HTML entities in the text run. C's HTML lexer resolves entities
-  // via expat during tokenization (htmllex.c), so both sizing and emit see the
-  // decoded UTF-8 text. Skipping this measured the raw "&#91;el..." (9-10 chars)
-  // instead of "[el..." (6 chars), inflating edge-label vnodes and shifting
-  // layout (corpus 1949: +18.7px). Plain-string labels already decode via the
-  // make_label path, so only HTML-delimited labels diverged.
-  return charDataTokens(htmlEntityUTF8(raw));
+  // C's HTML lexer resolves entities via scanEntity and expat during
+  // tokenization (htmllex.c), so both sizing and emit see the decoded text.
+  // Skipping this measured the raw "&#91;el..." (9-10 chars) instead of
+  // "[el..." (6 chars), inflating edge-label vnodes and shifting layout
+  // (corpus 1949: +18.7px).
+  return charDataTokens(decodeTextRefs(raw));
 }
 
 /** characterData on what expat decoded: a T_string when any character survives. */