@zumer/snapdom-plugins 3.0.0 → 4.0.0

package/vector/src/writer/subset.js

@@ -0,0 +1,387 @@
+/**
+ * TrueType subsetting: keep only the glyphs a document actually used, renumbered
+ * into a dense space, and rebuild a valid sfnt around them.
+ *
+ * This is the font compiler the embedding path deliberately went without at
+ * first. It earns its risk with one number: an unsubsetted face is the whole
+ * file (Inter is ~320 kB); a Latin document touches ~80 glyphs, and those plus
+ * their metrics and a fresh cmap are tens of kB. The saving is the glyf table,
+ * which is most of a font and most of which no page ever draws.
+ *
+ * The dense renumber is the whole trick, and it is shared with `createEmbeddedFont`:
+ * that writer already assigns each glyph a compact id in first-use order as it
+ * encodes the content stream (0 is `.notdef`), and writes THAT id as the CID. So
+ * the content stream, `/W` and `/ToUnicode` all speak the subset's numbering
+ * already; this file only has to place original glyph `g` at its new index and
+ * rewrite the composite-component references that point across the renumber.
+ *
+ * What is rebuilt, and what is dropped:
+ *  - `glyf` + `loca`: the retained outlines, composite component ids remapped,
+ *    long `loca` throughout so offsets never need the ×2 short encoding.
+ *  - `hmtx` + `hhea`: advances/lsbs for the dense set; `numberOfHMetrics` = count.
+ *  - `maxp`: `numGlyphs` = count; the other ceilings are copied (they bound).
+ *  - `cmap`: a fresh format 4 (+ format 12 when an astral code point is kept),
+ *    mapping the retained CODE POINTS to their new ids. The PDF encodes through
+ *    Identity-H and ignores this, but OTS and any standalone use want it valid.
+ *  - `head`, `name`, `OS/2`, `post`: copied; `post` is rewritten to v3 (no glyph
+ *    names — they are dead weight the PDF never reads).
+ *  - Hinting (`fpgm`/`prep`/`cvt`/`gasp`) and layout (`GSUB`/`GPOS`/`GDEF`) are
+ *    DROPPED. No shaping happens on this path, so the glyphs render unhinted and
+ *    unshaped — correct for an invisible layer and honest for short furniture.
+ *
+ * CFF (`OTTO`) fonts are not subset here — their charstrings are a different
+ * compiler — so the caller embeds those whole.
+ */
+
+const U16 = (b, at) => (b[at] << 8) | b[at + 1]
+const U32 = (b, at) => ((b[at] << 24) | (b[at + 1] << 16) | (b[at + 2] << 8) | b[at + 3]) >>> 0
+const TAG = (b, at) => String.fromCharCode(b[at], b[at + 1], b[at + 2], b[at + 3])
+
+// glyf composite flags this file reads.
+const ARG_1_AND_2_ARE_WORDS = 0x0001
+const WE_HAVE_A_SCALE = 0x0008
+const MORE_COMPONENTS = 0x0020
+const WE_HAVE_AN_X_AND_Y_SCALE = 0x0040
+const WE_HAVE_A_TWO_BY_TWO = 0x0080
+
+/** The sfnt table directory as tag → {offset, length}. */
+function directory(b) {
+  const numTables = U16(b, 4)
+  const tables = new Map()
+  for (let i = 0; i < numTables; i++) {
+    const at = 12 + i * 16
+    tables.set(TAG(b, at), { offset: U32(b, at + 8), length: U32(b, at + 12) })
+  }
+  return tables
+}
+
+/** glyph `gid`'s bytes in the source, or an empty view when it has no outline. */
+function glyphSlice(b, loca, gid) {
+  const start = loca[gid]
+  const end = loca[gid + 1]
+  return end > start ? b.subarray(start, end) : b.subarray(0, 0)
+}
+
+/** loca as absolute byte offsets into glyf, in either stored format. */
+function readLoca(b, locaT, glyfOffset, numGlyphs, longLoca) {
+  const out = new Array(numGlyphs + 1)
+  for (let i = 0; i <= numGlyphs; i++) {
+    out[i] = glyfOffset + (longLoca ? U32(b, locaT.offset + i * 4) : U16(b, locaT.offset + i * 2) * 2)
+  }
+  return out
+}
+
+/**
+ * Every glyph id a composite glyph reaches, transitively. A component may itself
+ * be composite, so the walk is a queue, not one pass; a self-reference or a cycle
+ * is bounded by the `seen` set.
+ */
+function componentIds(b, loca, gid, into) {
+  const g = glyphSlice(b, loca, gid)
+  if (g.length < 10 || (U16(g, 0) & 0x8000) === 0) return // not composite (numberOfContours >= 0)
+  let at = 10
+  for (;;) {
+    const flags = U16(g, at)
+    const componentGid = U16(g, at + 2)
+    into.add(componentGid)
+    at += 4
+    at += (flags & ARG_1_AND_2_ARE_WORDS) ? 4 : 2
+    if (flags & WE_HAVE_A_SCALE) at += 2
+    else if (flags & WE_HAVE_AN_X_AND_Y_SCALE) at += 4
+    else if (flags & WE_HAVE_A_TWO_BY_TWO) at += 8
+    if (!(flags & MORE_COMPONENTS)) break
+  }
+}
+
+/** A copy of glyph `gid`'s bytes with every component id remapped through `map`. */
+function remapComposite(bytes, map) {
+  if (bytes.length < 10 || (U16(bytes, 0) & 0x8000) === 0) return bytes
+  const out = bytes.slice()
+  let at = 10
+  for (;;) {
+    const flags = U16(out, at)
+    const newGid = map.get(U16(out, at + 2))
+    out[at + 2] = (newGid >> 8) & 0xff
+    out[at + 3] = newGid & 0xff
+    at += 4
+    at += (flags & ARG_1_AND_2_ARE_WORDS) ? 4 : 2
+    if (flags & WE_HAVE_A_SCALE) at += 2
+    else if (flags & WE_HAVE_AN_X_AND_Y_SCALE) at += 4
+    else if (flags & WE_HAVE_A_TWO_BY_TWO) at += 8
+    if (!(flags & MORE_COMPONENTS)) break
+  }
+  return out
+}
+
+/** sfnt table checksum: sum of big-endian uint32 over the 4-byte-padded data. */
+function checksum(bytes) {
+  let sum = 0
+  const n = (bytes.length + 3) & ~3
+  for (let i = 0; i < n; i += 4) {
+    const v = ((bytes[i] || 0) << 24) | ((bytes[i + 1] || 0) << 16) |
+      ((bytes[i + 2] || 0) << 8) | (bytes[i + 3] || 0)
+    sum = (sum + (v >>> 0)) >>> 0
+  }
+  return sum >>> 0
+}
+
+/** A format-4 (BMP) cmap subtable body for `pairs` of [codepoint, gid], sorted. */
+function cmapFormat4(pairs) {
+  const segments = []
+  let i = 0
+  while (i < pairs.length) {
+    const [startCp, startGid] = pairs[i]
+    let j = i
+    while (j + 1 < pairs.length &&
+      pairs[j + 1][0] === pairs[j][0] + 1 && pairs[j + 1][1] === pairs[j][1] + 1) j++
+    segments.push({ start: startCp, end: pairs[j][0], delta: (startGid - startCp) & 0xffff })
+    i = j + 1
+  }
+  segments.push({ start: 0xffff, end: 0xffff, delta: 1 }) // required terminator → gid 0
+  const segCount = segments.length
+  const segX2 = segCount * 2
+  let entropy = 0
+  while ((1 << (entropy + 1)) <= segCount) entropy++
+  const searchRange = 2 * (1 << entropy)
+  const words = []
+  words.push(4, segX2 + 16 /* length filled below */, 0, segX2, searchRange, entropy, segX2 - searchRange)
+  for (const s of segments) words.push(s.end)
+  words.push(0) // reservedPad
+  for (const s of segments) words.push(s.start)
+  for (const s of segments) words.push(s.delta)
+  for (const s of segments) words.push(0) // idRangeOffset — all pure-delta
+  const body = new Uint8Array(words.length * 2)
+  for (let k = 0; k < words.length; k++) {
+    body[k * 2] = (words[k] >> 8) & 0xff
+    body[k * 2 + 1] = words[k] & 0xff
+  }
+  body[2] = (body.length >> 8) & 0xff // real length now that it is known
+  body[3] = body.length & 0xff
+  return body
+}
+
+/** A format-12 (full Unicode) cmap subtable body for `pairs`, sorted. */
+function cmapFormat12(pairs) {
+  const groups = []
+  let i = 0
+  while (i < pairs.length) {
+    const [startCp, startGid] = pairs[i]
+    let j = i
+    while (j + 1 < pairs.length &&
+      pairs[j + 1][0] === pairs[j][0] + 1 && pairs[j + 1][1] === pairs[j][1] + 1) j++
+    groups.push([startCp, pairs[j][0], startGid])
+    i = j + 1
+  }
+  const body = new Uint8Array(16 + groups.length * 12)
+  const w32 = (at, v) => { body[at] = (v >>> 24) & 0xff; body[at + 1] = (v >>> 16) & 0xff; body[at + 2] = (v >>> 8) & 0xff; body[at + 3] = v & 0xff }
+  body[0] = 0; body[1] = 12 // format 12.0
+  w32(4, body.length)
+  w32(8, 0) // language
+  w32(12, groups.length)
+  groups.forEach(([s, e, g], k) => {
+    w32(16 + k * 12, s); w32(16 + k * 12 + 4, e); w32(16 + k * 12 + 8, g)
+  })
+  return body
+}
+
+/** The whole cmap table: a (3,1) format 4, plus a (3,10) format 12 when needed. */
+function buildCmap(codeToGid) {
+  const bmp = [...codeToGid].filter(([cp]) => cp <= 0xffff).sort((a, b) => a[0] - b[0])
+  const all = [...codeToGid].sort((a, b) => a[0] - b[0])
+  const astral = all.length > bmp.length
+  const subs = [{ platform: 3, encoding: 1, body: cmapFormat4(bmp) }]
+  if (astral) subs.push({ platform: 3, encoding: 10, body: cmapFormat12(all) })
+
+  const header = 4 + subs.length * 8
+  let offset = header
+  const out = []
+  const head = new Uint8Array(header)
+  head[0] = 0; head[1] = 0 // version
+  head[2] = (subs.length >> 8) & 0xff; head[3] = subs.length & 0xff
+  subs.forEach((s, i) => {
+    const at = 4 + i * 8
+    head[at] = (s.platform >> 8) & 0xff; head[at + 1] = s.platform & 0xff
+    head[at + 2] = (s.encoding >> 8) & 0xff; head[at + 3] = s.encoding & 0xff
+    head[at + 4] = (offset >>> 24) & 0xff; head[at + 5] = (offset >>> 16) & 0xff
+    head[at + 6] = (offset >>> 8) & 0xff; head[at + 7] = offset & 0xff
+    offset += s.body.length
+  })
+  out.push(head, ...subs.map(s => s.body))
+  return concat(out)
+}
+
+function concat(chunks) {
+  let n = 0
+  for (const c of chunks) n += c.length
+  const out = new Uint8Array(n)
+  let at = 0
+  for (const c of chunks) { out.set(c, at); at += c.length }
+  return out
+}
+
+/** A minimal post v3 table — no glyph names, which the PDF never reads. */
+function postV3(src) {
+  const out = new Uint8Array(32)
+  out.set(src.subarray(4, 32), 4) // keep italicAngle…maxMemType1 from the original
+  out[0] = 0; out[1] = 3; out[2] = 0; out[3] = 0 // version 3.0
+  return out
+}
+
+/**
+ * Subset a TrueType sfnt to the glyphs named by `directMap`.
+ *
+ * @param {Uint8Array} sfnt  the source font, already unpacked to plain sfnt
+ * @param {Map<number, number>} directMap  original glyph id → new (dense) id,
+ *   for exactly the glyphs the content stream references. Must map original 0
+ *   to new 0 (`.notdef`) and be dense over 0…directMap.size-1.
+ * @param {Map<number, number>} codeToOrig  code point → original glyph id, for
+ *   the retained characters — used to rebuild the cmap in the new numbering.
+ * @returns {{sfnt: Uint8Array}|{error: string}}
+ */
+export function subsetTrueType(sfnt, directMap, codeToOrig) {
+  const b = sfnt
+  const tables = directory(b)
+  const headT = tables.get('head')
+  const maxpT = tables.get('maxp')
+  const hheaT = tables.get('hhea')
+  const hmtxT = tables.get('hmtx')
+  const glyfT = tables.get('glyf')
+  const locaT = tables.get('loca')
+  if (!headT || !maxpT || !hheaT || !hmtxT || !glyfT || !locaT) {
+    return { error: 'subset needs head/maxp/hhea/hmtx/glyf/loca (TrueType outlines)' }
+  }
+  const numGlyphs = U16(b, maxpT.offset + 4)
+  const longLoca = U16(b, headT.offset + 50) === 1
+  const loca = readLoca(b, locaT, glyfT.offset, numGlyphs, longLoca)
+
+  // The retained set, closed over composite components. `newOf` extends
+  // directMap: components not directly used get ids after the dense prefix.
+  const newOf = new Map(directMap)
+  const queue = [...directMap.keys()]
+  while (queue.length) {
+    const gid = queue.pop()
+    if (gid >= numGlyphs) continue
+    const comps = new Set()
+    componentIds(b, loca, gid, comps)
+    for (const c of comps) {
+      if (!newOf.has(c)) { newOf.set(c, newOf.size); queue.push(c) }
+    }
+  }
+  const count = newOf.size
+  const origOf = new Array(count)
+  for (const [orig, nw] of newOf) origOf[nw] = orig
+
+  // glyf + loca in the new order. Each glyph padded to an even length; long loca
+  // throughout, so a large offset never overflows the short format.
+  const numH = U16(b, hheaT.offset + 34)
+  const glyphBytes = []
+  const locaOut = new Uint8Array((count + 1) * 4)
+  const putLoca = (i, v) => { const at = i * 4; locaOut[at] = (v >>> 24) & 0xff; locaOut[at + 1] = (v >>> 16) & 0xff; locaOut[at + 2] = (v >>> 8) & 0xff; locaOut[at + 3] = v & 0xff }
+  let glyfLen = 0
+  const hmtxOut = new Uint8Array(count * 4)
+  for (let nw = 0; nw < count; nw++) {
+    putLoca(nw, glyfLen)
+    const orig = origOf[nw]
+    let g = glyphSlice(b, loca, orig)
+    if (g.length) {
+      g = remapComposite(g, newOf)
+      glyphBytes.push(g)
+      glyfLen += g.length
+      if (g.length & 1) { glyphBytes.push(new Uint8Array(1)); glyfLen += 1 }
+    }
+    // hmtx: advance from the last real metric when orig sat in the monospaced
+    // tail, its own lsb otherwise; a subset keeps a full (advance,lsb) per glyph.
+    const hm = orig < numH ? orig : numH - 1
+    hmtxOut[nw * 4] = b[hmtxT.offset + hm * 4]
+    hmtxOut[nw * 4 + 1] = b[hmtxT.offset + hm * 4 + 1]
+    if (orig < numH) {
+      hmtxOut[nw * 4 + 2] = b[hmtxT.offset + orig * 4 + 2]
+      hmtxOut[nw * 4 + 3] = b[hmtxT.offset + orig * 4 + 3]
+    } else {
+      // A tail glyph's lsb lives in the leftSideBearings[] array that follows the
+      // hMetrics — not in a 4-byte hMetric. Reading it as one (above) would take
+      // the advance's bytes; dropping it (no else) would zero the bearing and
+      // shift the painted glyph sideways within its cell on a monospaced face.
+      const at = hmtxT.offset + numH * 4 + (orig - numH) * 2
+      hmtxOut[nw * 4 + 2] = b[at]
+      hmtxOut[nw * 4 + 3] = b[at + 1]
+    }
+  }
+  putLoca(count, glyfLen)
+  const glyfOut = concat(glyphBytes)
+
+  // Fixed-size tables, copied then patched.
+  const head = b.slice(headT.offset, headT.offset + headT.length)
+  head[50] = 0; head[51] = 1 // indexToLocFormat = long
+  head[8] = head[9] = head[10] = head[11] = 0 // checkSumAdjustment zeroed; set at the end
+  const maxp = b.slice(maxpT.offset, maxpT.offset + maxpT.length)
+  maxp[4] = (count >> 8) & 0xff; maxp[5] = count & 0xff
+  const hhea = b.slice(hheaT.offset, hheaT.offset + hheaT.length)
+  hhea[34] = (count >> 8) & 0xff; hhea[35] = count & 0xff // numberOfHMetrics
+
+  const codeToNew = new Map()
+  for (const [cp, orig] of codeToOrig) {
+    if (newOf.has(orig)) codeToNew.set(cp, newOf.get(orig))
+  }
+
+  const out = new Map()
+  out.set('head', head)
+  out.set('hhea', hhea)
+  out.set('maxp', maxp)
+  out.set('hmtx', hmtxOut)
+  out.set('loca', locaOut)
+  out.set('glyf', glyfOut)
+  out.set('cmap', buildCmap(codeToNew))
+  for (const tag of ['name', 'OS/2']) {
+    const t = tables.get(tag)
+    if (t) out.set(tag, b.slice(t.offset, t.offset + t.length))
+  }
+  const postT = tables.get('post')
+  if (postT) out.set('post', postV3(b.slice(postT.offset, postT.offset + postT.length)))
+
+  return { sfnt: assemble(out) }
+}
+
+/** Lay the tables out as a checksummed sfnt with a sorted directory. */
+function assemble(tableMap) {
+  const tags = [...tableMap.keys()].sort()
+  const numTables = tags.length
+  let entropy = 0
+  while ((1 << (entropy + 1)) <= numTables) entropy++
+  const searchRange = (1 << entropy) * 16
+
+  const headerLen = 12 + numTables * 16
+  let offset = headerLen
+  const records = []
+  for (const tag of tags) {
+    const data = tableMap.get(tag)
+    records.push({ tag, data, offset, length: data.length, checksum: checksum(data) })
+    offset += (data.length + 3) & ~3
+  }
+
+  const out = new Uint8Array(offset)
+  const w16 = (at, v) => { out[at] = (v >> 8) & 0xff; out[at + 1] = v & 0xff }
+  const w32 = (at, v) => { out[at] = (v >>> 24) & 0xff; out[at + 1] = (v >>> 16) & 0xff; out[at + 2] = (v >>> 8) & 0xff; out[at + 3] = v & 0xff }
+  w32(0, 0x00010000)
+  w16(4, numTables)
+  w16(6, searchRange)
+  w16(8, entropy)
+  w16(10, numTables * 16 - searchRange)
+  records.forEach((r, i) => {
+    const at = 12 + i * 16
+    for (let k = 0; k < 4; k++) out[at + k] = r.tag.charCodeAt(k) || 0x20
+    w32(at + 4, r.checksum)
+    w32(at + 8, r.offset)
+    w32(at + 12, r.length)
+    out.set(r.data, r.offset)
+  })
+
+  // head.checkSumAdjustment closes the loop: 0xB1B0AFBA minus the whole file's
+  // checksum with this field zero (it already is).
+  const headRec = records.find(r => r.tag === 'head')
+  const adjustment = (0xb1b0afba - checksum(out)) >>> 0
+  w32(headRec.offset + 8, adjustment)
+  return out
+}
+
+export default subsetTrueType

package/vector/svd/index.js

@@ -0,0 +1,72 @@
+/**
+ * SVD — Scalable Vector Document.
+ *
+ * The contract between the capture engine and every editor backend. The engine
+ * emits SVD; each backend is a pure `SVD -> output` function with no DOM, no
+ * `getComputedStyle`, no measuring. If a backend needs to measure something,
+ * the schema is missing a field — extend the schema, never measure downstream.
+ *
+ * Invariants (normative — a document violating any of these is a bug):
+ *
+ *  1. Units are CSS pixels, unrounded floats. Rounding belongs to the emitter.
+ *     Angles are degrees, clockwise.
+ *  2. `frame` is relative to the emitted parent. `abs` is relative to the
+ *     border box of the capture root, Y down. Both are always present: `abs`
+ *     survives reparenting and is what the test harness asserts against.
+ *  3. `children` is in DOM containment order. `paint.z` carries the back-to-front
+ *     order. The backend sorts by `paint.z`; it never re-derives it.
+ *  4. Colors are `[r, g, b, a]` floats 0..1 in sRGB, plus `srcCss` for round-trip.
+ *  5. `blur` is the CSS blur radius. SVG sigma is `blur / 2`. Stated once, here.
+ *  6. Gradient geometry is normalized to the unit square with a 2x3 matrix.
+ *     Never percentages — that is the documented origin of Figma's
+ *     "failed to invert transform" on SVG import.
+ *  7. Nothing relative, nothing external: assets are embedded or referenced by
+ *     local id.
+ *  8. A document that degraded something without saying so in `diagnostics` is
+ *     a bug. That rule outranks the rest of the schema.
+ */
+
+export const SVD_VERSION = 'svd/1.0'
+
+/** @typedef {'frame'|'group'|'shape'|'text'|'image'|'vector'} NodeType */
+export const NODE_TYPES = ['frame', 'group', 'shape', 'text', 'image', 'vector']
+
+/**
+ * Fidelity grades. Every node carries one; anything but `E` must have a
+ * matching entry in `diagnostics`.
+ *
+ *  E  exact        — the target primitive reproduces the CSS with no loss
+ *  A  approximate  — parametric, visually equivalent within tolerance
+ *  C  composed     — one CSS declaration became N primitives
+ *  R  raster       — rasterized, with the reason recorded
+ *  RH raster hybrid — background rasterized, vector content kept on top
+ *  O  omitted      — could not be represented; placeholder emitted
+ */
+export const FIDELITY = ['E', 'A', 'C', 'R', 'RH', 'O']
+
+export { validate, validateNode } from './schema.js'
+
+/** Empty document with every required top-level key present. */
+export function emptyDocument (capture = {}) {
+  return {
+    schema: SVD_VERSION,
+    generator: { name: '@zumer/snapdom-vector', version: '0.0.1', engine: 'dom' },
+    capture: {
+      url: '', selector: '', capturedAt: '',
+      viewport: { w: 0, h: 0, dpr: 1 },
+      docOrigin: { x: 0, y: 0 },
+      root: { w: 0, h: 0 },
+      conventions: {
+        units: 'px', y: 'down', angle: 'deg-cw',
+        colorSpace: 'srgb', blur: 'css-radius',
+      },
+      ...capture,
+    },
+    assets: {},
+    styles: { text: {} },
+    root: null,
+    nodes: {},
+    diagnostics: [],
+    report: {},
+  }
+}