Clone-fidelity fix waves 1-3 (+ wave 4 in flight): capture settling, media, iframes

Fixes from the cropin.com / ooni.com audit reviews:
- walker: preserve whitespace-only text in inline elements ("ofthe" bug);
  capture ::placeholder styles for inputs
- css: emit visibility (hidden-at-rest wordmark artifact); exempt list
  markers from inherited elision (lost bullets); ::placeholder rules
- seo: sanitize og:type to Next's enum (render crash -> dev badge leak);
  generated next.config disables devIndicators
- capture/stabilize: promote lazy-loader data attrs before snapshots
  (collapsed sections, viewport-inconsistent captures); settle autoplay
  carousels to home slide before every snapshot; force-reveal hidden
  videos for poster shots + log failures
- generate: ship downloaded video files as local <video src>; keep
  picture>source through IR prune with srcset rewrite (mobile-crop-on-
  desktop heroes); pin Tailwind named screens to computeBands boundaries
- capture/graft: capture cross-origin iframe subtrees (Klaviyo signup
  forms) with element-screenshot fallback; asset download retry +
  visual-assets-missing reporting

Checkpoint commit: wave 4 (reveal settling, 206 range-response fix,
hidden-geometry banding, bare-zero utility gating) is mid-implementation.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Samraaj Bath
2026-07-03 17:45:42 -07:00
co-authored by Claude Fable 5
parent da537e68b8
commit 9aed2540aa
31 changed files with 2579 additions and 121 deletions
+83
View File
@@ -0,0 +1,83 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Autoplay carousel fixture</title>
<style>
* { box-sizing: border-box; }
body { font-family: -apple-system, Segoe UI, Roboto, sans-serif; margin: 0; color: #1a1a1a; }
main { max-width: 720px; margin: 0 auto; padding: 40px 24px; }
/* Splide-like autoplaying carousel: clipped track, flex list moved by translateX,
pagination bullets, pause-on-hover (the library default this fixture mimics). */
.splide { position: relative; }
.splide__track { overflow: hidden; border-radius: 10px; }
.splide__list { display: flex; transition: transform .2s ease; margin: 0; padding: 0; list-style: none; }
.splide__slide { flex: 0 0 100%; min-width: 100%; height: 200px; display: flex; align-items: center; justify-content: center; font-size: 34px; color: #fff; }
.a0 { background: #4f46e5; } .a1 { background: #0891b2; } .a2 { background: #16a34a; } .a3 { background: #db2777; }
.splide__pagination { display: flex; gap: 8px; justify-content: center; margin-top: 12px; padding: 0; list-style: none; }
.splide__pagination__bullet { width: 10px; height: 10px; border-radius: 50%; border: 0; padding: 0; cursor: pointer; background: #d1d5db; }
.splide__pagination__bullet.is-active { background: #4f46e5; }
/* A control-less advanced track (what an interaction pass can leave behind):
no bullets, no arrows, not loop-mode — settle must pin it back to home. */
.swiper { overflow: hidden; border-radius: 10px; margin-top: 32px; }
.swiper-wrapper { display: flex; }
.swiper-slide { flex: 0 0 100%; min-width: 100%; height: 120px; display: flex; align-items: center; justify-content: center; font-size: 24px; color: #fff; background: #ea580c; }
</style>
</head>
<body>
<main>
<h1>Autoplay carousel</h1>
<div id="auto" class="splide" aria-roledescription="carousel" aria-label="Autoplay">
<div class="splide__track">
<ul class="splide__list">
<li class="splide__slide a0 is-active">Slide 1</li>
<li class="splide__slide a1">Slide 2</li>
<li class="splide__slide a2">Slide 3</li>
<li class="splide__slide a3">Slide 4</li>
</ul>
</div>
<ul class="splide__pagination">
<li><button class="splide__pagination__bullet is-active" aria-label="Go to slide 1"></button></li>
<li><button class="splide__pagination__bullet" aria-label="Go to slide 2"></button></li>
<li><button class="splide__pagination__bullet" aria-label="Go to slide 3"></button></li>
<li><button class="splide__pagination__bullet" aria-label="Go to slide 4"></button></li>
</ul>
</div>
<div id="stuck" class="swiper">
<div class="swiper-wrapper" style="transform: translateX(-100%);">
<div class="swiper-slide">A</div>
<div class="swiper-slide">B</div>
<div class="swiper-slide">C</div>
</div>
</div>
</main>
<script>
(() => {
const root = document.getElementById("auto");
const list = root.querySelector(".splide__list");
const slides = [...root.querySelectorAll(".splide__slide")];
const bullets = [...root.querySelectorAll(".splide__pagination__bullet")];
let i = 0;
let paused = false;
const go = (n) => {
i = ((n % slides.length) + slides.length) % slides.length;
list.style.transform = "translateX(-" + (i * 100) + "%)";
slides.forEach((s, k) => s.classList.toggle("is-active", k === i));
bullets.forEach((b, k) => b.classList.toggle("is-active", k === i));
};
bullets.forEach((b, k) => b.addEventListener("click", () => go(k)));
// pause-on-hover, the Splide/Slick default this fixture mimics
root.addEventListener("mouseenter", () => { paused = true; });
root.addEventListener("mouseleave", () => { paused = false; });
window.__autoplayTicks = 0;
setInterval(() => { if (!paused) { window.__autoplayTicks++; go(i + 1); } }, 250);
})();
</script>
</body>
</html>
+25
View File
@@ -0,0 +1,25 @@
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<title>Newsletter embed</title>
<style>
body { margin: 0; background: #1a1a2e; color: #ffffff; font-family: Arial, sans-serif; }
form { padding: 12px; }
label { display: block; margin-bottom: 6px; font-size: 13px; }
input { display: block; width: 200px; color: #ffffff; background: #333344; border: 1px solid #555577; padding: 6px; }
input::placeholder { color: rgb(200, 150, 100); }
button { margin-top: 8px; background: #e94560; color: #ffffff; border: 0; padding: 8px 16px; }
a { color: #8888ff; font-size: 12px; }
</style>
</head>
<body>
<form id="signup" action="/subscribe">
<label for="email">Email</label>
<input id="email" type="email" placeholder="Enter your email">
<button type="submit">Sign up</button>
<a href="#terms">Terms</a>
<img src="/pixel.png" alt="" width="10" height="10">
</form>
</body>
</html>
+23
View File
@@ -0,0 +1,23 @@
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<title>Iframe host</title>
<style>
body { margin: 0; font-family: Arial, sans-serif; background: #ffffff; }
h1 { margin: 20px; font-size: 24px; }
iframe.signup {
display: block;
margin: 30px 0 0 40px;
width: 320px;
height: 160px;
border: 0;
}
</style>
</head>
<body>
<h1>Host page</h1>
<!-- {{FRAME_ORIGIN}} is replaced by the test server with the second (cross-origin) server's origin. -->
<iframe class="signup" title="Newsletter" aria-label="Modal Overlay Box Frame" src="{{FRAME_ORIGIN}}/iframe-embed.html"></iframe>
</body>
</html>
+37
View File
@@ -0,0 +1,37 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Lazy-media fixture</title>
<style>
* { box-sizing: border-box; }
body { font-family: -apple-system, Segoe UI, Roboto, sans-serif; margin: 0; padding: 24px; }
img { display: block; }
.bg { width: 300px; height: 150px; background-size: cover; background-repeat: no-repeat; }
</style>
</head>
<body>
<h1>Lazy media</h1>
<!-- WP Rocket style: 0-size placeholder in src, real URL in data-lazy-src. -->
<img id="lazy1" src="data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7" data-lazy-src="og-image.png" alt="hero">
<!-- lazysizes style: no src at all, data-src + data-sizes="auto" (a swap flag, not a value). -->
<img id="lazy2" data-src="seo-icon.png" data-sizes="auto" alt="icon">
<picture>
<source id="lazysource" type="image/png" data-srcset="og-image.png 1x">
<img id="lazy3" data-src="og-image.png" alt="picture">
</picture>
<!-- Lazy background image (WP Rocket url(...) form). -->
<div id="lazybg" class="bg" data-bg="url('brand.svg')"></div>
<!-- Must NOT be promoted: the data attr is not a plausible URL. -->
<img id="notaurl" data-src="{{ placeholder }}" alt="template token">
<!-- Must NOT count as a promotion: src already equals the target. -->
<img id="already" src="seo-icon.png" data-src="seo-icon.png" alt="already swapped">
</body>
</html>
+18
View File
@@ -0,0 +1,18 @@
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<title>Placeholder styling</title>
<style>
body { margin: 0; font-family: Arial, sans-serif; }
input.styled { color: rgb(10, 20, 30); font-size: 16px; }
input.styled::placeholder { color: rgb(120, 30, 200); font-size: 14px; }
textarea.styled::placeholder { color: rgb(5, 100, 5); }
</style>
</head>
<body>
<input class="styled" type="email" placeholder="Enter your email">
<textarea class="styled" placeholder="Your message"></textarea>
<input class="plain" type="text" value="no placeholder here">
</body>
</html>
+223 -8
View File
@@ -3,6 +3,14 @@ import { join } from "node:path";
import { collectPage, type PageSnapshot, type FontFace } from "./walker.js"; import { collectPage, type PageSnapshot, type FontFace } from "./walker.js";
import { tagElements, captureInteractions, type InteractionCapture } from "./interactions.js"; import { tagElements, captureInteractions, type InteractionCapture } from "./interactions.js";
import { captureMotion, probeReveals, type MotionCapture } from "./motion.js"; import { captureMotion, probeReveals, type MotionCapture } from "./motion.js";
import {
promoteLazyMedia, settleCarousels, settleScrollReveals, neutralizePreReveal,
forceRevealForShot, restoreRevealForShot,
} from "./stabilize.js";
import {
enumerateFramesInPage, planForFrameUrl, graftFrameIntoSnapshot, frameHasRenderableContent,
MAX_GRAFT_FRAMES, FRAME_GRAFT_MAX_NODES, type FrameCandidate,
} from "./graft.js";
import { discoverBreakpoints } from "./breakpoints.js"; import { discoverBreakpoints } from "./breakpoints.js";
import { writeJSON, writeJSONCompact, writeBytes, ensureDir } from "../util/fsx.js"; import { writeJSON, writeJSONCompact, writeBytes, ensureDir } from "../util/fsx.js";
import { sha1_12, round } from "../util/canonical.js"; import { sha1_12, round } from "../util/canonical.js";
@@ -90,6 +98,23 @@ function viewportHeight(width: number): number {
return VIEWPORT_HEIGHTS[width] ?? Math.round(width * 0.66); return VIEWPORT_HEIGHTS[width] ?? Math.round(width * 0.66);
} }
// ---- Asset download retry (Fix: transient failures silently degrade to placeholders) ----
/** Types whose absence is visible in the clone (image → transparent GIF, video → blank,
* font → fallback face). These earn one retry; css/manifest/lottie degrade gracefully. */
const RETRYABLE_ASSET_TYPES = new Set(["image", "svg", "video", "font"]);
/** Fixed, bounded delay before the single retry — deterministic (no jitter/backoff). */
export const ASSET_RETRY_DELAY_MS = 750;
/** Should a failed download of `type` with HTTP `status` (null = network error / no
* response) be retried once? Transient states only: connection failures, 5xx, and 429.
* 4xx (404/403/…) are authoritative — retrying can't change them. */
export function isRetryableAssetFailure(type: string, status: number | null): boolean {
if (!RETRYABLE_ASSET_TYPES.has(type)) return false;
if (status === null) return true;
return status >= 500 || status === 429;
}
function extFromUrl(url: string): string { function extFromUrl(url: string): string {
try { try {
const p = new URL(url).pathname; const p = new URL(url).pathname;
@@ -137,6 +162,19 @@ function isCss(url: string, contentType: string | null): boolean {
return classifyAsset(url, contentType) === "css"; return classifyAsset(url, contentType) === "css";
} }
/** Container-magic check for accepted video bytes: ISO-BMFF (`ftyp` within the first
* 12 bytes — mp4/m4v/mov), webm/mkv (EBML 0x1A45DFA3), or ogg (`OggS`). A range
* fragment (e.g. an mp4's moov-atom tail served as a 206) fails all three, so a
* corrupt partial body is never stored as the asset. */
export function looksLikeVideoFile(bytes: Buffer): boolean {
if (bytes.length < 12) return false;
const head = bytes.subarray(0, 12);
if (head.includes("ftyp")) return true;
if (head[0] === 0x1a && head[1] === 0x45 && head[2] === 0xdf && head[3] === 0xa3) return true; // EBML (webm/mkv)
if (head[0] === 0x4f && head[1] === 0x67 && head[2] === 0x67 && head[3] === 0x53) return true; // "OggS"
return false;
}
async function autoScroll(page: import("playwright").Page, vpHeight: number): Promise<void> { async function autoScroll(page: import("playwright").Page, vpHeight: number): Promise<void> {
// Scroll through the page to trigger lazy-loaded images/backgrounds, then return // Scroll through the page to trigger lazy-loaded images/backgrounds, then return
// to the top so document-coordinate bboxes are measured from a settled layout. // to the top so document-coordinate bboxes are measured from a settled layout.
@@ -535,6 +573,10 @@ export async function captureSite(opts: {
const storeBytes = (url: string, type: string, bytes: Buffer): void => { const storeBytes = (url: string, type: string, bytes: Buffer): void => {
if (!bytes || bytes.length === 0) return; if (!bytes || bytes.length === 0) return;
// Reject non-container bytes for video from EVERY path (a range fragment or error
// page stored under first-stored-wins ships a corrupt file). Left unstored, the
// asset surfaces in visual_assets_missing instead.
if (type === "video" && !looksLikeVideoFile(bytes)) return;
const a = assetMap.get(url) ?? recordAsset(url, type, null, null, "network"); const a = assetMap.get(url) ?? recordAsset(url, type, null, null, "network");
if (a.storedAs) return; if (a.storedAs) return;
const ext = extFromUrl(url) || extFromContentType(a.contentType) || const ext = extFromUrl(url) || extFromContentType(a.contentType) ||
@@ -655,6 +697,10 @@ export async function captureSite(opts: {
const existing = assetMap.get(url); const existing = assetMap.get(url);
if (existing?.storedAs) return; if (existing?.storedAs) return;
if (status >= 400) return; if (status >= 400) return;
// A 206 body is a RANGE FRAGMENT (media seek), not the asset — storing it would
// ship a corrupt file under first-stored-wins and the full-download fallback
// would then skip the asset. Record only; the fallback pass fetches the 200 body.
if (status === 206) return;
bodyPromises.push( bodyPromises.push(
(async () => { (async () => {
try { try {
@@ -684,6 +730,42 @@ export async function captureSite(opts: {
if (!navigated) throw navErr; if (!navigated) throw navErr;
await settle(page); await settle(page);
// Stage 2: lazy-loader promotion. WP Rocket/lazysizes keep a 0-size placeholder in
// `src` with the real URL in data attrs; autoScroll outruns their IntersectionObserver
// (collapsed sections in the snapshot) and the interaction pass can trigger the swap
// midway (viewports then DISAGREE about the section's size). Promote once, before any
// snapshot, so every viewport measures the same loaded media.
const lazyPromoted = await promoteLazyMedia(page);
if (lazyPromoted) {
log({ event: "lazy_promoted", count: lazyPromoted });
await settle(page, 2000); // newly-real images reflow the layout
}
// Stage 2: reveal settling. Scroll reveals (Elementor waypoints, WOW/AOS class swaps)
// are the same class of one-shot load-state as lazy media: fire them ONCE, before any
// viewport snapshot, so every width records the POST-REVEAL steady state — otherwise
// the snapshot bakes `visibility:hidden` wrappers (or a mid-fade opacity) with no JS
// to ever reveal them, and the clone renders below-fold content blank.
// A scroll-locking consent wall would defeat the dwell walk, so clear it first (the
// per-viewport dismissal below still runs and owns the audit trail).
const preClicked = await clickDismiss(page);
if (preClicked.length) {
for (const d of preClicked) if (!dismissUnion.dismissed.includes(d)) dismissUnion.dismissed.push(d);
await settle(page, 1000);
}
// Motion capture needs the PRE-reveal hidden state, observable only before the first
// scroll — tag + probe now (captureMotion confirms the candidates at the canonical
// snapshot, reading each revealed element's entrance animation from computed style).
if (opts.motion) { await tagElements(page); await probeReveals(page); }
const revealSettle = await settleScrollReveals(page);
log({ event: "reveals_settled", ...revealSettle });
// Belt-and-braces: reveal any element STILL carrying a known-library pre-reveal marker
// (below the step bound, or keyed to a non-scroll trigger) via the library's own
// revealed state, so captured computed styles are genuine post-reveal values.
const neutralized = await neutralizePreReveal(page);
if (neutralized) log({ event: "prereveal_neutralized", count: neutralized });
await settle(page, 1500); // revealed content reflows the layout
for (const vw of viewports) { for (const vw of viewports) {
const vh = viewportHeight(vw); const vh = viewportHeight(vw);
await page.setViewportSize({ width: vw, height: vh }); await page.setViewportSize({ width: vw, height: vh });
@@ -710,12 +792,8 @@ export async function captureSite(opts: {
}; };
await applyDismiss("load"); await applyDismiss("load");
// Stage 5 (scroll reveals): at the FIRST viewport, before any scroll, tag elements // (Scroll-reveal probing happens once, before the viewport loop — see the reveal
// and snapshot the pre-reveal hidden state — scroll reveals fire on the first // settling pass above; by this point all one-shot reveals have already fired.)
// autoScroll and stay revealed, so this is the only moment their hidden state is
// observable. Idempotent + motion-gated; the settled snapshot is unchanged.
if (opts.motion && vw === viewports[0]) { await tagElements(page); await probeReveals(page); }
await autoScroll(page, vh); await autoScroll(page, vh);
await settle(page, 1500); await settle(page, 1500);
await applyDismiss("post-scroll"); await applyDismiss("post-scroll");
@@ -737,6 +815,15 @@ export async function captureSite(opts: {
}); });
await page.waitForTimeout(150); await page.waitForTimeout(150);
// Stage 2: settle autoplaying carousels (pause + navigate to the home slide)
// before EVERY snapshot — otherwise each viewport freezes a different track
// offset (per-band CSS then bakes four different slides), and the interaction
// pass between the canonical and last snapshots would leave the final viewport
// contaminated. Scoped to named-library tracks so motion.ts's marquee/rotator
// capture (which runs after the canonical snapshot) observes them unchanged.
const carousels = await settleCarousels(page);
if (carousels.roots) log({ event: "carousels_settled", viewport: vw, ...carousels });
// Stage 2: dynamic-media first frame. Materialize a still for poster-less // Stage 2: dynamic-media first frame. Materialize a still for poster-less
// videos so they don't render blank — canvas where the frame is readable, else // videos so they don't render blank — canvas where the frame is readable, else
// an element screenshot (the page cannot screenshot itself). Done once at the // an element screenshot (the page cannot screenshot itself). Done once at the
@@ -755,12 +842,27 @@ export async function captureSite(opts: {
} catch { /* ignore */ } } catch { /* ignore */ }
} }
for (const s of plan.shots) { for (const s of plan.shots) {
// A visibility:hidden video (entrance animation not yet fired) passes the size
// gate, but locator.screenshot auto-waits for visibility and times out —
// force-reveal the hidden ancestor chain for the shot, restore exactly after.
let shot = false;
try { try {
await page.evaluate(forceRevealForShot, s.sel);
const buf = await page.locator(s.sel).first().screenshot({ type: "jpeg", quality: 82, timeout: 5000, animations: "disabled" }); const buf = await page.locator(s.sel).first().screenshot({ type: "jpeg", quality: 82, timeout: 5000, animations: "disabled" });
recordAsset(s.url, "image", "image/jpeg", 200, "video-still"); recordAsset(s.url, "image", "image/jpeg", 200, "video-still");
storeBytes(s.url, "image", buf); storeBytes(s.url, "image", buf);
dismissUnion.videoStills++; dismissUnion.videoStills++;
} catch { /* element not shootable (off-screen/detached) — poster falls back to placeholder */ } shot = true;
} catch (e) {
log({ event: "video_still_error", viewport: vw, sel: s.sel, error: String(e).slice(0, 200) });
} finally {
await page.evaluate(restoreRevealForShot).catch(() => { /* ignore */ });
}
// A synthetic poster with no bytes behind it generates a transparent tile —
// on failure, remove the attr so the video renders as it did pre-capture.
if (!shot) {
await page.evaluate((sel) => { document.querySelector(sel)?.removeAttribute("poster"); }, s.sel).catch(() => { /* ignore */ });
}
} }
// Element screenshots scroll the target into view; restore scroll 0 so the // Element screenshots scroll the target into view; restore scroll 0 so the
// canonical DOM walk + screenshot match the generated app's default render. // canonical DOM walk + screenshot match the generated app's default render.
@@ -774,6 +876,95 @@ export async function captureSite(opts: {
// them (whitelisted), enabling interaction deltas + motion specs to map to cids. // them (whitelisted), enabling interaction deltas + motion specs to map to cids.
if ((opts.interactions || opts.motion) && vw === canonical) await tagElements(page); if ((opts.interactions || opts.motion) && vw === canonical) await tagElements(page);
// Stage 2.6: cross-origin iframe content. The in-page walker cannot see into a
// cross-origin frame (newsletter/form embeds), but Node CAN evaluate in it — run the
// SAME collectPage per meaningful frame here, then graft each subtree into this
// viewport's snapshot below (graft.ts). Frames that can't be grafted (media players,
// dead frames) get an element-screenshot recorded as the iframe's background at the
// canonical viewport, so the box at least PAINTS. Runs before the main collectPage so
// the fallback's inline background is part of the canonical computed style.
const frameCands: FrameCandidate[] = await Promise.race([
page.evaluate(enumerateFramesInPage),
new Promise<FrameCandidate[]>((res) => setTimeout(() => res([]), 8000)),
]).catch(() => [] as FrameCandidate[]);
const frameGrafts: Array<{ cand: FrameCandidate; snap: PageSnapshot }> = [];
let graftBudget = MAX_GRAFT_FRAMES;
let stillBudget = MAX_GRAFT_FRAMES; // fallback stills share the same per-page bound
let frameShotScrolled = false;
for (const cand of frameCands) {
if (!cand.visible) continue;
const plan = planForFrameUrl(cand.url);
if (plan === "skip") continue;
let frameSnap: PageSnapshot | null = null;
if (plan === "graft" && graftBudget > 0) {
try {
const handle = await page.$(`iframe[data-ditto-frame="${cand.idx}"]`);
const frame = handle ? await handle.contentFrame() : null;
if (frame) {
await frame.evaluate(ESBUILD_SHIM);
frameSnap = await Promise.race([
frame.evaluate(collectPage, { maxNodes: FRAME_GRAFT_MAX_NODES }),
new Promise<never>((_, rej) => setTimeout(() => rej(new Error("frame collect timeout")), 20_000)),
]);
}
await handle?.dispose();
} catch (e) {
log({ event: "frame_graft_error", viewport: vw, frame: cand.idx, url: cand.url.slice(0, 160), error: String(e).slice(0, 200) });
frameSnap = null;
}
}
if (frameSnap && frameHasRenderableContent(frameSnap.root)) {
graftBudget--;
frameGrafts.push({ cand, snap: frameSnap });
// Merge the frame's discoveries exactly like the main document's below: its
// assets/fonts flow through the same pipeline so grafted <img>/@font-face resolve.
for (const da of frameSnap.domAssets) {
const t = classifyAsset(da.url, null) ?? (da.kind === "video" ? "video" : da.kind === "svg" ? "svg" : "image");
recordAsset(da.url, t, null, null, `frame${cand.idx}:${da.via}`);
}
for (const u of frameSnap.cssUrls) {
const t = classifyAsset(u, null) ?? "other";
recordAsset(u, t, null, null, "css-url");
}
for (const ff of frameSnap.fontFaces) {
const key = `${ff.family}|${ff.weight ?? ""}|${ff.style ?? ""}|${ff.src}`;
if (!fontFaceMap.has(key)) fontFaceMap.set(key, ff);
}
} else if (vw === canonical && stillBudget > 0) {
stillBudget--;
// Screenshot fallback — synthetic-poster pattern (see captureVideoStills): the
// still's bytes ride the normal asset pipeline under a synthetic URL; the iframe's
// inline background then rewrites to the local file at generation.
const stillUrl = `https://clone-frame.local/${cand.idx}-${sha1_12(cand.url || String(cand.idx))}.jpg`;
const sel = `iframe[data-ditto-frame="${cand.idx}"]`;
try {
await page.evaluate(forceRevealForShot, sel);
const buf = await page.locator(sel).first().screenshot({ type: "jpeg", quality: 82, timeout: 5000, animations: "disabled" });
recordAsset(stillUrl, "image", "image/jpeg", 200, "iframe-still");
storeBytes(stillUrl, "image", buf);
frameShotScrolled = true;
await page.evaluate(({ s, u }) => {
const el = document.querySelector(s) as HTMLElement | null;
if (el) {
el.style.backgroundImage = `url("${u}")`;
el.style.backgroundSize = "100% 100%";
el.style.backgroundRepeat = "no-repeat";
}
}, { s: sel, u: stillUrl });
log({ event: "frame_still", viewport: vw, frame: cand.idx, url: cand.url.slice(0, 160) });
} catch (e) {
log({ event: "frame_still_error", viewport: vw, frame: cand.idx, error: String(e).slice(0, 200) });
} finally {
await page.evaluate(restoreRevealForShot).catch(() => { /* ignore */ });
}
}
}
// Element screenshots scroll the target into view; restore scroll 0 before the walk.
if (frameShotScrolled) {
await page.evaluate(() => window.scrollTo(0, 0));
await page.waitForTimeout(80);
}
// Bound the in-page DOM walk: page.evaluate has no default timeout, so a // Bound the in-page DOM walk: page.evaluate has no default timeout, so a
// pathologically large/animated DOM (e.g. asana.com) could hang forever. // pathologically large/animated DOM (e.g. asana.com) could hang forever.
const snapshot: PageSnapshot = await Promise.race([ const snapshot: PageSnapshot = await Promise.race([
@@ -781,6 +972,13 @@ export async function captureSite(opts: {
new Promise<never>((_, rej) => setTimeout(() => rej(new Error(`collectPage timeout vp${vw}`)), 60_000)), new Promise<never>((_, rej) => setTimeout(() => rej(new Error(`collectPage timeout vp${vw}`)), 60_000)),
]); ]);
// Graft the captured frame subtrees into this viewport's snapshot (offset bboxes,
// namespaced ids, frame-URL-absolutized src/href — see graft.ts).
for (const g of frameGrafts) {
const ok = graftFrameIntoSnapshot(snapshot, g.cand, g.snap);
log({ event: ok ? "frame_grafted" : "frame_graft_orphaned", viewport: vw, frame: g.cand.idx, url: g.cand.url.slice(0, 160), nodes: g.snap.doc.nodeCount });
}
// Merge discovered references from the snapshot (DOM + accessible CSS). // Merge discovered references from the snapshot (DOM + accessible CSS).
for (const da of snapshot.domAssets) { for (const da of snapshot.domAssets) {
const t = classifyAsset(da.url, null) ?? (da.kind === "manifest" ? "manifest" : da.kind === "video" ? "video" : da.kind === "svg" ? "svg" : "image"); const t = classifyAsset(da.url, null) ?? (da.kind === "manifest" ? "manifest" : da.kind === "video" ? "video" : da.kind === "svg" ? "svg" : "image");
@@ -895,6 +1093,10 @@ export async function captureSite(opts: {
if (a.storedAs) continue; if (a.storedAs) continue;
if (a.url.startsWith("data:")) continue; if (a.url.startsWith("data:")) continue;
if (!["image", "svg", "video", "font", "lottie", "css", "manifest"].includes(a.type)) continue; if (!["image", "svg", "video", "font", "lottie", "css", "manifest"].includes(a.type)) continue;
// One bounded retry for transiently-failed VISUAL assets (network error / 5xx / 429):
// a single flaky fetch otherwise degrades an image to the transparent-GIF placeholder.
for (let attempt = 0; attempt < 2; attempt++) {
let failStatus: number | null = null;
try { try {
const resp = await fallbackCtx.request.get(a.url, { timeout: 30000 }); const resp = await fallbackCtx.request.get(a.url, { timeout: 30000 });
a.status = resp.status(); a.status = resp.status();
@@ -907,9 +1109,22 @@ export async function captureSite(opts: {
recordAsset(u, t, null, null, "css-text"); recordAsset(u, t, null, null, "css-text");
}); });
if (a.type === "manifest") parseManifestForAssets(buf.toString("utf8"), a.url); if (a.type === "manifest") parseManifestForAssets(buf.toString("utf8"), a.url);
break;
} }
} catch { /* unreachable/signed — left as skipped */ } failStatus = resp.status();
} catch { failStatus = null; /* unreachable/signed — left as skipped */ }
if (attempt > 0 || !isRetryableAssetFailure(a.type, failStatus)) break;
log({ event: "asset_retry", url: a.url, type: a.type, status: failStatus });
await new Promise((r) => setTimeout(r, ASSET_RETRY_DELAY_MS));
} }
}
// Every visual asset that ultimately failed, in one machine-readable event: these are
// the boxes that will paint as placeholders (image → transparent GIF, video → blank).
const visualMissing = [...assetMap.values()]
.filter((a) => !a.storedAs && !a.url.startsWith("data:") && (a.type === "image" || a.type === "svg" || a.type === "video"))
.map((a) => ({ url: a.url, type: a.type, status: a.status }))
.sort((x, y) => x.url.localeCompare(y.url));
if (visualMissing.length) log({ event: "visual_assets_missing", count: visualMissing.length, assets: visualMissing });
const seoResources: SeoResource[] = []; const seoResources: SeoResource[] = [];
const fetchedSeo = new Set<string>(); const fetchedSeo = new Set<string>();
const fetchSeoResource = async (url: string, kind: SeoResource["kind"]): Promise<void> => { const fetchSeoResource = async (url: string, kind: SeoResource["kind"]): Promise<void> => {
+186
View File
@@ -0,0 +1,186 @@
import type { PageSnapshot, RawNode, RawChild } from "./walker.js";
/**
* Cross-origin iframe subtree graft (Stage 2). The in-page walker cannot see into a
* cross-origin iframe (Klaviyo/newsletter embeds), so the capture used to record an
* empty box and the clone painted a blank frame. Playwright CAN evaluate inside
* cross-origin frames from Node, so capture.ts runs the SAME collectPage in each
* meaningful frame and this module merges the returned subtree into the page snapshot
* as ordinary children of the iframe node:
* - bboxes shift by the iframe's content-box origin (frame-doc → page-doc coords);
* - id/for/#href are namespaced `f<idx>-…` so two frames (or frame + page) can't
* collide on DOM ids in the clone;
* - src/srcset/href absolutize against the FRAME document's URL (generation resolves
* attrs against the main page URL, which would break frame-relative paths);
* - the grafted <body> becomes a <div>, and the iframe node clips (overflow:hidden)
* exactly like a real frame viewport.
* Downstream the iframe-with-children renders as a positioned <div> (app.ts), so the
* embed's form paints as real, styleable DOM. Frames that can't be grafted fall back
* to an element screenshot recorded as the iframe's background (capture.ts).
*/
/** Determinism/perf caps: at most this many grafted frames per page snapshot… */
export const MAX_GRAFT_FRAMES = 10;
/** …and a per-frame walker node budget (a fraction of the main document's 12000). */
export const FRAME_GRAFT_MAX_NODES = 2000;
/** Minimum rendered size (both axes) for a frame to carry meaningful content. */
export const MIN_FRAME_DIM = 48;
export type FramePlan = "skip" | "still" | "graft";
// Ad/analytics/consent plumbing frames render nothing a visitor values — skip entirely.
const FRAME_SKIP_RE =
/(?:doubleclick\.net|googlesyndication\.com|googletagmanager\.com|google-analytics\.com|googleadservices\.com|adservice\.google|facebook\.com\/tr|connect\.facebook\.net|\brecaptcha\b|hcaptcha\.com|challenges\.cloudflare\.com|adsrvr\.org|amazon-adsystem\.com)/i;
// Media players are JS-built canvases/videos whose DOM graft is meaningless; an element
// screenshot (the poster frame + play chrome) is the faithful static paint.
const FRAME_STILL_RE =
/(?:youtube(?:-nocookie)?\.com\/embed|player\.vimeo\.com|\bwistia\b|fast\.wistia|players?\.brightcove|open\.spotify\.com|w\.soundcloud\.com|google\.com\/maps\/embed)/i;
/** How to materialize a frame's content, from its URL alone (deterministic). */
export function planForFrameUrl(url: string): FramePlan {
const u = (url || "").trim();
if (!u || u === "about:blank" || u.startsWith("javascript:")) return "skip";
if (FRAME_SKIP_RE.test(u)) return "skip";
if (FRAME_STILL_RE.test(u)) return "still";
return "graft";
}
export type FrameCandidate = {
idx: number; // stable per-page index, stamped as data-ditto-frame on the element
url: string; // the frame element's src (absolute)
visible: boolean; // rendered, ≥ MIN_FRAME_DIM on both axes
// Content-box origin in page document coordinates: the offset every frame-doc bbox
// shifts by (the child browsing context fills the iframe's content box).
contentX: number;
contentY: number;
};
/**
* Stamp every iframe with a stable `data-ditto-frame` index (DOM order; idempotent so
* later viewports reuse the same identity) and report each frame's geometry/visibility.
* Serialized into the page via page.evaluate — must stay self-contained.
*/
export function enumerateFramesInPage(): FrameCandidate[] {
const frames = Array.from(document.querySelectorAll("iframe"));
let next = 0;
for (const f of frames) {
const cur = f.getAttribute("data-ditto-frame");
if (cur !== null) next = Math.max(next, parseInt(cur, 10) + 1);
}
const sx = window.scrollX, sy = window.scrollY;
const out: FrameCandidate[] = [];
for (const f of frames) {
let idxAttr = f.getAttribute("data-ditto-frame");
if (idxAttr === null) { idxAttr = String(next++); f.setAttribute("data-ditto-frame", idxAttr); }
const cs = getComputedStyle(f);
const r = f.getBoundingClientRect();
const visible = cs.display !== "none" && cs.visibility !== "hidden" &&
parseFloat(cs.opacity || "1") > 0 && r.width >= 48 && r.height >= 48;
out.push({
idx: parseInt(idxAttr, 10),
url: (f as HTMLIFrameElement).src || "",
visible,
contentX: Math.round((r.x + sx + parseFloat(cs.borderLeftWidth || "0") + parseFloat(cs.paddingLeft || "0")) * 100) / 100,
contentY: Math.round((r.y + sy + parseFloat(cs.borderTopWidth || "0") + parseFloat(cs.paddingTop || "0")) * 100) / 100,
});
}
return out.sort((a, b) => a.idx - b.idx);
}
function isTextRaw(c: RawChild): c is { text: string } {
return (c as { text?: string }).text !== undefined;
}
/** Does the frame's captured tree paint anything (a visible element or real text)? */
export function frameHasRenderableContent(root: RawNode | undefined): boolean {
if (!root) return false;
const visit = (n: RawNode): boolean => {
for (const c of n.children) {
if (isTextRaw(c)) { if (c.text.trim().length > 0) return true; continue; }
if (c.visible || c.rawHTML) return true;
if (visit(c)) return true;
}
return false;
};
return visit(root);
}
/** Find the iframe RawNode stamped with the given data-ditto-frame index. */
export function findFrameNode(root: RawNode, idx: number): RawNode | null {
if (root.tag === "iframe" && root.attrs?.["data-ditto-frame"] === String(idx)) return root;
for (const c of root.children) {
if (isTextRaw(c)) continue;
const hit = findFrameNode(c, idx);
if (hit) return hit;
}
return null;
}
const ABS_URL_ATTRS = ["src", "poster", "data-lazy-src", "data-src", "data-original", "data-ll-src"];
const ABS_SRCSET_ATTRS = ["srcset", "data-lazy-srcset", "data-srcset"];
/**
* Merge one frame snapshot into the page snapshot as children of its iframe node.
* Returns true when the graft landed. Mutates `snapshot` (offsets, namespacing and the
* iframe's clipping are applied to the frame subtree copy embedded in it).
*/
export function graftFrameIntoSnapshot(
snapshot: PageSnapshot,
frame: { idx: number; contentX: number; contentY: number },
frameSnap: PageSnapshot,
): boolean {
const host = findFrameNode(snapshot.root, frame.idx);
if (!host || !frameSnap?.root) return false;
if (!frameHasRenderableContent(frameSnap.root)) return false;
const prefix = `f${frame.idx}-`;
const frameUrl = frameSnap.doc?.url || "";
const abs = (u: string): string => {
try { return new URL(u, frameUrl).href; } catch { return u; }
};
const absSrcset = (v: string): string =>
v.split(",").map((part) => {
const bits = part.trim().split(/\s+/);
if (bits[0] && !bits[0].startsWith("data:")) bits[0] = abs(bits[0]);
return bits.join(" ");
}).join(", ");
const nsIdList = (v: string): string =>
v.split(/\s+/).filter(Boolean).map((id) => prefix + id).join(" ");
const visit = (n: RawNode): void => {
n.bbox = {
...n.bbox,
x: Math.round((n.bbox.x + frame.contentX) * 100) / 100,
y: Math.round((n.bbox.y + frame.contentY) * 100) / 100,
};
const a = n.attrs ?? {};
delete a["data-cid-cap"]; // capture-ids belong to the main document only
if (a.id) a.id = prefix + a.id;
if (a.for) a.for = prefix + a.for;
for (const k of ["aria-labelledby", "aria-describedby", "aria-controls", "aria-owns", "aria-activedescendant"]) {
if (a[k]) a[k] = nsIdList(a[k]!);
}
if (a.href) a.href = a.href.startsWith("#") ? "#" + prefix + a.href.slice(1) : abs(a.href);
for (const k of ABS_URL_ATTRS) if (a[k] && !a[k]!.startsWith("data:")) a[k] = abs(a[k]!);
for (const k of ABS_SRCSET_ATTRS) if (a[k]) a[k] = absSrcset(a[k]!);
for (const c of n.children) if (!isTextRaw(c)) visit(c);
};
const wrapper = frameSnap.root; // the frame's <body>
visit(wrapper);
wrapper.tag = "div"; // a nested <body> is not valid; the box/styles replay identically
host.children = [wrapper];
// A real frame viewport clips its document; the replacement <div> must too, at every
// captured viewport (each per-viewport snapshot is grafted independently).
host.computed = { ...host.computed, overflow: "hidden", overflowX: "hidden", overflowY: "hidden" };
// An iframe is a REPLACED element: `display:inline` (the default) still honors its
// width/height. The <div> that replaces it at generation is not — inline would collapse
// the box — so translate to the behavior-equivalent inline-block.
if ((host.computed.display || "inline") === "inline") host.computed.display = "inline-block";
// Diagnostics only (nodeCount is logged, never gated).
snapshot.doc.nodeCount += frameSnap.doc?.nodeCount ?? 0;
// @keyframes referenced by grafted nodes must exist in the page snapshot's set.
if (frameSnap.keyframes?.length) snapshot.keyframes.push(...frameSnap.keyframes);
return true;
}
+64 -16
View File
@@ -41,7 +41,16 @@ export type RevealSpec = {
cap: string; // data-cid-cap of a scroll-revealed element cap: string; // data-cid-cap of a scroll-revealed element
opacity: string; // its hidden (pre-reveal) opacity, e.g. "0" opacity: string; // its hidden (pre-reveal) opacity, e.g. "0"
transform: string; // its hidden transform (slide/scale offset), or "none" transform: string; // its hidden transform (slide/scale offset), or "none"
transition: string; // the transition to animate the reveal with transition: string; // the transition to animate the reveal with ("" for the visibility family)
// visibility+entrance-class family (Elementor/WOW/AOS): hidden via `visibility:hidden`
// pre-scroll, revealed by a class swap that applies a keyframe animation. The clone
// re-hides with visibility (JS-applied, so non-JS/SSR still shows content) and replays
// the named animation when scrolled into view.
visibility?: "hidden";
animationName?: string; // entrance @keyframes name in the revealed state (e.g. fadeInUp)
animationDuration?: string; // e.g. "1.25s"
animationDelay?: string; // e.g. "0s"
animationTiming?: string; // e.g. "ease" / "cubic-bezier(...)"
}; };
export type MarqueeSpec = { export type MarqueeSpec = {
@@ -62,26 +71,41 @@ export type MotionCapture = {
}; };
/** /**
* Stage 5 (scroll reveals) — pre-scroll probe. Scroll-triggered reveals start hidden * Stage 5 (scroll reveals) — pre-scroll probe. Scroll-triggered reveals start hidden and
* (opacity:0 / transform offset) and animate in when scrolled into view; by the time the * animate in when scrolled into view; by the time the settled snapshot is taken (after
* settled snapshot is taken (after `autoScroll` has walked the page) they are already * the reveal-settling pass has walked the page) they are already revealed, so their
* revealed, so their hidden state must be sampled BEFORE the first scroll. Records, on * hidden state must be sampled BEFORE the first scroll. Records, on `window.__cloneReveal`,
* `window.__cloneReveal`, the pre-scroll opacity/transform/transition of every tagged * two candidate families over tagged elements — the set `captureMotion` later confirms
* element that is hidden-but-boxed with a real transition — the candidate reveal set * (kept only if the element ends up visible):
* `captureMotion` later confirms (kept only if the element ends up visible). Idempotent; * - **transition** — opacity≈0 with a real opacity/transform transition (the reveal
* call once at the first viewport, after settle, before `autoScroll`. * animates via the transition already on the element);
* - **visibility** — `visibility:hidden` with a real box (Elementor `.elementor-invisible`,
* WOW/AOS wrappers), revealed by a class swap that APPLIES a keyframe animation. The
* entrance animation only exists post-swap, so `captureMotion` reads it at confirm
* time from the revealed computed style. Only the OUTERMOST hidden element is recorded
* (visibility inherits — descendants are covered by the wrapper's reveal).
* Idempotent; call once at the canonical width, after settle, before the first scroll.
*/ */
export async function probeReveals(page: Page): Promise<void> { export async function probeReveals(page: Page): Promise<void> {
try { try {
await page.evaluate(() => { await page.evaluate(() => {
const out: Record<string, { opacity: string; transform: string; transition: string }> = {}; const out: Record<string, { opacity: string; transform: string; transition: string; family?: "visibility" }> = {};
for (const el of Array.from(document.querySelectorAll("[data-cid-cap]"))) { for (const el of Array.from(document.querySelectorAll("[data-cid-cap]"))) {
const cap = el.getAttribute("data-cid-cap"); if (!cap) continue; const cap = el.getAttribute("data-cid-cap"); if (!cap) continue;
let cs: CSSStyleDeclaration; try { cs = getComputedStyle(el); } catch { continue; } let cs: CSSStyleDeclaration; try { cs = getComputedStyle(el); } catch { continue; }
const op = parseFloat(cs.opacity || "1");
if (op > 0.05) continue; // only currently-hidden elements are reveal candidates
const r = (el as HTMLElement).getBoundingClientRect(); const r = (el as HTMLElement).getBoundingClientRect();
if (r.width < 8 || r.height < 8) continue; // must occupy real space (not a 0-box hidden node) if (r.width < 8 || r.height < 8) continue; // must occupy real space (not a 0-box hidden node)
if (cs.visibility === "hidden") {
// visibility family: record only the reveal ROOT (parent not also hidden).
const p = (el as HTMLElement).parentElement;
let parentHidden = false;
try { parentHidden = !!p && getComputedStyle(p).visibility === "hidden"; } catch { /* ignore */ }
if (parentHidden) continue;
out[cap] = { opacity: cs.opacity || "1", transform: "none", transition: "", family: "visibility" };
continue;
}
const op = parseFloat(cs.opacity || "1");
if (op > 0.05) continue; // only currently-hidden elements are reveal candidates
// must have a transition on opacity/transform/all (so the reveal animates, not snaps) // must have a transition on opacity/transform/all (so the reveal animates, not snaps)
const tp = (cs.transitionProperty || "").toLowerCase(); const tp = (cs.transitionProperty || "").toLowerCase();
const td = cs.transitionDuration || "0s"; const td = cs.transitionDuration || "0s";
@@ -311,17 +335,41 @@ export async function captureMotion(page: Page, opts?: { observeMs?: number; log
// ---- Scroll reveals: confirm the pre-scroll candidates (probeReveals) that are // ---- Scroll reveals: confirm the pre-scroll candidates (probeReveals) that are
// NOW visible — those genuinely revealed on scroll (vs. elements that stayed hidden). // NOW visible — those genuinely revealed on scroll (vs. elements that stayed hidden).
const reveals: Array<{ cap: string; opacity: string; transform: string; transition: string }> = []; const reveals: Array<{
cap: string; opacity: string; transform: string; transition: string;
visibility?: "hidden"; animationName?: string; animationDuration?: string; animationDelay?: string; animationTiming?: string;
}> = [];
try { try {
const probed = (window as unknown as { __cloneReveal?: Record<string, { opacity: string; transform: string; transition: string }> }).__cloneReveal || {}; const probed = (window as unknown as { __cloneReveal?: Record<string, { opacity: string; transform: string; transition: string; family?: "visibility" }> }).__cloneReveal || {};
// First value of a comma-joined animation longhand; timing functions carry inner
// commas (cubic-bezier/steps), so take the whole leading function when present.
const first = (v: string): string => (/^\s*(cubic-bezier\([^)]*\)|steps\([^)]*\)|[^,]+)/.exec(v || "")?.[1] ?? "").trim();
for (const cap of Object.keys(probed)) { for (const cap of Object.keys(probed)) {
const el = document.querySelector(`[data-cid-cap="${cap}"]`); const el = document.querySelector(`[data-cid-cap="${cap}"]`);
if (!el) continue; if (!el) continue;
const cs = getComputedStyle(el); const cs = getComputedStyle(el);
if (parseFloat(cs.opacity || "1") <= 0.05) continue; // still hidden → not a reveal, just hidden const p = probed[cap]!;
const r = (el as HTMLElement).getBoundingClientRect(); const r = (el as HTMLElement).getBoundingClientRect();
if (r.width < 8 || r.height < 8) continue; if (r.width < 8 || r.height < 8) continue;
reveals.push({ cap, ...probed[cap]! }); if (p.family === "visibility") {
if (cs.visibility === "hidden") continue; // never revealed → genuinely hidden content
// Revealed via class swap. The swap's entrance animation is now in the computed
// style (libraries keep the animated class); record it for replay.
const name = first(cs.animationName || "none");
reveals.push({
cap, opacity: p.opacity, transform: "none", transition: "",
visibility: "hidden",
...(name && name !== "none" ? {
animationName: name,
animationDuration: first(cs.animationDuration) || "1s",
animationDelay: first(cs.animationDelay) || "0s",
animationTiming: first(cs.animationTimingFunction) || "ease",
} : {}),
});
continue;
}
if (parseFloat(cs.opacity || "1") <= 0.05) continue; // still hidden → not a reveal, just hidden
reveals.push({ cap, opacity: p.opacity, transform: p.transform, transition: p.transition });
} }
} catch { /* ignore */ } } catch { /* ignore */ }
+411
View File
@@ -0,0 +1,411 @@
import type { Page } from "playwright";
/**
* Pre-snapshot stabilization (Stage 2). Two dynamic behaviors otherwise make the
* per-viewport snapshots disagree with each other and with what a settled visitor sees:
*
* - **Lazy-loader placeholders** (WP Rocket / lazysizes): the real URL lives in a data
* attribute while `src` holds a 0-size placeholder; `autoScroll` outruns their
* IntersectionObserver, so snapshots record collapsed sections — and the interaction
* pass can trigger the swap midway, leaving viewports INCONSISTENT (cropin's "Global
* presence" map: 0×0 at 375/768/1280, 898px at 1920). Promoting the data attrs to
* real ones ONCE, before any snapshot, makes every viewport measure the stable
* post-reveal size (validated against the live site).
* - **Autoplaying carousels** (Splide/Swiper-style transform tracks): each viewport
* snapshot freezes a DIFFERENT translateX offset which the generator bakes into
* per-band CSS (ooni's splide02: -375/0/-1280/-1920 across the four widths). Settling
* — autoplay paused, track at the home slide — before EVERY snapshot makes all
* viewports (including the post-interaction 1920 pass) see one canonical state.
*
* Scope guard for motion capture (motion.ts contract): carousel settling touches ONLY
* elements matching the named-library selectors below, and pauses only the track's own
* WAAPI/CSS animations. rAF-driven marquees (Framer Motion tickers) match none of these
* selectors and keep running, so `detectMarquees` still observes their velocity; paused
* WAAPI animations remain in `document.getAnimations()` with keyframes/timing intact.
*/
// ---- Lazy-media promotion (runs in the page) ----
/**
* Promote lazy-loader data attributes to real ones and wait (bounded) for the newly-real
* images to decode, so bboxes are measured loaded. Only values that look like URLs are
* promoted (some themes stash JSON/flags in data-src-like attrs); an `src` already equal
* to the target is left alone, so the pass is idempotent. Returns the number of elements
* changed. Serialized into the page via page.evaluate — must stay self-contained.
*/
export async function promoteLazyMediaInPage(): Promise<number> {
const sleep = (ms: number) => new Promise((r) => setTimeout(r, ms));
const urlish = (v: string | null | undefined): v is string => {
if (!v) return false;
const s = v.trim();
if (!s || s.length > 4096 || /[<>"'\s]/.test(s)) return false;
return /^(?:https?:)?\/\//i.test(s) || s.startsWith("/") || s.startsWith("./") || s.startsWith("../") ||
/^data:image\//i.test(s) || /^[\w][^:]*\.[a-z0-9]{2,5}(?:[?#]|$)/i.test(s);
};
const srcsetish = (v: string | null | undefined): v is string => {
if (!v) return false;
const first = v.split(",")[0]?.trim().split(/\s+/)[0];
return urlish(first);
};
const setAttr = (el: Element, name: string, value: string): boolean => {
if (el.getAttribute(name) === value) return false;
el.setAttribute(name, value);
return true;
};
const promotedImgs: HTMLImageElement[] = [];
let count = 0;
const SEL =
"img[data-lazy-src],img[data-src],img[data-lazy-srcset],img[data-srcset],img[data-lazy-sizes],img[data-sizes]," +
"source[data-lazy-srcset],source[data-srcset],iframe[data-lazy-src],iframe[data-src],[data-bg]";
for (const el of Array.from(document.querySelectorAll(SEL))) {
let changed = false;
const tag = el.tagName;
if (tag === "IMG" || tag === "IFRAME") {
const src = el.getAttribute("data-lazy-src") ?? el.getAttribute("data-src");
if (urlish(src)) changed = setAttr(el, "src", src.trim()) || changed;
}
if (tag === "IMG" || tag === "SOURCE") {
const srcset = el.getAttribute("data-lazy-srcset") ?? el.getAttribute("data-srcset");
if (srcsetish(srcset)) changed = setAttr(el, "srcset", srcset.trim()) || changed;
// lazysizes' `data-sizes="auto"` is a computed-at-swap flag, not a real sizes value.
const sizes = (el.getAttribute("data-lazy-sizes") ?? el.getAttribute("data-sizes"))?.trim();
if (sizes && sizes !== "auto") changed = setAttr(el, "sizes", sizes) || changed;
}
const bg = el.getAttribute("data-bg");
if (bg) {
// data-bg carries either a raw URL (lazysizes) or a full url(...) (WP Rocket).
const inner = bg.trim().replace(/^url\(\s*(['"]?)(.*?)\1\s*\)$/i, "$2").trim();
if (urlish(inner)) {
const want = `url("${inner}")`;
const st = (el as HTMLElement).style;
if (st.backgroundImage !== want) { st.backgroundImage = want; changed = true; }
}
}
if (changed) {
count++;
if (tag === "IMG") { el.setAttribute("loading", "eager"); promotedImgs.push(el as HTMLImageElement); }
}
}
if (promotedImgs.length) {
await Promise.race([
Promise.all(promotedImgs.map((img) => (typeof img.decode === "function" ? img.decode() : Promise.resolve()).catch(() => { /* broken URL — bbox stays as-is */ }))),
sleep(4000),
]);
}
return count;
}
/** Node-side wrapper: bounded + never fatal (a hung page just skips promotion). */
export async function promoteLazyMedia(page: Page): Promise<number> {
try {
return await Promise.race([
page.evaluate(promoteLazyMediaInPage),
new Promise<number>((res) => setTimeout(() => res(0), 8000)),
]);
} catch {
return 0;
}
}
// ---- Scroll-reveal settling (runs in the page) ----
/** Fixed dwell per scroll step. Reveal libraries fire from IntersectionObserver callbacks
* or throttled scroll handlers (WOW ~100ms, AOS 99ms debounce); the plain autoScroll's
* 60ms cadence outruns them, leaving reveal wrappers baked `visibility:hidden` in the
* snapshot. 400ms per step reliably clears every observed library. Deterministic constant. */
export const REVEAL_DWELL_MS = 400;
/** Bound the dwell walk for pathological/endless pages (~80 × 0.75 viewport ≈ 60 screens). */
export const REVEAL_MAX_STEPS = 80;
/** Bounded wait for FINITE entrance animations started by the walk to finish, so no
* viewport snapshot freezes a mid-fade frame (the ~5%-opacity ghost state). */
export const REVEAL_ANIMATION_WAIT_MS = 3000;
export type RevealSettleResult = { steps: number; animationsAwaited: number };
/**
* Deterministic dwell-scroll so every one-shot scroll reveal (Elementor waypoints,
* WOW/AOS class swaps, IntersectionObserver entrances) fires BEFORE any viewport
* snapshot — the same class of one-shot load-state as lazy media, settled the same way
* (once, before the viewport loop). Steps 0.75×viewport with a fixed dwell through the
* full scrollHeight, waits for the entrance animations it started to complete (infinite
* iterations excluded — they never finish by design), then restores scroll 0.
* Serialized into the page via page.evaluate — must stay self-contained.
*/
export async function settleScrollRevealsInPage(cfg: { dwellMs: number; maxSteps: number; animWaitMs: number }): Promise<RevealSettleResult> {
const sleep = (ms: number) => new Promise((r) => setTimeout(r, ms));
const step = Math.max(Math.round(window.innerHeight * 0.75), 200);
const maxScroll = () => document.documentElement.scrollHeight - window.innerHeight;
let y = 0;
let steps = 0;
while (y < maxScroll() && steps < cfg.maxSteps) {
y += step;
window.scrollTo(0, y);
await sleep(cfg.dwellMs);
steps++;
}
let animationsAwaited = 0;
try {
const anims = (document.getAnimations?.() ?? []).filter((a) => {
try {
if (a.playState !== "running") return false;
const t = (a.effect as KeyframeEffect | null)?.getTiming();
return t != null && t.iterations !== Infinity;
} catch { return false; }
});
animationsAwaited = anims.length;
await Promise.race([
Promise.allSettled(anims.map((a) => a.finished)),
sleep(cfg.animWaitMs),
]);
} catch { /* getAnimations unsupported — the fixed post-wait still applies */ }
window.scrollTo(0, 0);
await sleep(250); // fixed post-wait: let scroll-position-dependent styles re-settle at top
return { steps, animationsAwaited };
}
/** Node-side wrapper: bounded + never fatal (a hung page just skips settling). */
export async function settleScrollReveals(page: Page): Promise<RevealSettleResult> {
const empty: RevealSettleResult = { steps: 0, animationsAwaited: 0 };
const cfg = { dwellMs: REVEAL_DWELL_MS, maxSteps: REVEAL_MAX_STEPS, animWaitMs: REVEAL_ANIMATION_WAIT_MS };
const bound = cfg.maxSteps * cfg.dwellMs + cfg.animWaitMs + 8000;
try {
return await Promise.race([
page.evaluate(settleScrollRevealsInPage, cfg),
new Promise<RevealSettleResult>((res) => setTimeout(() => res(empty), bound)),
]);
} catch {
return empty;
}
}
/**
* Defensive follow-up to the dwell walk: elements STILL carrying a known pre-reveal
* marker (far below fold past the step bound, or keyed to a non-scroll trigger) are
* moved to the library's OWN revealed state — the classes the library would add/remove
* — never raw style overrides, so the captured computed styles are the library's
* genuine post-reveal values. Known-library allowlist only (same philosophy as the
* carousel selectors below); elements hidden for real reasons are untouched.
* Serialized into the page via page.evaluate — must stay self-contained.
*/
export function neutralizePreRevealInPage(): number {
const stillHidden = (el: Element): boolean => {
try {
const cs = getComputedStyle(el);
return cs.visibility === "hidden" || parseFloat(cs.opacity || "1") <= 0.05;
} catch { return false; }
};
let n = 0;
// Elementor waypoint reveals: `.elementor-invisible` is removed and `animated` + the
// entrance-animation class (data-settings._animation / .animation) added on reveal.
const elementorReveal = (el: Element): void => {
let anim = "";
try {
const s = JSON.parse(el.getAttribute("data-settings") || "{}") as Record<string, unknown>;
const v = s["_animation"] ?? s["animation"];
if (typeof v === "string") anim = v.trim();
} catch { /* malformed settings — reveal without the animation class */ }
el.classList.remove("elementor-invisible");
el.classList.add("animated");
if (anim && anim !== "none") el.classList.add(anim);
};
for (const el of Array.from(document.querySelectorAll(".elementor-invisible"))) { elementorReveal(el); n++; }
// Elementor variants where the invisible class was renamed but the entrance setting
// remains: still-hidden elements whose data-settings configure an animation.
for (const el of Array.from(document.querySelectorAll('[data-settings*="animation"]'))) {
if (!stillHidden(el)) continue;
elementorReveal(el);
n++;
}
// WOW.js: init() sets inline visibility:hidden; reveal sets it visible + adds `animated`
// (the keyframe class, e.g. fadeInUp, is already in the element's class list).
for (const el of Array.from(document.querySelectorAll(".wow:not(.animated)")) as HTMLElement[]) {
el.classList.add("animated");
el.style.visibility = "visible";
n++;
}
// AOS: [data-aos] elements are hidden by attribute selectors until `.aos-animate` lands.
for (const el of Array.from(document.querySelectorAll("[data-aos]:not(.aos-animate)"))) {
el.classList.add("aos-animate");
n++;
}
return n;
}
/** Node-side wrapper: bounded + never fatal. */
export async function neutralizePreReveal(page: Page): Promise<number> {
try {
return await Promise.race([
page.evaluate(neutralizePreRevealInPage),
new Promise<number>((res) => setTimeout(() => res(0), 8000)),
]);
} catch {
return 0;
}
}
// ---- Carousel settling (runs in the page) ----
export type CarouselSettleResult = { roots: number; normalized: number; neutralizedAnims: number };
/**
* Deterministically settle recognizable library carousels: engage the library's own
* pause path, neutralize any animation driving the track, and navigate to the REAL first
* slide (home). Navigation preference:
* 1. the exposed Swiper instance (`el.swiper`) — stop autoplay + slideTo(Loop)(0, 0);
* 2. the first pagination bullet (libraries resolve loop-mode clones themselves);
* 3. prev-arrow clicks back from the marked active slide's real index;
* 4. inline `translateX(0)` — only for a non-loop track with no controls (loop mode
* prepends clones, so 0 is not home there; such a track is left paused as-is).
* Autoplay pause is best-effort-deterministic: Swiper via its instance; Splide/Slick/Glide
* pause on pointer-enter by default, so a synthetic mouseenter/mouseover latches them
* (nothing dispatches the matching mouseleave). Ends with a bounded wait for every track
* transform to stop changing so the caller snapshots a settled frame.
*/
export async function settleCarouselsInPage(): Promise<CarouselSettleResult> {
const sleep = (ms: number) => new Promise((r) => setTimeout(r, ms));
const ROOT_SEL = ".splide, .swiper, .swiper-container, .slick-slider, .glide";
const TRACK_SEL = ".splide__list, .swiper-wrapper, .slick-track, .glide__slides";
const BULLET_SEL = ".splide__pagination__bullet, .swiper-pagination-bullet, .slick-dots button, .glide__bullet";
const PREV_SEL = ".splide__arrow--prev, .swiper-button-prev, .slick-prev, [data-glide-dir='<']";
const ACTIVE_SEL = ".is-active, .swiper-slide-active, .slick-current, .glide__slide--active";
const CLONE_SEL = ".splide__slide--clone, .swiper-slide-duplicate, .slick-cloned, .glide__slide--clone";
const txOf = (el: Element): number => {
try { return new DOMMatrixReadOnly(getComputedStyle(el).transform).m41; } catch { return 0; }
};
const roots = Array.from(document.querySelectorAll(ROOT_SEL));
const tracks: Element[] = [];
let normalized = 0;
let neutralizedAnims = 0;
for (const root of roots) {
const track = Array.from(root.querySelectorAll(TRACK_SEL)).find((t) => t.closest(ROOT_SEL) === root);
if (!track) continue;
tracks.push(track);
// pause-on-hover latch (Splide/Slick default-on; Glide's hoverpause): synthetic
// pointer-enter with no matching leave keeps autoplay paused for the snapshot.
for (const t of [root, track.parentElement, track]) {
if (!t) continue;
try {
t.dispatchEvent(new MouseEvent("mouseenter", { bubbles: false }));
t.dispatchEvent(new MouseEvent("mouseover", { bubbles: true }));
} catch { /* ignore */ }
}
// Neutralize the track's OWN animations only (scoped so motion.ts's marquee/rotator
// capture is untouched). CSS transitions/animations are CANCELED, not paused: pausing
// a CSSTransition disassociates it from style and it then HOLDS its frozen mid-flight
// transform, overriding the home navigation below (the declarative path reconstructs
// CSS motion from the IR, so canceling loses nothing). Pure WAAPI is PAUSED so it
// stays in getAnimations() with keyframes/timing intact for motion.ts to record.
try {
for (const a of (track as HTMLElement).getAnimations?.() ?? []) {
try {
const ctor = a.constructor.name;
if (ctor === "CSSTransition" || ctor === "CSSAnimation") a.cancel(); else a.pause();
neutralizedAnims++;
} catch { /* ignore */ }
}
} catch { /* ignore */ }
let home = false;
// 1) Swiper exposes its instance on the container.
const sw = (root as { swiper?: { autoplay?: { stop?: () => void }; slideTo?: (i: number, ms?: number) => void; slideToLoop?: (i: number, ms?: number) => void } }).swiper;
if (sw) {
try { sw.autoplay?.stop?.(); } catch { /* ignore */ }
try {
if (sw.slideToLoop) { sw.slideToLoop(0, 0); home = true; }
else if (sw.slideTo) { sw.slideTo(0, 0); home = true; }
} catch { /* ignore */ }
}
// 2) First pagination bullet (clicked even if hidden at this width — a destroyed
// breakpoint variant just ignores the click).
if (!home) {
const bullet = Array.from(root.querySelectorAll(BULLET_SEL)).find((b) => b.closest(ROOT_SEL) === root) as HTMLElement | undefined;
if (bullet) { try { bullet.click(); home = true; } catch { /* ignore */ } }
}
// 3) Step back from the marked active slide's real index with the prev arrow.
if (!home) {
const prev = Array.from(root.querySelectorAll(PREV_SEL)).find((b) => b.closest(ROOT_SEL) === root) as HTMLElement | undefined;
const active = Array.from(track.children).find((s) => s.matches(ACTIVE_SEL));
if (prev && active) {
const real = Array.from(track.children).filter((s) => !s.matches(CLONE_SEL));
const idx = real.indexOf(active);
if (idx >= 0) {
for (let k = 0; k < Math.min(idx, 30); k++) { try { prev.click(); } catch { break; } await sleep(90); }
home = true;
}
}
}
// 4) No controls: pin translateX(0) — home for a non-loop track. Loop mode prepends
// clones (0 shows a clone), so a control-less loop track is left paused as-is.
if (!home) {
const isLoop = /--loop\b/.test(root.className) || track.querySelector(CLONE_SEL) != null;
if (!isLoop && Math.abs(txOf(track)) > 0.5) {
(track as HTMLElement).style.transform = "translateX(0px)";
home = true;
}
}
if (home) normalized++;
}
// Bounded wait for the navigation transitions to land (and confirm nothing is still
// auto-advancing): every track transform stable for 3 consecutive samples.
if (tracks.length) {
let prevSig = tracks.map((t) => Math.round(txOf(t))).join(",");
let stable = 0;
for (let i = 0; i < 14 && stable < 3; i++) {
await sleep(140);
const sig = tracks.map((t) => Math.round(txOf(t))).join(",");
if (sig === prevSig) stable++; else { stable = 0; prevSig = sig; }
}
}
return { roots: tracks.length, normalized, neutralizedAnims };
}
/** Node-side wrapper: bounded + never fatal. */
export async function settleCarousels(page: Page): Promise<CarouselSettleResult> {
const empty: CarouselSettleResult = { roots: 0, normalized: 0, neutralizedAnims: 0 };
try {
return await Promise.race([
page.evaluate(settleCarouselsInPage),
new Promise<CarouselSettleResult>((res) => setTimeout(() => res(empty), 10_000)),
]);
} catch {
return empty;
}
}
// ---- Force-reveal for element screenshots (runs in the page) ----
/**
* A visibility:hidden video (entrance animation not yet fired) passes the size gate but
* `locator.screenshot` auto-waits for visibility and times out. Force the element and any
* hidden ancestor visible for the shot, recording each prior inline value so
* `restoreRevealForShot` puts everything back exactly. Returns how many were forced.
*/
export function forceRevealForShot(sel: string): number {
const el = document.querySelector(sel) as HTMLElement | null;
if (!el) return 0;
let n = 0;
for (let cur: HTMLElement | null = el; cur; cur = cur.parentElement) {
if (getComputedStyle(cur).visibility !== "hidden") continue;
const prev = cur.style.getPropertyValue("visibility");
const prio = cur.style.getPropertyPriority("visibility");
cur.setAttribute("data-clone-vis-restore", `${prio}|${prev}`);
cur.style.setProperty("visibility", "visible", "important");
n++;
}
return n;
}
/** Undo forceRevealForShot exactly (inline value + priority, or absence). */
export function restoreRevealForShot(): void {
for (const el of Array.from(document.querySelectorAll("[data-clone-vis-restore]")) as HTMLElement[]) {
const raw = el.getAttribute("data-clone-vis-restore") ?? "|";
el.removeAttribute("data-clone-vis-restore");
const i = raw.indexOf("|");
const prio = raw.slice(0, i);
const prev = raw.slice(i + 1);
if (prev) el.style.setProperty("visibility", prev, prio);
else el.style.removeProperty("visibility");
}
}
+33 -2
View File
@@ -47,6 +47,9 @@ export type RawNode = {
sizing?: RawSizing; sizing?: RawSizing;
before?: RawStyle; before?: RawStyle;
after?: RawStyle; after?: RawStyle;
// ::placeholder computed style for input/textarea with placeholder text. Without it the
// clone renders the browser's default gray, losing the authored placeholder color/type.
placeholder?: RawStyle;
rawHTML?: string; // set for inline <svg> rawHTML?: string; // set for inline <svg>
children: RawChild[]; children: RawChild[];
}; };
@@ -100,7 +103,9 @@ export type PageSnapshot = {
keyframes: string[]; // raw @keyframes blocks from accessible sheets keyframes: string[]; // raw @keyframes blocks from accessible sheets
}; };
export function collectPage(): PageSnapshot { // `| void` keeps the no-arg `page.evaluate(collectPage)` call sites type-compatible
// (Playwright types the missing argument as void); frame grafts pass { maxNodes }.
export function collectPage(opts?: { maxNodes?: number } | void): PageSnapshot {
// NOTE: This function is serialized and run in the browser via page.evaluate, // NOTE: This function is serialized and run in the browser via page.evaluate,
// so every constant/helper it uses must be declared INSIDE it (no module-scope // so every constant/helper it uses must be declared INSIDE it (no module-scope
// closure is available in the page). // closure is available in the page).
@@ -167,11 +172,25 @@ export function collectPage(): PageSnapshot {
"overflow", "objectFit", "overflow", "objectFit",
]; ];
// ::placeholder property set: the visual identity of placeholder text. Kept small —
// it inherits everything else from the input itself.
const PLACEHOLDER_PROPS: string[] = [
"color", "opacity", "fontFamily", "fontSize", "fontWeight", "fontStyle",
"letterSpacing", "textTransform",
];
const SKIP_TAGS = new Set([ const SKIP_TAGS = new Set([
"script", "style", "link", "meta", "noscript", "template", "base", "title", "head", "script", "style", "link", "meta", "noscript", "template", "base", "title", "head",
]); ]);
const MAX_NODES = 12000; // Text-level tags whose whitespace-only text still renders even as the FIRST/ONLY
// child (the lone-space case below); a block container's stray whitespace does not.
const INLINE_TEXT_TAGS = new Set([
"span", "strong", "em", "b", "i", "a", "u", "small", "sub", "sup", "code", "abbr", "time", "label",
]);
// Frame grafts pass a lower cap so one pathological embed can't dominate the snapshot.
const MAX_NODES = (opts && opts.maxNodes) || 12000;
const round2 = (n: number): number => Math.round((n || 0) * 100) / 100; const round2 = (n: number): number => Math.round((n || 0) * 100) / 100;
const scrollX = window.scrollX; const scrollX = window.scrollX;
@@ -358,6 +377,13 @@ export function collectPage(): PageSnapshot {
} }
} catch { /* ignore */ } } catch { /* ignore */ }
// ::placeholder: only meaningful on a control that shows placeholder text.
if ((tag === "input" || tag === "textarea") && (attrs.placeholder || "").trim()) {
try {
node.placeholder = grabStyle(window.getComputedStyle(el, "::placeholder"), PLACEHOLDER_PROPS);
} catch { /* ignore */ }
}
// Inline SVG → raw markup, no recursion. // Inline SVG → raw markup, no recursion.
if (tag === "svg") { if (tag === "svg") {
node.rawHTML = el.outerHTML; node.rawHTML = el.outerHTML;
@@ -379,6 +405,11 @@ export function collectPage(): PageSnapshot {
} else if (t.length > 0 && node.children.length > 0) { } else if (t.length > 0 && node.children.length > 0) {
// Preserve a single significant space between inline elements. // Preserve a single significant space between inline elements.
node.children.push({ text: " " }); node.children.push({ text: " " });
} else if (t.length > 0 && (/^inline/.test(cs.display) || INLINE_TEXT_TAGS.has(tag))) {
// Whitespace that is the FIRST/ONLY child of an inline element still renders
// (`of<strong> </strong>the` keeps its space); dropping it fuses the adjacent
// text runs. Scoped to inline parents so block containers stay empty.
node.children.push({ text: " " });
} }
continue; continue;
} }
+8 -4
View File
@@ -38,6 +38,9 @@ export type CloneResult = {
/** A stable, timestamp-free path to the app (via the `runs/<site>/latest` symlink), /** A stable, timestamp-free path to the app (via the `runs/<site>/latest` symlink),
* present in runs-layout mode when the symlink could be created. */ * present in runs-layout mode when the symlink could be created. */
stableAppDir?: string; stableAppDir?: string;
/** Visual assets (image/svg/video) that could not be downloaded — those boxes render
* as placeholders. Surfaced in the CLI summary; details in generated/assets.json. */
visualAssetsMissing?: number;
}; };
export function siteIdFromUrl(url: string): string { export function siteIdFromUrl(url: string): string {
@@ -455,7 +458,8 @@ export async function runClone(opts: CloneOptions): Promise<CloneResult> {
const gen = generateAll({ sourceDir, capture, viewports, sampleViewports: captureViewports, url: opts.url, outDir: generatedDir }); const gen = generateAll({ sourceDir, capture, viewports, sampleViewports: captureViewports, url: opts.url, outDir: generatedDir });
logBoth({ event: "ir_built", nodes: gen.ir.doc.nodeCount }); logBoth({ event: "ir_built", nodes: gen.ir.doc.nodeCount });
logBoth({ event: "inferred", sections: gen.sections.length, assets: gen.assetGraph.entries.length, fonts: gen.fontGraph.entries.length }); logBoth({ event: "inferred", sections: gen.sections.length, assets: gen.assetGraph.entries.length, fonts: gen.fontGraph.entries.length });
logBoth({ event: "generated", assetsCopied: gen.assetsCopied, assetsMissing: gen.assetsMissing.length }); const visualAssetsMissing = gen.assetGraph.entries.filter((e) => e.impact === "visual_missing").length;
logBoth({ event: "generated", assetsCopied: gen.assetsCopied, assetsMissing: gen.assetsMissing.length, visualAssetsMissing });
// When the source capture lacks native probe flags (this sandbox can't reach the // When the source capture lacks native probe flags (this sandbox can't reach the
// live site through the egress proxy), optionally iterate render→regen so the LOCAL clone-probe // live site through the egress proxy), optionally iterate render→regen so the LOCAL clone-probe
@@ -479,7 +483,7 @@ export async function runClone(opts: CloneOptions): Promise<CloneResult> {
stableAppDir = writeLatestPointer(runsDir, siteId, runDir); stableAppDir = writeLatestPointer(runsDir, siteId, runDir);
} }
return { runDir, sourceDir, appDir: out ? out.appDir : appDir, sourceUrl: opts.url, stableAppDir }; return { runDir, sourceDir, appDir: out ? out.appDir : appDir, sourceUrl: opts.url, stableAppDir, visualAssetsMissing };
} }
/** Record the newest run for a site in the runs layout: a `latest.json` breadcrumb (used by /** Record the newest run for a site in the runs layout: a `latest.json` breadcrumb (used by
@@ -576,11 +580,11 @@ async function main(): Promise<void> {
// --serve installs deps + starts the dev server after cloning; --open also launches the browser. // --serve installs deps + starts the dev server after cloning; --open also launches the browser.
const open = hasAnyFlag(args, ["--open"]); const open = hasAnyFlag(args, ["--open"]);
const serve = open || hasAnyFlag(args, ["--serve"]); const serve = open || hasAnyFlag(args, ["--serve"]);
const finish = async (res: { appDir: string; stableAppDir?: string }) => { const finish = async (res: { appDir: string; stableAppDir?: string; visualAssetsMissing?: number }) => {
if (serve) { if (serve) {
await serveApp(res.appDir, { open }); await serveApp(res.appDir, { open });
} else { } else {
process.stderr.write(doneSummary({ url, appDir: res.appDir, framework, stableAppDir: res.stableAppDir })); process.stderr.write(doneSummary({ url, appDir: res.appDir, framework, stableAppDir: res.stableAppDir, visualAssetsMissing: res.visualAssetsMissing }));
} }
}; };
const vpArg = firstFlagValue(args, ["--dev-viewports", "--viewports"]); const vpArg = firstFlagValue(args, ["--dev-viewports", "--viewports"]);
+9
View File
@@ -20,6 +20,9 @@ export type DoneSummaryInput = {
framework: "next" | "vite"; framework: "next" | "vite";
/** Stable path (a `runs/<site>/latest` symlink target) preferred as the `cd` target when present. */ /** Stable path (a `runs/<site>/latest` symlink target) preferred as the `cd` target when present. */
stableAppDir?: string; stableAppDir?: string;
/** Visual assets (image/svg/video) that failed to download those boxes paint as
* placeholders in the clone. 0/undefined line omitted. */
visualAssetsMissing?: number;
}; };
/** Double-quote a path for a POSIX shell so spaces and wrapping don't break copy-paste. */ /** Double-quote a path for a POSIX shell so spaces and wrapping don't break copy-paste. */
@@ -46,6 +49,12 @@ export function doneSummary(input: DoneSummaryInput): string {
" Or re-run with --serve to install deps and start the dev server for you", " Or re-run with --serve to install deps and start the dev server for you",
" (add --open to launch the browser too).", " (add --open to launch the browser too).",
]; ];
if (input.visualAssetsMissing) {
const n = input.visualAssetsMissing;
lines.push("");
lines.push(`${n} visual asset${n === 1 ? "" : "s"} could not be downloaded and will render as`);
lines.push(` placeholders — see generated/assets.json (classification "skipped").`);
}
if (input.stableAppDir && input.stableAppDir !== input.appDir) { if (input.stableAppDir && input.stableAppDir !== input.appDir) {
lines.push(""); lines.push("");
lines.push(` The path above is a stable pointer to the newest clone. This exact run:`); lines.push(` The path above is a stable pointer to the newest clone. This exact run:`);
+74 -19
View File
@@ -195,6 +195,35 @@ function resolveUrl(url: string, base: string): string {
try { return new URL(url, base).href; } catch { return url; } try { return new URL(url, base).href; } catch { return url; }
} }
/** Srcset candidates rewritten to materialized local assets, order preserved;
* candidates whose URL did not materialize are dropped (never placeholders
* srcset wins over src, so a placeholder would beat the rewritten fallback). */
function keptSrcsetCandidates(value: string, assetMap: Map<string, string>, sourceUrl: string): string[] {
return value.split(",").map((p) => p.trim()).filter(Boolean).map((seg) => {
const sp = seg.split(/\s+/);
const abs = resolveUrl(sp[0] ?? "", sourceUrl);
const local = assetMap.get(abs);
return local ? [local, ...sp.slice(1)].join(" ") : null;
}).filter((x): x is string => x !== null);
}
/** True when a <video>'s own src or any child <source src> materialized locally
* the clone can then ship the real file instead of the poster-only fallback. */
function videoHasLocalSource(node: IRNode, assetMap: Map<string, string>, sourceUrl: string): boolean {
if (node.attrs.src && assetMap.get(resolveUrl(node.attrs.src, sourceUrl))) return true;
return node.children.some((c) => !isTextChild(c) && c.tag === "source"
&& !!c.attrs.src && !!assetMap.get(resolveUrl(c.attrs.src, sourceUrl)));
}
/** Whether a <source> element still points at anything after asset rewriting
* (materialized src, or 1 surviving srcset candidate). A source with none is
* omitted rather than emitted pointing at placeholders. */
function sourceHasLocalCandidate(c: IRNode, assetMap: Map<string, string>, sourceUrl: string): boolean {
if (c.attrs.src && assetMap.get(resolveUrl(c.attrs.src, sourceUrl))) return true;
if (c.attrs.srcset && keptSrcsetCandidates(c.attrs.srcset, assetMap, sourceUrl).length > 0) return true;
return false;
}
/** Default single-page link rewrite: a clone is self-contained, so a link that points /** Default single-page link rewrite: a clone is self-contained, so a link that points
* back to the SOURCE origin is rewritten to an app-relative path (`/enterprise`) instead * back to the SOURCE origin is rewritten to an app-relative path (`/enterprise`) instead
* of the absolute source URL (`https://www.source.com/enterprise`) otherwise every nav * of the absolute source URL (`https://www.source.com/enterprise`) otherwise every nav
@@ -284,39 +313,48 @@ export function propsList(node: IRNode, assetMap: Map<string, string>, sourceUrl
const prim = ctx?.primitives?.get(node.id); const prim = ctx?.primitives?.get(node.id);
if (prim) props.push(['"data-component"', JSON.stringify(prim)]); if (prim) props.push(['"data-component"', JSON.stringify(prim)]);
// Stage 2: a <video> is rendered as its (first-frame) poster — a streamed source // A <video> whose file materialized locally ships it, mirroring the captured
// has no deterministic frame and its request aborts at snapshot time. Drop the // playback attrs (autoplay/loop/muted/playsInline) faithfully. Otherwise it is
// streaming src + autoplay so only the poster paints; keep the poster (rewritten // rendered as its (first-frame) poster — a streamed source has no deterministic
// to a local still below). <source>/<track> children are dropped in renderNode. // frame and its request aborts at snapshot time — dropping the streaming src +
// autoplay so only the poster paints; keep the poster (rewritten to a local
// still below). <source>/<track> children are filtered in emitChildren.
const isVideo = node.tag === "video"; const isVideo = node.tag === "video";
// An <iframe> embeds a third-party, non-deterministic document; reproducing it const videoLocal = isVideo && videoHasLocalSource(node, assetMap, sourceUrl);
// An <iframe> embeds a third-party, non-deterministic document; reproducing it live
// would break self-containment (rubric Gate 2) and can't be deterministic anyway. // would break self-containment (rubric Gate 2) and can't be deterministic anyway.
// Keep the element as a placeholder sized by its captured box, but drop the // When capture grafted the embedded document's subtree (graft.ts) the node renders as
// document-loading attrs so it paints an empty frame instead of pulling content. // a <div> with those children (see resolveTag) — drop the frame-only attrs entirely
// (width/height would be invalid on a div; CSS keys the box). Otherwise keep the
// element as a placeholder sized by its captured box, minus the document-loading
// attrs, so it paints an empty frame instead of pulling content.
const isIframe = node.tag === "iframe"; const isIframe = node.tag === "iframe";
const iframeGrafted = isIframe && node.children.some((c) => !isTextChild(c));
const attrKeys = Object.keys(node.attrs).sort(); const attrKeys = Object.keys(node.attrs).sort();
for (const key of attrKeys) { for (const key of attrKeys) {
let value = node.attrs[key]!; let value = node.attrs[key]!;
if (key === "class" || key === "style" || key === "data-cid-cap") continue; if (key === "class" || key === "style" || key === "data-cid-cap") continue;
if (isVideo && (key === "src" || key === "autoplay" || key === "loop" || key === "preload")) continue; if (isVideo && !videoLocal && (key === "src" || key === "autoplay" || key === "loop" || key === "preload")) continue;
if (isIframe && (key === "src" || key === "srcdoc" || key === "name")) continue; if (isIframe && (key === "src" || key === "srcdoc" || key === "name")) continue;
if (iframeGrafted && (key === "width" || key === "height")) continue;
if (ASSET_ATTRS.has(key)) { if (ASSET_ATTRS.has(key)) {
const abs = resolveUrl(value, sourceUrl); const abs = resolveUrl(value, sourceUrl);
const local = assetMap.get(abs); const local = assetMap.get(abs);
value = local ?? TRANSPARENT_GIF; // never point back to a remote origin // A poster that missed the asset map is DROPPED — a guaranteed-blank overlay
// hides the element's own background, strictly worse than showing it. A missed
// video/source src likewise (a placeholder is not playable; a local <source>
// child may still carry the file). Only <img> keeps the transparent-GIF
// fallback: never point back to a remote origin.
if (!local && (key === "poster" || node.tag === "video" || node.tag === "source")) continue;
value = local ?? TRANSPARENT_GIF;
} else if (key === "srcset") { } else if (key === "srcset") {
// Keep only candidates we actually materialized; drop the rest. Lazy-load // Keep only candidates we actually materialized; drop the rest. Lazy-load
// libraries seed srcset with 1x1 placeholders (data: GIFs) and swap in the // libraries seed srcset with 1x1 placeholders (data: GIFs) and swap in the
// real URLs via JS — replaying those placeholders would beat the rewritten // real URLs via JS — replaying those placeholders would beat the rewritten
// `src` (srcset wins over src) and paint a blank box. If nothing survives, // `src` (srcset wins over src) and paint a blank box. If nothing survives,
// omit srcset so the browser falls back to the real local `src`. // omit srcset so the browser falls back to the real local `src`.
const kept = value.split(",").map((p) => p.trim()).filter(Boolean).map((seg) => { const kept = keptSrcsetCandidates(value, assetMap, sourceUrl);
const sp = seg.split(/\s+/);
const abs = resolveUrl(sp[0] ?? "", sourceUrl);
const local = assetMap.get(abs);
return local ? [local, ...sp.slice(1)].join(" ") : null;
}).filter((x): x is string => x !== null);
if (kept.length === 0) continue; if (kept.length === 0) continue;
value = kept.join(", "); value = kept.join(", ");
} else if (key === "href") { } else if (key === "href") {
@@ -340,7 +378,7 @@ export function propsList(node: IRNode, assetMap: Map<string, string>, sourceUrl
} }
} }
if (isVideo && !props.some(([k]) => k === "preload")) props.push(["preload", JSON.stringify("none")]); if (isVideo && !videoLocal && !props.some(([k]) => k === "preload")) props.push(["preload", JSON.stringify("none")]);
if (node.rawHTML && node.tag === "svg") { if (node.rawHTML && node.tag === "svg") {
// Strip the Stage-4 capture-id (`data-cid-cap`) the interaction pass stamps on // Strip the Stage-4 capture-id (`data-cid-cap`) the interaction pass stamps on
@@ -504,6 +542,10 @@ export function resolveTag(node: IRNode, insideInteractive: boolean, insideTable
const disp = (node.computedByVp[1280] ?? Object.values(node.computedByVp)[0])?.display ?? ""; const disp = (node.computedByVp[1280] ?? Object.values(node.computedByVp)[0])?.display ?? "";
tag = /inline(?!-block|-flex|-grid)/.test(disp) ? "span" : "div"; tag = /inline(?!-block|-flex|-grid)/.test(disp) ? "span" : "div";
} }
// An iframe with grafted children (capture/graft.ts) is a real subtree, not an embed:
// children inside <iframe> are unrendered fallback content, so emit a <div> container
// carrying the iframe's box/styles (CSS is keyed by cid, so the geometry is identical).
if (tag === "iframe" && node.children.some((c) => !isTextChild(c))) tag = "div";
if (TABLE_SCOPED.has(tag) && !insideTable) tag = "div"; // orphan table element → neutral box if (TABLE_SCOPED.has(tag) && !insideTable) tag = "div"; // orphan table element → neutral box
if (violatesContentModel(node, tag)) tag = "div"; if (violatesContentModel(node, tag)) tag = "div";
return tag; return tag;
@@ -579,7 +621,11 @@ function emitChildren(children: IRChild[], parentTag: string | null, assetMap: M
textBuf += c.text; textBuf += c.text;
continue; continue;
} }
if (parentTag === "video" && (c.tag === "source" || c.tag === "track")) continue; if (parentTag === "video" && c.tag === "track") continue; // caption files are not captured
// A <picture>/<video> `<source>` is emitted only when a candidate materialized
// locally: a video source otherwise falls back to poster-only rendering, a
// picture source must never point its media band at a placeholder.
if ((parentTag === "video" || parentTag === "picture") && c.tag === "source" && !sourceHasLocalCandidate(c, assetMap, sourceUrl)) continue;
// Section split: a section-root child is hoisted into its own module and replaced // Section split: a section-root child is hoisted into its own module and replaced
// by a `<HeroSection />` placeholder. Rendered once (subtree → module body); the // by a `<HeroSection />` placeholder. Rendered once (subtree → module body); the
// composed DOM is identical to inlining (same tags/cids/classes). // composed DOM is identical to inlining (same tags/cids/classes).
@@ -1117,8 +1163,11 @@ function emitVariantSkeleton(componentName: string, instances: IRNode[], variant
instances.forEach((n, k) => { runText![k] += (n.children[i] as IRTextNode).text; }); instances.forEach((n, k) => { runText![k] += (n.children[i] as IRTextNode).text; });
continue; continue;
} }
if (tag === "video" && (repr.children[i] as IRNode).tag === "source") continue;
if (tag === "video" && (repr.children[i] as IRNode).tag === "track") continue; if (tag === "video" && (repr.children[i] as IRNode).tag === "track") continue;
// Mirror emitChildren: a <source> child survives only with a materialized candidate
// (judged on the representative — instances share the capture's asset set).
if ((tag === "video" || tag === "picture") && (repr.children[i] as IRNode).tag === "source"
&& !sourceHasLocalCandidate(repr.children[i] as IRNode, assetMap, sourceUrl)) continue;
flushText(); flushText();
const subNodes = instances.map((n) => n.children[i] as IRNode); const subNodes = instances.map((n) => n.children[i] as IRNode);
const sub = emitVariantSkeleton(componentName, subNodes, variants, childInteractive, indent + 1, dataRows, styleRows, cids, gen, styleGen, slots, assetMap, sourceUrl, ctx, childTable, [...ancestors, repr.tag]); const sub = emitVariantSkeleton(componentName, subNodes, variants, childInteractive, indent + 1, dataRows, styleRows, cids, gen, styleGen, slots, assetMap, sourceUrl, ctx, childTable, [...ancestors, repr.tag]);
@@ -1498,8 +1547,11 @@ function emitSkeleton(instances: IRNode[], insideInteractive: boolean, indent: n
instances.forEach((n, k) => { runText![k] += (n.children[i] as IRTextNode).text; }); instances.forEach((n, k) => { runText![k] += (n.children[i] as IRTextNode).text; });
continue; continue;
} }
if (tag === "video" && (repr.children[i] as IRNode).tag === "source") continue;
if (tag === "video" && (repr.children[i] as IRNode).tag === "track") continue; if (tag === "video" && (repr.children[i] as IRNode).tag === "track") continue;
// Mirror emitChildren: a <source> child survives only with a materialized candidate
// (judged on the representative — instances share the capture's asset set).
if ((tag === "video" || tag === "picture") && (repr.children[i] as IRNode).tag === "source"
&& !sourceHasLocalCandidate(repr.children[i] as IRNode, assetMap, sourceUrl)) continue;
flushText(); flushText();
const sub = emitSkeleton(instances.map((n) => n.children[i] as IRNode), childInteractive, indent + 1, dataRows, styleRows, cids, gen, styleGen, assetMap, sourceUrl, ctx, childTable, [...ancestors, repr.tag]); const sub = emitSkeleton(instances.map((n) => n.children[i] as IRNode), childInteractive, indent + 1, dataRows, styleRows, cids, gen, styleGen, assetMap, sourceUrl, ctx, childTable, [...ancestors, repr.tag]);
if (sub === null) return null; if (sub === null) return null;
@@ -2478,6 +2530,7 @@ export function generateApp(input: GenerateInput, tokensCss: string): { pageTsx:
const globals = tailwindGlobalsCss({ const globals = tailwindGlobalsCss({
reset: RESET_CSS, fontCss: fontGraph.css, tokensCss: tokensCss + (tw.colorDefsCss ? "\n" + tw.colorDefsCss : ""), reset: RESET_CSS, fontCss: fontGraph.css, tokensCss: tokensCss + (tw.colorDefsCss ? "\n" + tw.colorDefsCss : ""),
htmlBg, bodyFont: SYSTEM_FALLBACK, clip, colorTokens: tw.colorTokens, viewports: ir.doc.viewports, htmlBg, bodyFont: SYSTEM_FALLBACK, clip, colorTokens: tw.colorTokens, viewports: ir.doc.viewports,
canonical: ir.doc.canonicalViewport,
}); });
writeText(join(rootDir, "globals.css"), framework === "vite" ? viteGlobalsCss(globals) : globals); writeText(join(rootDir, "globals.css"), framework === "vite" ? viteGlobalsCss(globals) : globals);
} else { } else {
@@ -2685,6 +2738,8 @@ const nextConfig = {
reactStrictMode: false, reactStrictMode: false,
eslint: { ignoreDuringBuilds: true }, eslint: { ignoreDuringBuilds: true },
typescript: { ignoreBuildErrors: true }, typescript: { ignoreBuildErrors: true },
// The dev-tools badge would leak into reviewer/validator screenshots.
devIndicators: false,
}; };
export default nextConfig; export default nextConfig;
`; `;
+2 -1
View File
@@ -44,6 +44,7 @@ function signature(nr: NodeRule): string {
for (const b of nr.bands) s += `@${b.media}{${serDecls(b.decls)}}`; for (const b of nr.bands) s += `@${b.media}{${serDecls(b.decls)}}`;
if (nr.before) { s += "::before{" + serDecls(nr.before.base); for (const b of nr.before.bands) s += `@${b.media}{${serDecls(b.decls)}}`; s += "}"; } if (nr.before) { s += "::before{" + serDecls(nr.before.base); for (const b of nr.before.bands) s += `@${b.media}{${serDecls(b.decls)}}`; s += "}"; }
if (nr.after) { s += "::after{" + serDecls(nr.after.base); for (const b of nr.after.bands) s += `@${b.media}{${serDecls(b.decls)}}`; s += "}"; } if (nr.after) { s += "::after{" + serDecls(nr.after.base); for (const b of nr.after.bands) s += `@${b.media}{${serDecls(b.decls)}}`; s += "}"; }
if (nr.placeholder) { s += "::placeholder{" + serDecls(nr.placeholder.base); for (const b of nr.placeholder.bands) s += `@${b.media}{${serDecls(b.decls)}}`; s += "}"; }
return s; return s;
} }
@@ -125,7 +126,7 @@ function splitRule(nr: NodeRule): { typo: NodeRule; layout: NodeRule; box: NodeR
const b0 = partition(nr.base); const b0 = partition(nr.base);
const typo: NodeRule = { base: b0.typo, bands: [] }; const typo: NodeRule = { base: b0.typo, bands: [] };
const layout: NodeRule = { base: b0.layout, bands: [] }; const layout: NodeRule = { base: b0.layout, bands: [] };
const box: NodeRule = { base: b0.box, bands: [], before: nr.before, after: nr.after }; const box: NodeRule = { base: b0.box, bands: [], before: nr.before, after: nr.after, placeholder: nr.placeholder };
for (const band of nr.bands) { for (const band of nr.bands) {
const bs = partition(band.decls); const bs = partition(band.decls);
if (bs.typo.size) typo.bands.push({ media: band.media, decls: bs.typo }); if (bs.typo.size) typo.bands.push({ media: band.media, decls: bs.typo });
+92 -14
View File
@@ -91,6 +91,11 @@ const GENERIC: Array<{ prop: string; def: string | string[] }> = [
{ prop: "bottom", def: "auto" }, { prop: "left", def: "auto" }, { prop: "bottom", def: "auto" }, { prop: "left", def: "auto" },
{ prop: "float", def: "none" }, { prop: "clear", def: "none" }, { prop: "float", def: "none" }, { prop: "clear", def: "none" },
{ prop: "zIndex", def: "auto" }, { prop: "zIndex", def: "auto" },
// visibility is INHERITED, so the parent-equality skip keeps descendants of a
// hidden subtree clean; without this entry a node hidden at the canonical
// viewport could never emit `visibility:hidden` at all (the only other path is
// per-band and gated on being shown at base).
{ prop: "visibility", def: "visible" },
{ prop: "opacity", def: "1" }, { prop: "isolation", def: "auto" }, { prop: "opacity", def: "1" }, { prop: "isolation", def: "auto" },
{ prop: "mixBlendMode", def: "normal" }, { prop: "mixBlendMode", def: "normal" },
{ prop: "minWidth", def: ["0px", "auto"] }, { prop: "maxWidth", def: "none" }, { prop: "minWidth", def: ["0px", "auto"] }, { prop: "maxWidth", def: "none" },
@@ -2305,7 +2310,14 @@ function declsForViewport(
} }
// Inherited: skip when equal to parent's value (inheritance handles it). // Inherited: skip when equal to parent's value (inheritance handles it).
if (INHERITED.has(prop) && parentComputed && parentComputed[prop] === value) continue; // Exception: the reset (`ul, ol, menu { list-style: none; }`) breaks the list-marker
// inheritance chain on those tags, so parent-equality is not a safe elision there —
// a source <ul> with `disc` equals its parent's initial `disc` yet must still emit
// or the reset erases the markers. <li> inherits from the ul, which now emits.
if (INHERITED.has(prop) && parentComputed && parentComputed[prop] === value) {
const listMarkerReset = (prop === "listStyleType" || prop === "listStylePosition") && /^(ul|ol|menu)$/.test(tag);
if (!listMarkerReset) continue;
}
let outValue = value; let outValue = value;
if (prop === "backgroundImage" || prop === "maskImage" || prop === "filter" || prop === "clipPath") { if (prop === "backgroundImage" || prop === "maskImage" || prop === "filter" || prop === "clipPath") {
@@ -2586,7 +2598,41 @@ function pseudoDecls(style: StyleMap, assetMap: Map<string, string>): Map<string
// the grouped class produces identical computed styles (fidelity-neutral). // the grouped class produces identical computed styles (fidelity-neutral).
export type BandRule = { media: string; decls: Map<string, string> }; export type BandRule = { media: string; decls: Map<string, string> };
export type PseudoRule = { base: Map<string, string>; bands: BandRule[] }; export type PseudoRule = { base: Map<string, string>; bands: BandRule[] };
export type NodeRule = { base: Map<string, string>; bands: BandRule[]; before?: PseudoRule; after?: PseudoRule }; export type NodeRule = { base: Map<string, string>; bands: BandRule[]; before?: PseudoRule; after?: PseudoRule; placeholder?: PseudoRule };
/** ::placeholder declarations for a form control. Color always emits (the UA default is
* its own gray, NOT inherited from the input, so equality with the host proves nothing);
* font/spacing props DO inherit from the input inside the pseudo, so they emit only when
* they differ from the host's own computed value. */
function placeholderDecls(style: StyleMap, host: StyleMap | undefined): Map<string, string> {
const decls = new Map<string, string>();
if (style.color) decls.set("color", style.color);
if (style.opacity && style.opacity !== "1") decls.set("opacity", style.opacity);
for (const prop of ["fontFamily", "fontSize", "fontWeight", "fontStyle", "letterSpacing", "textTransform"]) {
const v = style[prop];
if (!v || (host && host[prop] === v)) continue;
decls.set(kebab(prop), v);
}
return decls;
}
/** Banded ::placeholder rule (mirrors collectPseudoRule's base+delta shape). */
function collectPlaceholderRule(styleByVp: Record<number, StyleMap>, hostByVp: Record<number, StyleMap>, baseVp: number, bands: Band[], tokenResolver?: TokenResolver): PseudoRule | undefined {
const baseStyle = styleByVp[baseVp] ?? Object.values(styleByVp)[0];
if (!baseStyle) return undefined;
const out: PseudoRule = { base: finalizeDecls(placeholderDecls(baseStyle, hostByVp[baseVp]), tokenResolver), bands: [] };
for (const b of bands) {
if (!b.media) continue;
const st = styleByVp[b.vp];
if (!st) continue; // control not rendered at this width — the host rule already hides it
const vpDecls = finalizeDecls(placeholderDecls(st, hostByVp[b.vp]), tokenResolver);
const delta = new Map<string, string>();
for (const [k, v] of vpDecls) if (out.base.get(k) !== v) delta.set(k, v);
for (const [k] of out.base) if (!vpDecls.has(k)) delta.set(k, resetValue(k));
if (delta.size > 0) out.bands.push({ media: b.media, decls: delta });
}
return out.base.size > 0 ? out : undefined;
}
/** Collect a pseudo-element's banded rule (its size/position can be responsive /** Collect a pseudo-element's banded rule (its size/position can be responsive
* e.g. flex spacer pseudo-elements in horizontal scrollers). `hostContentWidthByVp` * e.g. flex spacer pseudo-elements in horizontal scrollers). `hostContentWidthByVp`
@@ -2940,20 +2986,47 @@ export function collectNodeRules(ir: IR, assetMap: Map<string, string>, includeN
// Node was not in the observed DOM at this width (responsive conditional // Node was not in the observed DOM at this width (responsive conditional
// rendering): hide it so the clone matches the source at this viewport. // rendering): hide it so the clone matches the source at this viewport.
if (!node.computedByVp[b.vp]) { nr.bands.push({ media: b.media, decls: new Map([["display", "none"]]) }); continue; } if (!node.computedByVp[b.vp]) { nr.bands.push({ media: b.media, decls: new Map([["display", "none"]]) }); continue; }
// The node isn't PAINTED at this width — either it goes `display:none` itself, or an ancestor // The node isn't PAINTED at this width. HOW it is hidden decides what to emit, because only
// hides the whole subtree. Its box renders nothing, so any width/height/inset/padding override // `display:none` takes the box out of layout — a `visibility:hidden` box still occupies space
// here is invisible: pure breakpoint noise. Emit ONLY the hide, and only when the node ITSELF // and can extend the scrollable area. The base rule bakes CANONICAL geometry unconditionally,
// is the one turning off (was shown at base); if an ancestor hides it, emit nothing at all — // so skipping every override here can park e.g. a desktop `left:548px` slider arrow inside a
// that ancestor's own `display:none` already removes this node with it. // 375px viewport: invisible, but +210px of sideways scroll.
const ownNone = (node.computedByVp[b.vp]?.display || "") === "none"; const vpCs = node.computedByVp[b.vp]!;
if (ownNone || !node.visibleByVp[b.vp]) { const ownNone = (vpCs.display || "") === "none";
// The node ITSELF is visibility:hidden here (its parent isn't → not inherited from an
// ancestor whose own rule already carries the hide).
const ownHidden = !ownNone && /^(hidden|collapse)$/.test(vpCs.visibility || "") &&
!/^(hidden|collapse)$/.test(parentNode?.computedByVp[b.vp]?.visibility || "");
if (ownHidden) {
const bb = node.bboxByVp[b.vp];
if (!bb || bb.width <= 0 || bb.height <= 0) {
// The hidden box occupied NOTHING in the capture (0×0 — e.g. an uninitialised swiper
// arrow collapsed by its container). `display:none` reproduces "invisible, takes no
// space" exactly and guarantees it cannot extend scroll bounds at this width.
nr.bands.push({ media: b.media, decls: new Map([["display", "none"]]) });
continue;
}
// Non-zero bbox: the invisible box occupies real layout space. Fall through to the normal
// per-viewport delta so it sits where the capture measured it at THIS width — not at the
// canonical geometry the base rule baked. The hide itself rides along in the delta
// (declsForViewport emits `visibility:hidden`; parent-equality keeps it own-only).
} else if (ownNone || !node.visibleByVp[b.vp]) {
const shownAtBase = node.visibleByVp[baseVp] && (node.computedByVp[baseVp]?.display || "") !== "none"; const shownAtBase = node.visibleByVp[baseVp] && (node.computedByVp[baseVp]?.display || "") !== "none";
if (ownNone && shownAtBase) nr.bands.push({ media: b.media, decls: new Map([["display", "none"]]) }); if (ownNone) {
else if (shownAtBase) { // Own display:none removes the box from layout entirely. Emit the hide even when the node
const vpCs = node.computedByVp[b.vp]; // is ALSO hidden at base — a visibility:hidden base still bakes an OCCUPYING box (see
// above), so without this band the canonical geometry would render at this width. Skip
// only when the base itself is display:none (the band would be redundant).
if ((node.computedByVp[baseVp]?.display || "") !== "none") {
nr.bands.push({ media: b.media, decls: new Map([["display", "none"]]) });
}
} else if (shownAtBase) {
// Hidden by an ancestor (or zero-size / opacity:0): geometry overrides are breakpoint
// noise — the ancestor's own hide (or the reveal replay, for scroll-reveal opacity)
// covers it. Emit only the hide the node itself carries.
const hide = new Map<string, string>(); const hide = new Map<string, string>();
if (vpCs && pf(vpCs.opacity) === 0 && !animOwned.has("opacity")) hide.set("opacity", "0"); if (pf(vpCs.opacity) === 0 && !animOwned.has("opacity")) hide.set("opacity", "0");
if (vpCs && /^(hidden|collapse)$/.test(vpCs.visibility || "")) hide.set("visibility", "hidden"); if (/^(hidden|collapse)$/.test(vpCs.visibility || "")) hide.set("visibility", "hidden");
if (hide.size) nr.bands.push({ media: b.media, decls: hide }); if (hide.size) nr.bands.push({ media: b.media, decls: hide });
} }
continue; continue;
@@ -2997,6 +3070,9 @@ export function collectNodeRules(ir: IR, assetMap: Map<string, string>, includeN
nr.after = collectPseudoRule(node.afterByVp, baseVp, bands, assetMap, tokenResolver, hostCW, padW, padH); nr.after = collectPseudoRule(node.afterByVp, baseVp, bands, assetMap, tokenResolver, hostCW, padW, padH);
} }
} }
if (node.placeholderByVp) {
nr.placeholder = collectPlaceholderRule(node.placeholderByVp, node.computedByVp, baseVp, bands, tokenResolver);
}
rules.set(node.id, nr); rules.set(node.id, nr);
for (const c of node.children) if (!isTextChild(c)) walk(c, node, childFluid, childLayoutParent, childCb, childDefiniteHeight); for (const c of node.children) if (!isTextChild(c)) walk(c, node, childFluid, childLayoutParent, childCb, childDefiniteHeight);
@@ -3026,9 +3102,11 @@ export function assembleCss(
if (nr.base.size > 0) baseRules.push(formatRule(sel, nr.base)); if (nr.base.size > 0) baseRules.push(formatRule(sel, nr.base));
if (nr.before) baseRules.push(formatRule(`${sel}::before`, nr.before.base)); if (nr.before) baseRules.push(formatRule(`${sel}::before`, nr.before.base));
if (nr.after) baseRules.push(formatRule(`${sel}::after`, nr.after.base)); if (nr.after) baseRules.push(formatRule(`${sel}::after`, nr.after.base));
if (nr.placeholder) baseRules.push(formatRule(`${sel}::placeholder`, nr.placeholder.base));
for (const b of nr.bands) bandRules.get(b.media)?.push(formatRule(sel, b.decls)); for (const b of nr.bands) bandRules.get(b.media)?.push(formatRule(sel, b.decls));
if (nr.before) for (const b of nr.before.bands) bandRules.get(b.media)?.push(formatRule(`${sel}::before`, b.decls)); if (nr.before) for (const b of nr.before.bands) bandRules.get(b.media)?.push(formatRule(`${sel}::before`, b.decls));
if (nr.after) for (const b of nr.after.bands) bandRules.get(b.media)?.push(formatRule(`${sel}::after`, b.decls)); if (nr.after) for (const b of nr.after.bands) bandRules.get(b.media)?.push(formatRule(`${sel}::after`, b.decls));
if (nr.placeholder) for (const b of nr.placeholder.bands) bandRules.get(b.media)?.push(formatRule(`${sel}::placeholder`, b.decls));
} }
const parts: string[] = []; const parts: string[] = [];
if (keyframes) parts.push(keyframes); if (keyframes) parts.push(keyframes);
+64 -15
View File
@@ -19,7 +19,13 @@ import type { MotionCapture, WaapiAnim, RotatorSpec, RevealSpec, MarqueeSpec } f
export type RTWaapi = { cid: string; keyframes: Array<Record<string, string | number>>; duration: number; delay: number; easing: string; iterations: number; direction: string; fill: string }; export type RTWaapi = { cid: string; keyframes: Array<Record<string, string | number>>; duration: number; delay: number; easing: string; iterations: number; direction: string; fill: string };
export type RTRotator = { cid: string; texts: string[]; intervalMs: number }; export type RTRotator = { cid: string; texts: string[]; intervalMs: number };
export type RTReveal = { cid: string; opacity: string; transform: string; transition: string }; // Reveal families: transition (hidden via opacity/transform, revealed by the element's own
// transition) and visibility+entrance-class (hidden via visibility, revealed with a named
// keyframe animation — Elementor/WOW/AOS; the @keyframes ship in the page CSS).
export type RTReveal = {
cid: string; opacity: string; transform: string; transition: string;
visibility?: "hidden"; animationName?: string; animationDuration?: string; animationDelay?: string; animationTiming?: string;
};
export type RTMarquee = { cid: string; pxPerSec: number; periodPx: number }; export type RTMarquee = { cid: string; pxPerSec: number; periodPx: number };
export type MotionSpec = { waapi: RTWaapi[]; rotators: RTRotator[]; reveals: RTReveal[]; marquees: RTMarquee[] }; export type MotionSpec = { waapi: RTWaapi[]; rotators: RTRotator[]; reveals: RTReveal[]; marquees: RTMarquee[] };
@@ -58,7 +64,16 @@ export function buildMotionSpec(ir: IR, motion: MotionCapture | undefined, inclu
for (const rv of (motion.reveals ?? []) as RevealSpec[]) { for (const rv of (motion.reveals ?? []) as RevealSpec[]) {
const cid = map.get(rv.cap); const cid = map.get(rv.cap);
if (!ok(cid)) continue; if (!ok(cid)) continue;
reveals.push({ cid, opacity: rv.opacity, transform: rv.transform, transition: rv.transition }); reveals.push({
cid, opacity: rv.opacity, transform: rv.transform, transition: rv.transition,
...(rv.visibility === "hidden" ? { visibility: rv.visibility } : {}),
...(rv.animationName ? {
animationName: rv.animationName,
animationDuration: rv.animationDuration,
animationDelay: rv.animationDelay,
animationTiming: rv.animationTiming,
} : {}),
});
} }
const marquees: RTMarquee[] = []; const marquees: RTMarquee[] = [];
for (const m of (motion.marquees ?? []) as MarqueeSpec[]) { for (const m of (motion.marquees ?? []) as MarqueeSpec[]) {
@@ -92,7 +107,10 @@ import { useEffect } from "react";
type RTWaapi = { cid: string; keyframes: Array<Record<string, string | number>>; duration: number; delay: number; easing: string; iterations: number; direction: string; fill: string }; type RTWaapi = { cid: string; keyframes: Array<Record<string, string | number>>; duration: number; delay: number; easing: string; iterations: number; direction: string; fill: string };
type RTRotator = { cid: string; texts: string[]; intervalMs: number }; type RTRotator = { cid: string; texts: string[]; intervalMs: number };
type RTReveal = { cid: string; opacity: string; transform: string; transition: string }; type RTReveal = {
cid: string; opacity: string; transform: string; transition: string;
visibility?: "hidden"; animationName?: string; animationDuration?: string; animationDelay?: string; animationTiming?: string;
};
type RTMarquee = { cid: string; pxPerSec: number; periodPx: number }; type RTMarquee = { cid: string; pxPerSec: number; periodPx: number };
export type MotionSpec = { waapi: RTWaapi[]; rotators: RTRotator[]; reveals: RTReveal[]; marquees: RTMarquee[] }; export type MotionSpec = { waapi: RTWaapi[]; rotators: RTRotator[]; reveals: RTReveal[]; marquees: RTMarquee[] };
@@ -111,7 +129,8 @@ export default function DittoMotion({ spec }: { spec: MotionSpec }) {
const intervals: ReturnType<typeof setInterval>[] = []; const intervals: ReturnType<typeof setInterval>[] = [];
const rotators: Array<{ el: HTMLElement; original: string | null }> = []; const rotators: Array<{ el: HTMLElement; original: string | null }> = [];
const anims: Animation[] = []; const anims: Animation[] = [];
const revealed: Array<() => void> = []; // per-reveal "show now" fns (also the cleanup) // per-reveal "show now" fns (also the cleanup); animate=false jumps to the settled frame
const revealed: Array<(animate: boolean) => void> = [];
let io: IntersectionObserver | null = null; let io: IntersectionObserver | null = null;
let forceTimer: ReturnType<typeof setTimeout> | null = null; let forceTimer: ReturnType<typeof setTimeout> | null = null;
@@ -153,29 +172,59 @@ export default function DittoMotion({ spec }: { spec: MotionSpec }) {
intervals.push(setInterval(() => { i = (i + 1) % r.texts.length; el.textContent = r.texts[i]!; }, Math.max(400, r.intervalMs))); intervals.push(setInterval(() => { i = (i + 1) % r.texts.length; el.textContent = r.texts[i]!; }, Math.max(400, r.intervalMs)));
} }
// Scroll reveals: hide each element (opacity/transform) with the captured transition, // Scroll reveals: re-hide each element (JS-applied on mount, so non-JS/SSR still shows
// then reveal (clear the inline overrides → transitions to the base CSS) when it scrolls // the content), then reveal when it scrolls into view. Two families:
// into view. A force-reveal timer guarantees nothing stays hidden if the observer misses. // - transition — hide via opacity/transform, reveal by transitioning to full;
// - visibility+entrance-class (Elementor/WOW/AOS) — hide via visibility with the baked
// entrance animation suppressed, reveal by restarting the captured @keyframes
// (they ship in the page CSS under the captured animation-name).
// A force-reveal timer guarantees nothing stays hidden if the observer misses.
if (spec.reveals.length) { if (spec.reveals.length) {
// Reveal to the full resting state. Setting 1/none (not clearing to base) is correct for // Reveal to the full resting state. Setting 1/none (not clearing to base) is correct for
// every reveal — the revealed state is always full + un-offset — and is REQUIRED for // every reveal — the revealed state is always full + un-offset — and is REQUIRED for
// scroll-scrub panels whose captured base CSS is a frozen mid-scrub value (opacity 0.63). // scroll-scrub panels whose captured base CSS is a frozen mid-scrub value (opacity 0.63).
const show = (el: HTMLElement) => { el.style.opacity = "1"; el.style.transform = "none"; }; // animate=false (validator settle path) jumps straight to the settled frame so no
const byEl = new Map<Element, HTMLElement>(); // measurement can catch a mid-entrance value.
const show = (el: HTMLElement, rv: RTReveal, animate: boolean) => {
el.style.opacity = "1"; el.style.transform = "none";
if (rv.visibility === "hidden") el.style.visibility = "visible";
if (rv.animationName) {
if (animate) {
// restart the entrance from t=0: none -> name starts a fresh animation
el.style.animationName = "none";
void el.offsetWidth;
el.style.animationName = rv.animationName;
el.style.animationDuration = rv.animationDuration || "1s";
el.style.animationDelay = rv.animationDelay || "0s";
el.style.animationTimingFunction = rv.animationTiming || "ease";
el.style.animationFillMode = "both";
el.style.animationIterationCount = "1";
} else {
el.style.animationName = "none"; // settled frame: keep the entrance suppressed
}
}
};
const shows = new Map<Element, (animate: boolean) => void>();
for (const rv of spec.reveals) { for (const rv of spec.reveals) {
const el = byCid(rv.cid); const el = byCid(rv.cid);
if (!el) continue; if (!el) continue;
if (rv.visibility === "hidden") {
el.style.visibility = "hidden";
el.style.animationName = "none"; // don't burn the baked entrance while hidden
} else {
el.style.transition = rv.transition; el.style.transition = rv.transition;
el.style.opacity = rv.opacity; el.style.opacity = rv.opacity;
if (rv.transform !== "none") el.style.transform = rv.transform; if (rv.transform !== "none") el.style.transform = rv.transform;
byEl.set(el, el); }
revealed.push(() => show(el)); const fn = (animate: boolean) => show(el, rv, animate);
shows.set(el, fn);
revealed.push(fn);
} }
io = new IntersectionObserver((entries) => { io = new IntersectionObserver((entries) => {
for (const e of entries) if (e.isIntersecting) { const el = byEl.get(e.target); if (el) { show(el); io!.unobserve(e.target); } } for (const e of entries) if (e.isIntersecting) { const f = shows.get(e.target); if (f) { f(true); io!.unobserve(e.target); } }
}, { rootMargin: "0px 0px -8% 0px" }); }, { rootMargin: "0px 0px -8% 0px" });
for (const el of byEl.keys()) io.observe(el); for (const el of shows.keys()) io.observe(el);
forceTimer = setTimeout(() => { for (const f of revealed) f(); }, 4000); forceTimer = setTimeout(() => { for (const f of revealed) f(true); }, 4000);
} }
const stopAll = () => { const stopAll = () => {
@@ -185,7 +234,7 @@ export default function DittoMotion({ spec }: { spec: MotionSpec }) {
for (const a of anims) { try { a.cancel(); } catch { /* ignore */ } } for (const a of anims) { try { a.cancel(); } catch { /* ignore */ } }
if (io) io.disconnect(); if (io) io.disconnect();
if (forceTimer) clearTimeout(forceTimer); if (forceTimer) clearTimeout(forceTimer);
for (const f of revealed) f(); // reveal everything → base CSS settled frame for (const f of revealed) f(false); // reveal everything, settled → base CSS graded frame
}; };
// Measurement hook: restore the fully-settled/revealed base for grading. // Measurement hook: restore the fully-settled/revealed base for grading.
(window as any).__dittoMotionStop = stopAll; (window as any).__dittoMotionStop = stopAll;
+15 -1
View File
@@ -273,6 +273,15 @@ function iconUrl(icon: SeoInventory["icons"][number]): string {
return icon.localPath || icon.href; return icon.localPath || icon.href;
} }
// Next validates metadata.openGraph.type against this fixed enum and THROWS at
// render on anything else (e.g. Shopify's `product.group`). Unsupported values
// are emitted via metadata.other instead so the tag survives without the crash.
const NEXT_OG_TYPES = new Set([
"website", "article", "book", "profile",
"music.song", "music.album", "music.playlist", "music.radio_station",
"video.movie", "video.episode", "video.tv_show", "video.other",
]);
function metadataObject(report: SeoInventory): Record<string, unknown> { function metadataObject(report: SeoInventory): Record<string, unknown> {
const metadata: Record<string, unknown> = { title: report.title || "Cloned Page" }; const metadata: Record<string, unknown> = { title: report.title || "Cloned Page" };
if (report.description) metadata.description = report.description; if (report.description) metadata.description = report.description;
@@ -297,9 +306,13 @@ function metadataObject(report: SeoInventory): Record<string, unknown> {
const ogSiteName = firstValue(ogEntries, "og:site_name"); const ogSiteName = firstValue(ogEntries, "og:site_name");
const ogUrl = firstValue(ogEntries, "og:url"); const ogUrl = firstValue(ogEntries, "og:url");
const ogImages = ogEntries.filter((entry) => entry.property?.toLowerCase() === "og:image").map((entry) => entry.content); const ogImages = ogEntries.filter((entry) => entry.property?.toLowerCase() === "og:image").map((entry) => entry.content);
const other: Record<string, string> = {};
if (ogTitle) og.title = ogTitle; if (ogTitle) og.title = ogTitle;
if (ogDescription) og.description = ogDescription; if (ogDescription) og.description = ogDescription;
if (ogType) og.type = ogType; if (ogType) {
if (NEXT_OG_TYPES.has(ogType)) og.type = ogType;
else other["og:type"] = ogType;
}
if (ogSiteName) og.siteName = ogSiteName; if (ogSiteName) og.siteName = ogSiteName;
if (ogUrl) og.url = ogUrl; if (ogUrl) og.url = ogUrl;
if (ogImages.length) og.images = ogImages; if (ogImages.length) og.images = ogImages;
@@ -335,6 +348,7 @@ function metadataObject(report: SeoInventory): Record<string, unknown> {
} }
if (Object.keys(icons).length) metadata.icons = icons; if (Object.keys(icons).length) metadata.icons = icons;
if (report.manifest) metadata.manifest = report.manifest.localPath || report.manifest.href; if (report.manifest) metadata.manifest = report.manifest.localPath || report.manifest.href;
if (Object.keys(other).length) metadata.other = other;
return metadata; return metadata;
} }
+62 -15
View File
@@ -99,6 +99,14 @@ const SPACE_SCALE = new Set<string>([
]); ]);
// Props that take 100% → the `-full` named utility (exact: w-full ≡ width:100%). // Props that take 100% → the `-full` named utility (exact: w-full ≡ width:100%).
const PCT_FULL = new Set<string>(["width", "height", "min-width", "max-width", "min-height", "max-height"]); const PCT_FULL = new Set<string>(["width", "height", "min-width", "max-width", "min-height", "max-height"]);
// ARB props whose named Tailwind v4 scale really includes a bare `0` step: the spacing-scale props
// (w-0, top-0, leading-0, indent-0, …) plus the numeric scales grow/shrink/basis/order/z/opacity.
// font-size, letter-spacing, the radius corners and object-position have NO `-0` utility — `text-0`
// / `tracking-0` compile to NOTHING in v4 (a silent no-op: a map label captured at font-size:0
// painted at the inherited 20px) — so their zeros must stay arbitrary (`text-[0px]`).
const ZERO_NAMED = new Set<string>([
...SPACE_SCALE, "flex-grow", "flex-shrink", "flex-basis", "order", "z-index", "opacity",
]);
const ORIGIN_NAMED: Record<string, string> = { const ORIGIN_NAMED: Record<string, string> = {
center: "origin-center", top: "origin-top", "top right": "origin-top-right", right: "origin-right", center: "origin-center", top: "origin-top", "top right": "origin-top-right", right: "origin-right",
"bottom right": "origin-bottom-right", bottom: "origin-bottom", "bottom left": "origin-bottom-left", "bottom right": "origin-bottom-right", bottom: "origin-bottom", "bottom left": "origin-bottom-left",
@@ -141,8 +149,9 @@ function transformNeedsOrigin(v: string | undefined): boolean {
return !!v && v !== "none" && translateOffsets(v) === null; return !!v && v !== "none" && translateOffsets(v) === null;
} }
/** Translate ONE computed declaration into a Tailwind utility (no responsive prefix). */ /** Translate ONE computed declaration into a Tailwind utility (no responsive prefix).
function declToUtil(prop: string, value: string): string { * Exported for tests (the zero-value gating is load-bearing: an invalid utility is a SILENT no-op). */
export function declToUtil(prop: string, value: string): string {
// Colors → named theme tokens when tokenized (bg-primary), else arbitrary value. // Colors → named theme tokens when tokenized (bg-primary), else arbitrary value.
if (prop === "color") { const n = tokenName(value); return n ? `text-${n}` : `text-[color:${arb(value)}]`; } if (prop === "color") { const n = tokenName(value); return n ? `text-${n}` : `text-[color:${arb(value)}]`; }
if (prop === "background-color") { const n = tokenName(value); return n ? `bg-${n}` : `bg-[${arb(value)}]`; } if (prop === "background-color") { const n = tokenName(value); return n ? `bg-${n}` : `bg-[${arb(value)}]`; }
@@ -229,7 +238,11 @@ function declToUtil(prop: string, value: string): string {
const n = spacingSteps(value); const n = spacingSteps(value);
if (n !== null) return n < 0 ? `-${ARB[prop]}-${-n}` : `${ARB[prop]}-${n}`; if (n !== null) return n < 0 ? `-${ARB[prop]}-${-n}` : `${ARB[prop]}-${n}`;
} }
if (value === "0" || value === "0px" || value === "0rem") return `${ARB[prop]}-0`; // bare/unitless zero // Bare/unitless zero — the named `-0` only where that utility actually exists (ZERO_NAMED);
// elsewhere emit the exact arbitrary length so the declaration really compiles.
if (value === "0" || value === "0px" || value === "0rem") {
return ZERO_NAMED.has(prop) ? `${ARB[prop]}-0` : `${ARB[prop]}-[0px]`;
}
return `${ARB[prop]}-[${arb(value)}]`; return `${ARB[prop]}-[${arb(value)}]`;
} }
if (/^border-(top|right|bottom|left)-width$/.test(prop)) { if (/^border-(top|right|bottom|left)-width$/.test(prop)) {
@@ -261,25 +274,47 @@ function declToUtil(prop: string, value: string): string {
/** Responsive variant prefix for a band's media query. For the standard capture ladder /** Responsive variant prefix for a band's media query. For the standard capture ladder
* (375/768/1280/1920, canonical 1280) the per-viewport bands are the midpoints 571/1024/1600, * (375/768/1280/1920, canonical 1280) the per-viewport bands are the midpoints 571/1024/1600,
* which we map to the CONVENTIONAL Tailwind breakpoints a human would author `max-md:` * which we map to the CONVENTIONAL Tailwind breakpoint NAMES a human would author `max-md:`
* (mobile, <768), `md:max-lg:` (tablet, 7681023), `2xl:` (wide, 1536) leaving 1280 as the * (mobile), `md:max-lg:` (tablet), `2xl:` (wide) leaving 1280 as the unprefixed base. The
* unprefixed base. Each reproduces the captured width EXACTLY (the style gate measures only at * generated @theme redefines those screens to the band boundaries (bandScreens), so every named
* 375/768/1280/1920), while between-width behaviour follows the standard breakpoints instead of * variant flips at EXACTLY the same width as the ditto.css band media query (Tailwind's stock
* arbitrary midpoint pixels. Non-standard viewport sets fall back to the exact arbitrary band. */ * 768/1024/1536 would open windows e.g. 15361600 where utilities and ditto.css rules
function prefixFor(media: string): string { * disagree). Non-standard viewport sets fall back to the exact arbitrary band. */
export function prefixFor(media: string): string {
const min = /min-width:\s*(\d+)px/.exec(media); const min = /min-width:\s*(\d+)px/.exec(media);
const max = /max-width:\s*(\d+)px/.exec(media); const max = /max-width:\s*(\d+)px/.exec(media);
const lo = min ? +min[1]! : 0; const lo = min ? +min[1]! : 0;
const hi = max ? +max[1]! : Infinity; const hi = max ? +max[1]! : Infinity;
if (!min && hi === 571) return "max-md:"; // 375 sample → below md (768) if (!min && hi === 571) return "max-md:"; // 375 sample band (md pinned to 572)
if (lo === 572 && hi === 1024) return "md:max-lg:"; // 768 sample → md … below lg (1024) if (lo === 572 && hi === 1024) return "md:max-lg:"; // 768 sample band (lg pinned to 1025)
if (lo === 1601 && !max) return "2xl:"; // 1920 sample → at/above 2xl (1536) if (lo === 1601 && !max) return "2xl:"; // 1920 sample band (2xl pinned to 1601)
let p = ""; let p = "";
if (min) p += `min-[${lo}px]:`; if (min) p += `min-[${lo}px]:`;
if (max) p += `max-[${hi}px]:`; // v4 max-* is STRICT (`width < X`) while the band's max-width is inclusive → pin to hi+1.
if (max) p += `max-[${hi + 1}px]:`;
return p; return p;
} }
/** Named-screen overrides for the generated @theme, derived from the SAME bands prefixFor
* names. Invariant: for every band, the Tailwind variant prefix and the ditto.css media
* query produce the same min/max boundaries. Tailwind v4 compiles `X:` to `width >= X`
* and `max-X:` to `width < X`, so a band min pins the screen directly and a band max pins
* it to max+1. Bands prefixFor leaves arbitrary (`min-[…px]:`) embed their exact bounds
* and need no override. */
export function bandScreens(viewports: number[], canonical: number): Array<[string, number]> {
const out = new Map<string, number>();
for (const b of computeBands(viewports, canonical)) {
if (!b.media) continue;
const min = /min-width:\s*(\d+)px/.exec(b.media);
const max = /max-width:\s*(\d+)px/.exec(b.media);
for (const m of prefixFor(b.media).matchAll(/(max-)?(sm|md|lg|xl|2xl):/g)) {
out.set(m[2]!, m[1] ? +max![1]! + 1 : +min![1]!);
}
}
const order = ["sm", "md", "lg", "xl", "2xl"];
return [...out.entries()].sort((a, b) => order.indexOf(a[0]) - order.indexOf(b[0]));
}
function fmtRule(sel: string, decls: Map<string, string>): string { function fmtRule(sel: string, decls: Map<string, string>): string {
return `${sel}{${[...decls].map(([k, v]) => `${k}:${v}`).join(";")}}`; return `${sel}{${[...decls].map(([k, v]) => `${k}:${v}`).join(";")}}`;
} }
@@ -1141,6 +1176,13 @@ export function buildTailwind(ir: IR, assetMap: Map<string, string>, colorVar?:
classOf.set(cid, dedupeUtils([...(classOf.get(cid) ?? "").split(" ").filter(Boolean), ...filteredPseudoUtils]).join(" ")); classOf.set(cid, dedupeUtils([...(classOf.get(cid) ?? "").split(" ").filter(Boolean), ...filteredPseudoUtils]).join(" "));
utilCount += filteredPseudoUtils.length; utilCount += filteredPseudoUtils.length;
} }
// ::placeholder → a raw [data-cid] ditto.css rule (rare — only styled form controls),
// matching the pseudo fallback path above: exact colors, no Tailwind-escape round-trip.
if (nr.placeholder) {
const psel = `${sel}::placeholder`;
if (nr.placeholder.base.size) extraParts.push(fmtRule(psel, nr.placeholder.base));
for (const b of nr.placeholder.bands) extraParts.push(`${b.media} {\n${fmtRule(psel, b.decls)}\n}`);
}
} }
// Stage 4 hover/focus → Tailwind variant utilities folded into each node's className (replacing the // Stage 4 hover/focus → Tailwind variant utilities folded into each node's className (replacing the
@@ -1181,9 +1223,14 @@ export function buildTailwind(ir: IR, assetMap: Map<string, string>, colorVar?:
* so utilities override them. */ * so utilities override them. */
export function tailwindGlobalsCss(opts: { export function tailwindGlobalsCss(opts: {
reset: string; fontCss: string; tokensCss: string; htmlBg: string; bodyFont: string; reset: string; fontCss: string; tokensCss: string; htmlBg: string; bodyFont: string;
clip: string; colorTokens: string[]; viewports: number[]; clip: string; colorTokens: string[]; viewports: number[]; canonical: number;
}): string { }): string {
const screens = [...opts.viewports].sort((a, b) => a - b).map((v) => ` --breakpoint-vp${v}: ${v}px;`).join("\n"); const screens = [
...[...opts.viewports].sort((a, b) => a - b).map((v) => ` --breakpoint-vp${v}: ${v}px;`),
// Named screens pinned to the ditto.css band boundaries so `md:`/`2xl:` utilities
// and the banded rules flip at the same width (see bandScreens).
...bandScreens(opts.viewports, opts.canonical).map(([n, px]) => ` --breakpoint-${n}: ${px}px;`),
].join("\n");
const colors = opts.colorTokens.map((n) => ` --color-${n}: var(--${n});`).join("\n"); const colors = opts.colorTokens.map((n) => ` --color-${n}: var(--${n});`).join("\n");
return `@layer theme, base, components, utilities; return `@layer theme, base, components, utilities;
@import "tailwindcss/theme.css" layer(theme); @import "tailwindcss/theme.css" layer(theme);
+3 -5
View File
@@ -96,11 +96,9 @@ export function materializeAssets(graph: AssetGraph, sourceDir: string, appPubli
for (const e of graph.entries) { for (const e of graph.entries) {
if (e.classification !== "downloaded" || !e.storedFile || !e.localPath) continue; if (e.classification !== "downloaded" || !e.storedFile || !e.localPath) continue;
if (e.type === "css") continue; // not served by the app if (e.type === "css") continue; // not served by the app
// A <video> is rendered as its poster still (streaming sources are dropped in // Video files ARE copied: generation emits a local <video src>/<source src>
// generation), so the materialized video file is never referenced — skip // whenever the file materialized (only streaming/undownloaded videos fall back
// copying it to public/ to save disk and build time. Its first-frame poster is // to poster-only rendering).
// a separate image asset that is still materialized.
if (e.type === "video") continue;
if (seenPublicPaths?.has(e.localPath)) continue; if (seenPublicPaths?.has(e.localPath)) continue;
const src = join(storeDir, e.storedFile); const src = join(storeDir, e.storedFile);
const from = fileExists(src) ? src : join(cssDir, e.storedFile); const from = fileExists(src) ? src : join(cssDir, e.storedFile);
+18 -5
View File
@@ -34,6 +34,9 @@ export type IRNode = {
sizingByVp?: Record<number, RawSizing>; sizingByVp?: Record<number, RawSizing>;
beforeByVp?: Record<number, StyleMap>; beforeByVp?: Record<number, StyleMap>;
afterByVp?: Record<number, StyleMap>; afterByVp?: Record<number, StyleMap>;
// ::placeholder computed style (input/textarea with placeholder text) — emitted as a
// `::placeholder` rule so form controls keep their authored placeholder color.
placeholderByVp?: Record<number, StyleMap>;
children: IRChild[]; children: IRChild[];
}; };
@@ -156,10 +159,12 @@ function resolveLazyAttrs(attrs: Record<string, string> | undefined): Record<str
return out; return out;
} }
// `<iframe>` is intentionally NOT noise: it is kept as a sized placeholder box (its // `<iframe>` is intentionally NOT noise: capture grafts an embedded document's subtree
// external document is dropped at generation — see propsList — so the clone stays // as the iframe's children (capture/graft.ts) so form embeds (Klaviyo et al) render as
// self-contained while preserving the layout the embed occupied). Truly invisible // real content — generation then emits the iframe as a positioned <div>. When the graft
// iframes (tracking/chat, 0-size or display:none) are removed by the visibility prune. // wasn't possible the iframe is kept as a sized placeholder box (document-loading attrs
// dropped at generation — see propsList — so the clone stays self-contained). Truly
// invisible iframes (tracking/chat, 0-size or display:none) are removed by the prune.
const NOISE_TAGS = new Set(["next-route-announcer"]); const NOISE_TAGS = new Set(["next-route-announcer"]);
// Third-party overlay widgets (chat launchers, cookie/consent bars, captcha badges) are // Third-party overlay widgets (chat launchers, cookie/consent bars, captcha badges) are
@@ -306,6 +311,7 @@ export function buildIR(sourceDir: string, viewports: number[], opts?: { motion?
const sizingByVp: Record<number, RawSizing> = {}; const sizingByVp: Record<number, RawSizing> = {};
const beforeByVp: Record<number, StyleMap> = {}; const beforeByVp: Record<number, StyleMap> = {};
const afterByVp: Record<number, StyleMap> = {}; const afterByVp: Record<number, StyleMap> = {};
const placeholderByVp: Record<number, StyleMap> = {};
for (const vw of viewports) { for (const vw of viewports) {
const match = matched[vw]; const match = matched[vw];
@@ -316,6 +322,7 @@ export function buildIR(sourceDir: string, viewports: number[], opts?: { motion?
if (match.sizing) sizingByVp[vw] = match.sizing; if (match.sizing) sizingByVp[vw] = match.sizing;
if (match.before) beforeByVp[vw] = match.before; if (match.before) beforeByVp[vw] = match.before;
if (match.after) afterByVp[vw] = match.after; if (match.after) afterByVp[vw] = match.after;
if (match.placeholder) placeholderByVp[vw] = match.placeholder;
} }
const node: IRNode = { const node: IRNode = {
@@ -333,6 +340,7 @@ export function buildIR(sourceDir: string, viewports: number[], opts?: { motion?
if (Object.keys(sizingByVp).length) node.sizingByVp = sizingByVp; if (Object.keys(sizingByVp).length) node.sizingByVp = sizingByVp;
if (Object.keys(beforeByVp).length) node.beforeByVp = beforeByVp; if (Object.keys(beforeByVp).length) node.beforeByVp = beforeByVp;
if (Object.keys(afterByVp).length) node.afterByVp = afterByVp; if (Object.keys(afterByVp).length) node.afterByVp = afterByVp;
if (Object.keys(placeholderByVp).length) node.placeholderByVp = placeholderByVp;
if (!raw.rawHTML) { if (!raw.rawHTML) {
const canonKids = elementChildren(raw); const canonKids = elementChildren(raw);
@@ -381,11 +389,16 @@ export function buildIR(sourceDir: string, viewports: number[], opts?: { motion?
keptChildren.push(c); keptChildren.push(c);
continue; continue;
} }
// `<source>` carries a <picture>/<video>'s media/art-direction candidates but never
// paints (0×0 at every viewport), so the visibility prune would always drop it —
// losing responsive variants (the lazy-loaded mobile img.src then serves every
// width). Keep it structurally; generation emits it only when its file materialized.
const keepSource = c.tag === "source" && (node.tag === "picture" || node.tag === "video");
if (keepAll) { if (keepAll) {
prune(c, true); prune(c, true);
keptChildren.push(c); keptChildren.push(c);
if (Object.values(c.visibleByVp).some(Boolean) || childHasVisible(c)) hasVisibleDescendant = true; if (Object.values(c.visibleByVp).some(Boolean) || childHasVisible(c)) hasVisibleDescendant = true;
} else if (prune(c, false)) { } else if (prune(c, false) || keepSource) {
keptChildren.push(c); keptChildren.push(c);
if (Object.values(c.visibleByVp).some(Boolean) || childHasVisible(c)) hasVisibleDescendant = true; if (Object.values(c.visibleByVp).some(Boolean) || childHasVisible(c)) hasVisibleDescendant = true;
} }
+1
View File
@@ -438,6 +438,7 @@ export function generateSiteApp(opts: {
const globals = tailwindGlobalsCss({ const globals = tailwindGlobalsCss({
reset: RESET_CSS, fontCss: unionFontCss(routes), tokensCss, htmlBg, bodyFont: SYSTEM_FALLBACK, clip, reset: RESET_CSS, fontCss: unionFontCss(routes), tokensCss, htmlBg, bodyFont: SYSTEM_FALLBACK, clip,
colorTokens: [...interner.tokens], viewports: entry.ir.doc.viewports, colorTokens: [...interner.tokens], viewports: entry.ir.doc.viewports,
canonical: entry.ir.doc.canonicalViewport,
}); });
writeText(join(appDir, "src", isVite ? "globals.css" : join("app", "globals.css")), isVite ? viteGlobalsCss(globals) : globals); writeText(join(appDir, "src", isVite ? "globals.css" : join("app", "globals.css")), isVite ? viteGlobalsCss(globals) : globals);
} else { } else {
+37
View File
@@ -0,0 +1,37 @@
import { describe, it } from "node:test";
import assert from "node:assert/strict";
import { isRetryableAssetFailure, ASSET_RETRY_DELAY_MS } from "../src/capture/capture.js";
describe("asset download retry decision", () => {
it("retries visual/font assets on network error (no response)", () => {
for (const type of ["image", "svg", "video", "font"]) {
assert.equal(isRetryableAssetFailure(type, null), true, `${type} + network error retries`);
}
});
it("retries visual/font assets on transient server states (5xx, 429)", () => {
assert.equal(isRetryableAssetFailure("image", 500), true);
assert.equal(isRetryableAssetFailure("image", 503), true);
assert.equal(isRetryableAssetFailure("video", 502), true);
assert.equal(isRetryableAssetFailure("font", 429), true);
});
it("does NOT retry authoritative 4xx failures (404/403/410)", () => {
assert.equal(isRetryableAssetFailure("image", 404), false);
assert.equal(isRetryableAssetFailure("image", 403), false);
assert.equal(isRetryableAssetFailure("video", 410), false);
assert.equal(isRetryableAssetFailure("font", 401), false);
});
it("does NOT retry non-visual asset types at all", () => {
assert.equal(isRetryableAssetFailure("css", null), false);
assert.equal(isRetryableAssetFailure("manifest", 500), false);
assert.equal(isRetryableAssetFailure("lottie", 503), false);
assert.equal(isRetryableAssetFailure("other", null), false);
});
it("uses a fixed, bounded retry delay (deterministic — no jitter)", () => {
assert.equal(typeof ASSET_RETRY_DELAY_MS, "number");
assert.ok(ASSET_RETRY_DELAY_MS > 0 && ASSET_RETRY_DELAY_MS <= 2000);
});
});
+102
View File
@@ -0,0 +1,102 @@
import { describe, it } from "node:test";
import assert from "node:assert/strict";
import type { IR, IRNode, IRChild, StyleMap, BBox } from "../src/normalize/ir.js";
import { generateCss } from "../src/generate/css.js";
const VPS = [375, 1280];
const CANONICAL = 1280;
/** Computed style with the minimum the emitter reads, per viewport. */
function computed(over: StyleMap = {}): StyleMap {
return { display: "block", position: "static", visibility: "visible", listStyleType: "disc", listStylePosition: "outside", ...over };
}
function node(id: string, tag: string, cs: StyleMap, children: IRChild[] = [], visible = true): IRNode {
const computedByVp: Record<number, StyleMap> = {};
const bboxByVp: Record<number, BBox> = {};
const visibleByVp: Record<number, boolean> = {};
for (const vp of VPS) {
computedByVp[vp] = { ...cs };
bboxByVp[vp] = { x: 0, y: 0, width: vp, height: 100 };
visibleByVp[vp] = visible;
}
return { id, tag, attrs: {}, visibleByVp, bboxByVp, computedByVp, children };
}
function irWith(root: IRNode): IR {
return {
doc: {
sourceUrl: "https://example.test/css",
title: "CSS Fixture",
lang: "en",
charset: "UTF-8",
metaViewport: "width=device-width, initial-scale=1",
viewports: VPS,
sampleViewports: VPS,
canonicalViewport: CANONICAL,
perViewport: Object.fromEntries(VPS.map((vp) => [vp, { scrollHeight: 800, scrollWidth: vp, htmlBg: "rgb(255, 255, 255)", bodyBg: "rgb(255, 255, 255)", bodyColor: "rgb(0, 0, 0)", bodyFont: "Arial" }])),
nodeCount: 4,
keyframes: [],
},
root,
};
}
/** The base-rule body for a selector (first non-banded `.c<id>{…}` block). */
function baseRule(css: string, id: string): string {
const m = css.match(new RegExp(`\\.c${id}\\{([^}]*)\\}`));
return m?.[1] ?? "";
}
describe("generateCss list markers", () => {
it("re-establishes list-style on a <ul> whose disc equals the parent's initial value", () => {
// list-style-type's initial value is `disc` on EVERY element, so a real <ul disc>
// equals its parent <div>'s computed value — but the reset (`ul, ol, menu
// { list-style: none; }`) breaks the inheritance chain, so it must still emit.
const li = node("n2", "li", computed({ display: "list-item" }));
const ul = node("n1", "ul", computed(), [li]);
const root = node("n0", "body", computed(), [ul]);
const css = generateCss(irWith(root), new Map());
assert.ok(baseRule(css, "n1").includes("list-style-type:disc"));
// The <li> inherits from the ul (not reset), so parent-equality elision still applies.
assert.ok(!baseRule(css, "n2").includes("list-style-type"));
});
it("does not emit list-style-type on non-list tags at the initial disc", () => {
const div = node("n1", "div", computed());
const root = node("n0", "body", computed(), [div]);
const css = generateCss(irWith(root), new Map());
assert.ok(!baseRule(css, "n1").includes("list-style-type"));
});
});
describe("generateCss visibility", () => {
it("emits visibility:hidden for a node hidden at the canonical viewport", () => {
const hidden = node("n1", "div", computed({ visibility: "hidden" }), [], false);
const shown = node("n2", "div", computed());
const root = node("n0", "body", computed(), [hidden, shown]);
const css = generateCss(irWith(root), new Map());
assert.ok(baseRule(css, "n1").includes("visibility:hidden"));
assert.ok(!baseRule(css, "n2").includes("visibility"));
});
it("restores inherited visibility at a band where the node is shown", () => {
const n = node("n1", "div", computed({ visibility: "hidden" }), [], false);
n.computedByVp[375]!.visibility = "visible";
n.visibleByVp[375] = true;
const root = node("n0", "body", computed(), [n]);
const css = generateCss(irWith(root), new Map());
assert.ok(baseRule(css, "n1").includes("visibility:hidden"));
const band = css.match(/@media \(max-width: \d+px\) \{\n([\s\S]*?)\n\}/);
assert.ok(band?.[1]?.includes("visibility:inherit"));
});
it("stays silent on the descendants of a hidden subtree (inheritance covers them)", () => {
const child = node("n2", "div", computed({ visibility: "hidden" }), [], false);
const parent = node("n1", "div", computed({ visibility: "hidden" }), [child], false);
const root = node("n0", "body", computed(), [parent]);
const css = generateCss(irWith(root), new Map());
assert.ok(baseRule(css, "n1").includes("visibility:hidden"));
assert.ok(!baseRule(css, "n2").includes("visibility"));
});
});
+255
View File
@@ -0,0 +1,255 @@
import { describe, it, before, after } from "node:test";
import assert from "node:assert/strict";
import { createServer, type Server } from "node:http";
import { mkdtempSync, readFileSync, rmSync } from "node:fs";
import { tmpdir } from "node:os";
import { join, dirname } from "node:path";
import { fileURLToPath } from "node:url";
import type { PageSnapshot, RawNode, RawChild } from "../src/capture/walker.js";
import {
planForFrameUrl, graftFrameIntoSnapshot, frameHasRenderableContent, findFrameNode,
} from "../src/capture/graft.js";
import { captureSite } from "../src/capture/capture.js";
import { buildIR, isTextChild, type IRNode } from "../src/normalize/ir.js";
import { resolveTag, propsList } from "../src/generate/app.js";
import { readJSON } from "../src/util/fsx.js";
const FIXTURES = join(dirname(fileURLToPath(import.meta.url)), "..", "fixtures");
// ---- Synthetic snapshot helpers (unit tests for the merge/offset logic) ----
function isText(c: RawChild): c is { text: string } {
return (c as { text?: string }).text !== undefined;
}
function raw(tag: string, attrs: Record<string, string> = {}, children: RawChild[] = [], over: Partial<RawNode> = {}): RawNode {
return {
tag, attrs,
computed: { display: "block", position: "static" },
bbox: { x: 0, y: 0, width: 100, height: 50 },
visible: true,
children,
...over,
};
}
function pageSnap(root: RawNode, url = "https://host.test/page"): PageSnapshot {
return {
doc: {
url, title: "t",
head: { description: "", canonical: "", ogTitle: "", ogDescription: "", ogImage: "", ogType: "", ogSiteName: "", twitterCard: "", themeColor: "" },
lang: "en", charset: "UTF-8", viewportWidth: 800, viewportHeight: 600,
scrollWidth: 800, scrollHeight: 600, htmlBg: "", bodyBg: "", bodyColor: "", bodyFont: "",
metaViewport: "", nodeCount: 3, truncated: false,
},
root, cssVars: {}, fontFaces: [], cssUrls: [], domAssets: [], keyframes: [],
};
}
describe("planForFrameUrl", () => {
it("skips blank/js frames and ad/analytics/captcha domains", () => {
assert.equal(planForFrameUrl(""), "skip");
assert.equal(planForFrameUrl("about:blank"), "skip");
assert.equal(planForFrameUrl("javascript:void(0)"), "skip");
assert.equal(planForFrameUrl("https://googleads.g.doubleclick.net/pagead/ads"), "skip");
assert.equal(planForFrameUrl("https://www.googletagmanager.com/ns.html?id=GTM-X"), "skip");
assert.equal(planForFrameUrl("https://www.google.com/recaptcha/api2/anchor"), "skip");
});
it("screenshots media-player embeds instead of grafting their JS-built DOM", () => {
assert.equal(planForFrameUrl("https://www.youtube.com/embed/abc123"), "still");
assert.equal(planForFrameUrl("https://player.vimeo.com/video/1"), "still");
assert.equal(planForFrameUrl("https://www.google.com/maps/embed?pb=x"), "still");
});
it("grafts everything else (form/newsletter embeds)", () => {
assert.equal(planForFrameUrl("https://static-forms.klaviyo.com/forms/abc"), "graft");
assert.equal(planForFrameUrl("http://127.0.0.1:4001/iframe-embed.html"), "graft");
});
});
describe("graftFrameIntoSnapshot", () => {
const mkHost = (): PageSnapshot =>
pageSnap(raw("body", {}, [
raw("iframe", { "data-ditto-frame": "0", src: "https://frame.test/embed" }, [], {
bbox: { x: 40, y: 30, width: 320, height: 160 },
}),
]));
const mkFrame = (): PageSnapshot => {
const input = raw("input", { id: "email", placeholder: "Enter your email" }, [], {
bbox: { x: 12, y: 34, width: 200, height: 30 },
});
const label = raw("label", { for: "email" }, [{ text: "Email" }]);
const anchor = raw("a", { href: "#terms" }, [{ text: "Terms" }]);
const img = raw("img", { src: "/pixel.png", srcset: "/pixel.png 1x, /pixel@2x.png 2x" });
const form = raw("form", { id: "signup" }, [label, input, anchor, img], {
bbox: { x: 0, y: 0, width: 320, height: 160 },
});
return pageSnap(raw("body", {}, [form]), "https://frame.test/embed");
};
it("grafts the frame body as a <div> child of the iframe with offset bboxes", () => {
const host = mkHost();
const ok = graftFrameIntoSnapshot(host, { idx: 0, contentX: 40, contentY: 30 }, mkFrame());
assert.equal(ok, true);
const iframe = findFrameNode(host.root, 0)!;
assert.equal(iframe.children.length, 1);
const wrapper = iframe.children[0] as RawNode;
assert.equal(wrapper.tag, "div", "the grafted <body> is retagged to <div>");
const form = wrapper.children[0] as RawNode;
assert.deepEqual([form.bbox.x, form.bbox.y], [40, 30], "frame-doc coords shift by the content-box origin");
const input = form.children.find((c) => !isText(c) && (c as RawNode).tag === "input") as RawNode;
assert.deepEqual([input.bbox.x, input.bbox.y], [52, 64]);
});
it("namespaces ids/for/#href so two frames cannot collide, and absolutizes frame-relative URLs", () => {
const host = mkHost();
graftFrameIntoSnapshot(host, { idx: 0, contentX: 40, contentY: 30 }, mkFrame());
const iframe = findFrameNode(host.root, 0)!;
const form = (iframe.children[0] as RawNode).children[0] as RawNode;
assert.equal(form.attrs.id, "f0-signup");
const byTag = (t: string): RawNode => form.children.find((c) => !isText(c) && (c as RawNode).tag === t) as RawNode;
assert.equal(byTag("input").attrs.id, "f0-email");
assert.equal(byTag("label").attrs.for, "f0-email");
assert.equal(byTag("a").attrs.href, "#f0-terms");
assert.equal(byTag("img").attrs.src, "https://frame.test/pixel.png");
assert.equal(byTag("img").attrs.srcset, "https://frame.test/pixel.png 1x, https://frame.test/pixel@2x.png 2x");
});
it("makes the iframe clip like a real frame viewport (overflow:hidden)", () => {
const host = mkHost();
graftFrameIntoSnapshot(host, { idx: 0, contentX: 40, contentY: 30 }, mkFrame());
const iframe = findFrameNode(host.root, 0)!;
assert.equal(iframe.computed.overflow, "hidden");
assert.equal(iframe.computed.overflowY, "hidden");
});
it("refuses to graft an empty/invisible frame (screenshot fallback territory)", () => {
const host = mkHost();
const empty = pageSnap(raw("body", {}, [raw("div", {}, [], { visible: false, bbox: { x: 0, y: 0, width: 0, height: 0 } })]), "https://frame.test/embed");
assert.equal(frameHasRenderableContent(empty.root), false);
assert.equal(graftFrameIntoSnapshot(host, { idx: 0, contentX: 40, contentY: 30 }, empty), false);
assert.equal(findFrameNode(host.root, 0)!.children.length, 0);
});
it("merges the frame's @keyframes into the page snapshot", () => {
const host = mkHost();
const frame = mkFrame();
frame.keyframes = ["@keyframes spin { to { transform: rotate(360deg); } }"];
graftFrameIntoSnapshot(host, { idx: 0, contentX: 40, contentY: 30 }, frame);
assert.ok(host.keyframes.includes("@keyframes spin { to { transform: rotate(360deg); } }"));
});
});
// ---- Cross-origin integration: capture → snapshot graft → IR → generated tag ----
describe("cross-origin iframe capture (integration)", () => {
let hostServer: Server;
let frameServer: Server;
let hostUrl = "";
let frameOrigin = "";
let outDir = "";
// A 1x1 transparent PNG so the frame's <img src="/pixel.png"> resolves and downloads.
const PIXEL = Buffer.from(
"iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNkYPhfDwAChwGA60e6kgAAAABJRU5ErkJggg==",
"base64",
);
before(async () => {
const frameHtml = readFileSync(join(FIXTURES, "iframe-embed.html"), "utf8");
frameServer = createServer((req, res) => {
if (req.url?.startsWith("/pixel.png")) {
res.writeHead(200, { "content-type": "image/png" });
res.end(PIXEL);
return;
}
res.writeHead(200, { "content-type": "text/html" });
res.end(frameHtml);
});
await new Promise<void>((r) => frameServer.listen(0, "127.0.0.1", r));
frameOrigin = `http://127.0.0.1:${(frameServer.address() as { port: number }).port}`;
const hostHtml = readFileSync(join(FIXTURES, "iframe-host.html"), "utf8").replaceAll("{{FRAME_ORIGIN}}", frameOrigin);
hostServer = createServer((_req, res) => {
res.writeHead(200, { "content-type": "text/html" });
res.end(hostHtml);
});
await new Promise<void>((r) => hostServer.listen(0, "127.0.0.1", r));
// localhost:portA vs localhost:portB are DIFFERENT origins — the in-page walker
// cannot see into the frame; only the Node-side graft can.
hostUrl = `http://127.0.0.1:${(hostServer.address() as { port: number }).port}/`;
outDir = mkdtempSync(join(tmpdir(), "ditto-iframe-graft-"));
});
after(async () => {
hostServer?.close();
frameServer?.close();
rmSync(outDir, { recursive: true, force: true });
});
it("grafts the cross-origin form into the snapshot, merges its assets, and generates a <div>", async () => {
const capture = await captureSite({
url: hostUrl,
outDir,
viewports: [800],
breakpoints: false,
screenshots: false,
});
// 1) The snapshot carries the grafted subtree under the iframe node.
const snap = readJSON<PageSnapshot>(join(outDir, "capture", "dom-800.json"));
const iframe = findFrameNode(snap.root, 0);
assert.ok(iframe, "the iframe was stamped and captured");
assert.equal(iframe!.children.length, 1, "the frame document was grafted");
const wrapper = iframe!.children[0] as RawNode;
assert.equal(wrapper.tag, "div");
const findIn = (n: RawNode, pred: (x: RawNode) => boolean): RawNode | null => {
if (pred(n)) return n;
for (const c of n.children) {
if (isText(c)) continue;
const hit = findIn(c, pred);
if (hit) return hit;
}
return null;
};
const input = findIn(wrapper, (n) => n.tag === "input")!;
assert.ok(input, "the email input is part of the graft");
assert.equal(input.attrs.id, "f0-email", "frame-internal ids are namespaced");
assert.equal(input.attrs.placeholder, "Enter your email");
assert.equal(input.placeholder?.color, "rgb(200, 150, 100)", "::placeholder captured inside the frame");
// The iframe sits at margin-left:40px — grafted geometry is in page coordinates.
assert.ok(input.bbox.x >= 40, `input.bbox.x (${input.bbox.x}) offset by the iframe origin`);
const label = findIn(wrapper, (n) => n.tag === "label")!;
assert.equal(label.attrs.for, "f0-email");
const button = findIn(wrapper, (n) => n.tag === "button")!;
assert.ok(button.visible, "the Sign up button is visible in the graft");
// 2) The frame's assets flow through the normal pipeline (absolute frame URL, downloaded).
const pixel = capture.assets.find((a) => a.url === `${frameOrigin}/pixel.png`);
assert.ok(pixel, "the frame-relative <img> was discovered under the frame's origin");
assert.ok(pixel!.storedAs, "and its bytes were stored");
// 3) IR keeps the grafted children as ordinary nodes; generation emits a <div>.
const ir = buildIR(outDir, [800]);
const findIr = (n: IRNode, pred: (x: IRNode) => boolean): IRNode | null => {
if (pred(n)) return n;
for (const c of n.children) {
if (isTextChild(c)) continue;
const hit = findIr(c, pred);
if (hit) return hit;
}
return null;
};
const irIframe = findIr(ir.root, (n) => n.tag === "iframe")!;
assert.ok(irIframe, "iframe survives into the IR");
assert.ok(irIframe.children.some((c) => !isTextChild(c)), "grafted children survive the IR prune");
assert.equal(resolveTag(irIframe, false), "div", "a grafted iframe renders as a positioned <div>");
const props = new Map(propsList(irIframe, new Map(), hostUrl));
assert.equal(props.get("src"), undefined, "document-loading attrs stay dropped");
const irInput = findIr(ir.root, (n) => n.tag === "input")!;
assert.ok(irInput.placeholderByVp?.[800], "placeholder style flows into the IR");
});
});
+100
View File
@@ -0,0 +1,100 @@
import { describe, it } from "node:test";
import assert from "node:assert/strict";
import { mkdtempSync, mkdirSync, writeFileSync } from "node:fs";
import { tmpdir } from "node:os";
import { join } from "node:path";
import type { RawNode, RawChild } from "../src/capture/walker.js";
import { buildIR, isTextChild, type IRNode } from "../src/normalize/ir.js";
const VPS = [375, 1280];
function raw(tag: string, attrs: Record<string, string> = {}, children: RawChild[] = [], visible = true): RawNode {
return {
tag, attrs,
computed: { display: visible ? "block" : "none", position: "static", visibility: "visible" },
bbox: { x: 0, y: 0, width: visible ? 640 : 0, height: visible ? 360 : 0 },
visible,
children,
};
}
/** Minimal PageSnapshot JSON around a body tree (same tree at every viewport). */
function snapshot(vp: number, root: RawNode): object {
return {
doc: {
url: "https://example.test/page", title: "Fixture",
head: { description: "", canonical: "", ogTitle: "", ogDescription: "", ogImage: "", ogType: "", ogSiteName: "", twitterCard: "", themeColor: "" },
lang: "en", charset: "UTF-8", viewportWidth: vp, viewportHeight: 800,
scrollWidth: vp, scrollHeight: 800, htmlBg: "rgb(255, 255, 255)", bodyBg: "rgb(255, 255, 255)",
bodyColor: "rgb(0, 0, 0)", bodyFont: "Arial", metaViewport: "width=device-width, initial-scale=1",
nodeCount: 10, truncated: false,
},
root, cssVars: {}, fontFaces: [], cssUrls: [], domAssets: [], keyframes: [],
};
}
function buildFixtureIR(root: RawNode): IRNode {
const sourceDir = mkdtempSync(join(tmpdir(), "ditto-ir-prune-"));
mkdirSync(join(sourceDir, "capture"), { recursive: true });
for (const vp of VPS) {
writeFileSync(join(sourceDir, "capture", `dom-${vp}.json`), JSON.stringify(snapshot(vp, structuredClone(root))));
}
return buildIR(sourceDir, VPS).root;
}
function findByTag(node: IRNode, tag: string): IRNode | null {
if (node.tag === tag) return node;
for (const c of node.children) {
if (isTextChild(c)) continue;
const hit = findByTag(c, tag);
if (hit) return hit;
}
return null;
}
describe("IR prune keeps <source> media candidates", () => {
it("keeps invisible source children of <picture> and <video>, still pruning other invisibles", () => {
// <source> is 0×0 at EVERY viewport (never painted), which the visibility prune
// reads as unobserved — real case (ooni.com): 47 <source> in dom-1280.json, 0 in
// ir.json, so the mobile img.src got baked into desktop layouts.
const picture = raw("picture", {}, [
raw("source", { srcset: "/img/hero-1280.jpg 1x", media: "(min-width: 768px)" }, [], false),
raw("img", { src: "/img/hero-375.jpg", alt: "hero" }),
]);
const video = raw("video", { autoplay: "" }, [
raw("source", { src: "/media/hero.mp4", type: "video/mp4" }, [], false),
]);
const noise = raw("div", {}, [raw("span", {}, [], false)], false); // invisible everywhere → pruned
const body = raw("body", {}, [picture, video, noise]);
const root = buildFixtureIR(body);
const pic = findByTag(root, "picture");
assert.ok(pic, "picture survives");
const picTags = pic!.children.filter((c) => !isTextChild(c)).map((c) => (c as IRNode).tag);
assert.deepEqual(picTags, ["source", "img"]);
const src = findByTag(pic!, "source")!;
assert.equal(src.attrs.srcset, "/img/hero-1280.jpg 1x");
assert.equal(src.attrs.media, "(min-width: 768px)");
const vid = findByTag(root, "video");
assert.ok(vid, "video survives");
const vidTags = vid!.children.filter((c) => !isTextChild(c)).map((c) => (c as IRNode).tag);
assert.deepEqual(vidTags, ["source"]);
// The carve-out is scoped: unrelated invisible-everywhere subtrees still prune.
assert.equal(findByTag(root, "span"), null);
});
it("does not keep a <source> outside picture/video, nor resurrect a pruned picture", () => {
const orphan = raw("div", {}, [raw("source", { srcset: "/x.jpg" }, [], false)]);
// A picture invisible everywhere with no visible descendants is still pruned whole.
const ghost = raw("picture", {}, [raw("source", { srcset: "/y.jpg" }, [], false)], false);
const body = raw("body", {}, [orphan, ghost]);
const root = buildFixtureIR(body);
assert.equal(findByTag(root, "source"), null);
assert.equal(findByTag(root, "picture"), null);
});
});
+133
View File
@@ -0,0 +1,133 @@
import { describe, it } from "node:test";
import assert from "node:assert/strict";
import type { IRNode, IRChild, StyleMap, BBox } from "../src/normalize/ir.js";
import { propsList, renderChildrenJsx } from "../src/generate/app.js";
const VPS = [375, 1280];
const SOURCE = "https://example.test/page";
const GIF = "data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7";
function el(id: string, tag: string, attrs: Record<string, string> = {}, children: IRChild[] = [], visible = true): IRNode {
const computedByVp: Record<number, StyleMap> = {};
const bboxByVp: Record<number, BBox> = {};
const visibleByVp: Record<number, boolean> = {};
for (const vp of VPS) {
computedByVp[vp] = { display: "block", position: "static", visibility: "visible" };
bboxByVp[vp] = { x: 0, y: 0, width: visible ? 640 : 0, height: visible ? 360 : 0 };
visibleByVp[vp] = visible;
}
return { id, tag, attrs, visibleByVp, bboxByVp, computedByVp, children };
}
function props(node: IRNode, assetMap: Map<string, string>): Map<string, string> {
return new Map(propsList(node, assetMap, SOURCE));
}
describe("video src emission", () => {
it("ships the local file and mirrors captured playback attrs when the src materialized", () => {
const assetMap = new Map([
["https://example.test/media/hero.mp4", "/assets/cloned/videos/ab.mp4"],
["https://example.test/img/poster.jpg", "/assets/cloned/images/cd.jpg"],
]);
const video = el("n1", "video", {
src: "/media/hero.mp4", poster: "/img/poster.jpg",
autoplay: "", loop: "", muted: "", playsinline: "",
});
const p = props(video, assetMap);
assert.equal(p.get("src"), JSON.stringify("/assets/cloned/videos/ab.mp4"));
assert.equal(p.get("poster"), JSON.stringify("/assets/cloned/images/cd.jpg"));
assert.equal(p.get("autoPlay"), "true");
assert.equal(p.get("loop"), "true");
assert.equal(p.get("muted"), "true");
assert.equal(p.get("playsInline"), "true");
// preload is mirrored, not forced: none captured → none emitted.
assert.equal(p.get("preload"), undefined);
});
it("falls back to poster-only when no video file materialized", () => {
const assetMap = new Map([
["https://example.test/img/poster.jpg", "/assets/cloned/images/cd.jpg"],
]);
const video = el("n1", "video", { src: "/media/hero.mp4", poster: "/img/poster.jpg", autoplay: "", loop: "" });
const p = props(video, assetMap);
assert.equal(p.get("src"), undefined);
assert.equal(p.get("autoPlay"), undefined);
assert.equal(p.get("loop"), undefined);
assert.equal(p.get("poster"), JSON.stringify("/assets/cloned/images/cd.jpg"));
assert.equal(p.get("preload"), JSON.stringify("none"));
});
it("treats a materialized child <source> as a shippable video (src carried by the child)", () => {
const assetMap = new Map([["https://example.test/media/hero.webm", "/assets/cloned/videos/ef.webm"]]);
const source = el("n2", "source", { src: "/media/hero.webm", type: "video/webm" }, [], false);
const video = el("n1", "video", { autoplay: "", muted: "" }, [source]);
const p = props(video, assetMap);
assert.equal(p.get("autoPlay"), "true");
assert.equal(p.get("preload"), undefined);
const jsx = renderChildrenJsx([video], assetMap, SOURCE, 0);
assert.match(jsx, /<source[^>]*src="\/assets\/cloned\/videos\/ef\.webm"[^>]*\/>/);
assert.match(jsx, /type="video\/webm"/);
});
it("drops non-materialized <source> children (poster-only video keeps none)", () => {
const source = el("n2", "source", { src: "/media/hero.webm", type: "video/webm" }, [], false);
const video = el("n1", "video", { autoplay: "" }, [source]);
const jsx = renderChildrenJsx([video], new Map(), SOURCE, 0);
assert.ok(!jsx.includes("<source"));
assert.ok(!jsx.includes("autoPlay"));
});
});
describe("poster fallback policy", () => {
it("DROPS a poster that missed the asset map instead of substituting the transparent GIF", () => {
const video = el("n1", "video", { poster: "https://clone-still.local/0-abc.jpg" });
const p = props(video, new Map());
assert.equal(p.get("poster"), undefined);
assert.ok(![...p.values()].includes(JSON.stringify(GIF)));
});
it("keeps the transparent-GIF fallback for a missed <img> src", () => {
const img = el("n1", "img", { src: "/img/gone.png", alt: "" });
const p = props(img, new Map());
assert.equal(p.get("src"), JSON.stringify(GIF));
});
});
describe("picture>source rewriting", () => {
const mkPicture = () => {
const desktop = el("n2", "source", {
srcset: "/img/hero-1280.jpg 1x, /img/hero-2560.jpg 2x",
media: "(min-width: 768px)", type: "image/jpeg", sizes: "100vw",
}, [], false);
const gone = el("n3", "source", { srcset: "/img/never-downloaded.avif", media: "(min-width: 1024px)" }, [], false);
const img = el("n4", "img", { src: "/img/hero-375.jpg", alt: "hero" });
return el("n1", "picture", {}, [desktop, gone, img]);
};
it("emits surviving srcset candidates with media/type/sizes preserved, omits sources with none", () => {
const assetMap = new Map([
["https://example.test/img/hero-1280.jpg", "/assets/cloned/images/a.jpg"],
["https://example.test/img/hero-375.jpg", "/assets/cloned/images/m.jpg"],
]);
const jsx = renderChildrenJsx([mkPicture()], assetMap, SOURCE, 0);
// Desktop source survives with ONLY the materialized candidate; void element.
assert.match(jsx, /<source[^>]*srcSet="\/assets\/cloned\/images\/a\.jpg 1x"[^>]*\/>/);
assert.ok(!jsx.includes("hero-2560"));
assert.match(jsx, /media="\(min-width: 768px\)"/);
assert.match(jsx, /type="image\/jpeg"/);
assert.match(jsx, /sizes="100vw"/);
// The fully-missed source is omitted entirely (no placeholder-pointing variant).
assert.ok(!jsx.includes("never-downloaded"));
assert.ok(!jsx.includes("(min-width: 1024px)"));
// The <img> fallback keeps its captured (rewritten) src.
assert.match(jsx, /<img[^>]*src="\/assets\/cloned\/images\/m\.jpg"/);
// <source> stays a void element: no children, no closing tag.
assert.ok(!jsx.includes("</source>"));
});
it("omits every source when nothing materialized, keeping the img fallback", () => {
const jsx = renderChildrenJsx([mkPicture()], new Map(), SOURCE, 0);
assert.ok(!jsx.includes("<source"));
assert.match(jsx, new RegExp(`<img[^>]*src="${GIF.replace(/[+/]/g, (c) => "\\" + c)}"`));
});
});
+164
View File
@@ -0,0 +1,164 @@
import { describe, it, before, after } from "node:test";
import assert from "node:assert/strict";
import { readFileSync } from "node:fs";
import { join, dirname } from "node:path";
import { fileURLToPath } from "node:url";
import { chromium, type Browser, type Page } from "playwright";
import { collectPage, type RawNode, type RawChild } from "../src/capture/walker.js";
import type { IR, IRNode, IRChild, StyleMap, BBox } from "../src/normalize/ir.js";
import { generateCss } from "../src/generate/css.js";
import { buildTailwind } from "../src/generate/tailwind.js";
const FIXTURES = join(dirname(fileURLToPath(import.meta.url)), "..", "fixtures");
function isText(c: RawChild): c is { text: string } {
return (c as { text?: string }).text !== undefined;
}
function findBy(root: RawNode, pred: (n: RawNode) => boolean): RawNode | null {
if (pred(root)) return root;
for (const c of root.children) {
if (isText(c)) continue;
const hit = findBy(c, pred);
if (hit) return hit;
}
return null;
}
describe("walker ::placeholder capture", () => {
let browser: Browser;
let page: Page;
before(async () => {
browser = await chromium.launch();
page = await browser.newPage();
});
after(async () => {
await browser.close();
});
it("captures the styled ::placeholder color/font of input and textarea", async () => {
await page.setContent(readFileSync(join(FIXTURES, "placeholder.html"), "utf8"));
await page.evaluate("globalThis.__name = globalThis.__name || ((fn) => fn);");
const snap = await page.evaluate(collectPage);
const input = findBy(snap.root, (n) => n.tag === "input" && n.attrs.class === "styled")!;
assert.ok(input.placeholder, "styled input carries a placeholder style");
assert.equal(input.placeholder!.color, "rgb(120, 30, 200)");
assert.equal(input.placeholder!.fontSize, "14px");
const textarea = findBy(snap.root, (n) => n.tag === "textarea")!;
assert.ok(textarea.placeholder, "styled textarea carries a placeholder style");
assert.equal(textarea.placeholder!.color, "rgb(5, 100, 5)");
});
it("does not attach placeholder style to a control without placeholder text", async () => {
await page.setContent(readFileSync(join(FIXTURES, "placeholder.html"), "utf8"));
await page.evaluate("globalThis.__name = globalThis.__name || ((fn) => fn);");
const snap = await page.evaluate(collectPage);
const plain = findBy(snap.root, (n) => n.tag === "input" && n.attrs.class === "plain")!;
assert.equal(plain.placeholder, undefined);
});
});
// ---- Generated CSS ----
const VPS = [375, 1280];
const CANONICAL = 1280;
function computed(over: StyleMap = {}): StyleMap {
return { display: "block", position: "static", visibility: "visible", ...over };
}
function node(id: string, tag: string, cs: StyleMap, children: IRChild[] = []): IRNode {
const computedByVp: Record<number, StyleMap> = {};
const bboxByVp: Record<number, BBox> = {};
const visibleByVp: Record<number, boolean> = {};
for (const vp of VPS) {
computedByVp[vp] = { ...cs };
bboxByVp[vp] = { x: 0, y: 0, width: 200, height: 40 };
visibleByVp[vp] = true;
}
return { id, tag, attrs: {}, visibleByVp, bboxByVp, computedByVp, children };
}
function irWith(root: IRNode): IR {
return {
doc: {
sourceUrl: "https://example.test/placeholder",
title: "Placeholder Fixture",
lang: "en",
charset: "UTF-8",
metaViewport: "width=device-width, initial-scale=1",
viewports: VPS,
sampleViewports: VPS,
canonicalViewport: CANONICAL,
perViewport: Object.fromEntries(VPS.map((vp) => [vp, { scrollHeight: 800, scrollWidth: vp, htmlBg: "rgb(255, 255, 255)", bodyBg: "rgb(255, 255, 255)", bodyColor: "rgb(0, 0, 0)", bodyFont: "Arial" }])),
nodeCount: 2,
keyframes: [],
},
root,
};
}
describe("generateCss ::placeholder emission", () => {
it("emits a ::placeholder rule with the captured color", () => {
const input = node("n1", "input", computed({ color: "rgb(10, 20, 30)", fontSize: "16px" }));
input.placeholderByVp = {
375: { color: "rgb(120, 30, 200)", fontSize: "14px" },
1280: { color: "rgb(120, 30, 200)", fontSize: "14px" },
};
const root = node("n0", "body", computed(), [input]);
const css = generateCss(irWith(root), new Map());
const m = css.match(/\.cn1::placeholder\{([^}]*)\}/);
assert.ok(m, "a .cn1::placeholder rule is emitted");
assert.ok(m![1]!.includes("color:rgb(120, 30, 200)"));
// fontSize differs from the host's own 16px, so it must be emitted too.
assert.ok(m![1]!.includes("font-size:14px"));
});
it("omits font props the placeholder inherits from the input itself", () => {
const input = node("n1", "input", computed({ color: "rgb(10, 20, 30)", fontSize: "16px", fontFamily: "Arial" }));
input.placeholderByVp = {
375: { color: "rgb(200, 150, 100)", fontSize: "16px", fontFamily: "Arial" },
1280: { color: "rgb(200, 150, 100)", fontSize: "16px", fontFamily: "Arial" },
};
const root = node("n0", "body", computed(), [input]);
const css = generateCss(irWith(root), new Map());
const m = css.match(/\.cn1::placeholder\{([^}]*)\}/);
assert.ok(m);
assert.ok(m![1]!.includes("color:rgb(200, 150, 100)"));
assert.ok(!m![1]!.includes("font-size"), "inherited font-size is not re-declared");
assert.ok(!m![1]!.includes("font-family"), "inherited font-family is not re-declared");
});
it("bands a placeholder color that changes across viewports", () => {
const input = node("n1", "input", computed({ color: "rgb(10, 20, 30)" }));
input.placeholderByVp = {
375: { color: "rgb(1, 2, 3)" },
1280: { color: "rgb(120, 30, 200)" },
};
const root = node("n0", "body", computed(), [input]);
const css = generateCss(irWith(root), new Map());
assert.ok(/\.cn1::placeholder\{[^}]*rgb\(120, 30, 200\)/.test(css), "base carries the canonical color");
assert.ok(/@media \(max-width: \d+px\)[\s\S]*\.cn1::placeholder\{[^}]*rgb\(1, 2, 3\)/.test(css), "mobile band overrides the color");
});
it("emits no ::placeholder rule for nodes without placeholder styles", () => {
const input = node("n1", "input", computed({ color: "rgb(10, 20, 30)" }));
const root = node("n0", "body", computed(), [input]);
const css = generateCss(irWith(root), new Map());
assert.ok(!css.includes("::placeholder"));
});
it("tailwind mode (the default pipeline) carries the rule in ditto.css keyed by data-cid", () => {
const input = node("n1", "input", computed({ color: "rgb(10, 20, 30)" }));
input.placeholderByVp = {
375: { color: "rgb(120, 30, 200)" },
1280: { color: "rgb(120, 30, 200)" },
};
const root = node("n0", "body", computed(), [input]);
const tw = buildTailwind(irWith(root), new Map());
assert.ok(/\[data-cid="n1"\]::placeholder\s*\{[^}]*color:\s*rgb\(120, 30, 200\)/.test(tw.pseudoCss),
`pseudoCss carries the placeholder rule:\n${tw.pseudoCss}`);
});
});
+64
View File
@@ -0,0 +1,64 @@
import { describe, it } from "node:test";
import assert from "node:assert/strict";
import { computeBands } from "../src/generate/css.js";
import { bandScreens, prefixFor, tailwindGlobalsCss } from "../src/generate/tailwind.js";
/** Interval covered by a ditto.css band media query (integer px, inclusive). */
function bandInterval(media: string | null): [number, number] {
if (!media) return [0, Infinity];
const min = /min-width:\s*(\d+)px/.exec(media);
const max = /max-width:\s*(\d+)px/.exec(media);
return [min ? +min[1]! : 0, max ? +max[1]! : Infinity];
}
/** Integer interval covered by a Tailwind variant prefix under the given screens.
* v4 semantics for named AND arbitrary variants: `X:` = width >= X; `max-X:` = width < X. */
function prefixInterval(prefix: string, screens: Map<string, number>): [number, number] {
let lo = 0, hi = Infinity;
for (const m of prefix.matchAll(/(max-)?(?:(sm|md|lg|xl|2xl)|(?:min-)?\[(\d+)px\]):/g)) {
const px = m[3] ? +m[3] : screens.get(m[2]!);
assert.ok(px !== undefined, `screen defined for ${m[0]}`);
if (m[1]) hi = Math.min(hi, px! - 1);
else lo = Math.max(lo, px!);
}
return [lo, hi];
}
function assertAgreement(viewports: number[], canonical: number): void {
const screens = new Map(bandScreens(viewports, canonical));
for (const b of computeBands(viewports, canonical)) {
if (!b.media) continue;
assert.deepEqual(
prefixInterval(prefixFor(b.media), screens),
bandInterval(b.media),
`band vp${b.vp} (${b.media}) matches its Tailwind prefix`,
);
}
}
describe("tailwind screens agree with computeBands boundaries", () => {
it("standard 375/768/1280/1920 ladder: md/lg/2xl pinned to the band midpoint boundaries", () => {
const screens = new Map(bandScreens([375, 768, 1280, 1920], 1280));
// Midpoints 571/1024/1600 → bands ≤571, 5721024, base, ≥1601.
assert.deepEqual([...screens], [["md", 572], ["lg", 1025], ["2xl", 1601]]);
// NOT Tailwind's stock 768/1024/1536 — stock 2xl (1536) would flip utility-classed
// nodes to the 1920 layout in 15361600 while ditto.css still holds the 1280 layout.
assert.notEqual(screens.get("2xl"), 1536);
assertAgreement([375, 768, 1280, 1920], 1280);
});
it("odd viewport set: bands fall back to exact arbitrary prefixes (no named screens needed)", () => {
assert.deepEqual(bandScreens([320, 900, 1440], 900), []);
assertAgreement([320, 900, 1440], 900);
});
it("emits the derived screens into the generated @theme", () => {
const css = tailwindGlobalsCss({
reset: "", fontCss: "", tokensCss: "", htmlBg: "#fff", bodyFont: "sans-serif",
clip: "", colorTokens: [], viewports: [375, 768, 1280, 1920], canonical: 1280,
});
assert.ok(css.includes("--breakpoint-md: 572px;"));
assert.ok(css.includes("--breakpoint-lg: 1025px;"));
assert.ok(css.includes("--breakpoint-2xl: 1601px;"));
});
});
+25
View File
@@ -16,6 +16,7 @@ import {
seoRouteFiles, seoRouteFiles,
} from "../src/generate/seo.js"; } from "../src/generate/seo.js";
import { agentsMd, architectureMd } from "../src/generate/docs.js"; import { agentsMd, architectureMd } from "../src/generate/docs.js";
import { NEXT_CONFIG } from "../src/generate/app.js";
function fixtureIr(): IR { function fixtureIr(): IR {
return { return {
@@ -169,4 +170,28 @@ describe("SEO inventory and emission", () => {
assert.ok(agentsMd(docsInput).includes("src/app/ditto")); assert.ok(agentsMd(docsInput).includes("src/app/ditto"));
assert.ok(architectureMd(docsInput).includes("data-ditto-id")); assert.ok(architectureMd(docsInput).includes("data-ditto-id"));
}); });
it("passes a Next-supported og:type through to openGraph", () => {
const ir = fixtureIr();
ir.doc.head!.meta!.push({ property: "og:type", content: "article" });
const report = buildSeoInventory(ir, fixtureAssets(), fixtureCapture());
const metadata = metadataExport(report);
assert.ok(metadata.includes('"type": "article"'));
assert.ok(!metadata.includes('"og:type"'));
});
it("routes an unsupported og:type into metadata.other (Next throws on unknown enum values)", () => {
const ir = fixtureIr();
ir.doc.head!.meta!.push({ property: "og:type", content: "product.group" });
const report = buildSeoInventory(ir, fixtureAssets(), fixtureCapture());
const metadata = metadataExport(report);
assert.ok(metadata.includes('"og:type": "product.group"'));
assert.ok(!metadata.includes('"type": "product.group"'));
});
});
describe("generated Next config", () => {
it("disables the dev-tools badge so it cannot leak into screenshots", () => {
assert.ok(NEXT_CONFIG.includes("devIndicators: false"));
});
}); });
+166
View File
@@ -0,0 +1,166 @@
import { describe, it, before, after } from "node:test";
import assert from "node:assert/strict";
import { join, dirname } from "node:path";
import { fileURLToPath, pathToFileURL } from "node:url";
import { chromium, type Browser, type Page } from "playwright";
import {
promoteLazyMediaInPage,
settleCarouselsInPage,
forceRevealForShot,
restoreRevealForShot,
} from "../src/capture/stabilize.js";
const FIXTURES = join(dirname(fileURLToPath(import.meta.url)), "..", "fixtures");
// tsx/esbuild wraps functions with a __name() helper for stack traces; the serialized
// page functions carry those calls, so shim it (same as capture.ts's init script).
const ESBUILD_SHIM = "globalThis.__name = globalThis.__name || ((fn) => fn);";
describe("stabilize: lazy-media promotion", () => {
let browser: Browser;
let page: Page;
before(async () => {
browser = await chromium.launch();
page = await browser.newPage();
await page.addInitScript(ESBUILD_SHIM);
await page.goto(pathToFileURL(join(FIXTURES, "lazy.html")).href);
});
after(async () => {
await browser.close();
});
it("promotes data-lazy-src/data-src/data-srcset/data-bg and measures the loaded size", async () => {
const count = await page.evaluate(promoteLazyMediaInPage);
// lazy1 (src), lazy2 (src), lazysource (srcset), lazy3 (src), lazybg (background)
assert.equal(count, 5);
const state = await page.evaluate(() => {
const attr = (id: string, name: string) => document.getElementById(id)?.getAttribute(name) ?? null;
const img = (id: string) => document.getElementById(id) as HTMLImageElement;
return {
lazy1Src: attr("lazy1", "src"),
lazy1Loading: attr("lazy1", "loading"),
lazy1Width: img("lazy1").naturalWidth,
lazy2Src: attr("lazy2", "src"),
lazy2Sizes: attr("lazy2", "sizes"), // data-sizes="auto" is a flag, not a value
sourceSrcset: attr("lazysource", "srcset"),
lazy3Src: attr("lazy3", "src"),
bg: (document.getElementById("lazybg") as HTMLElement).style.backgroundImage,
notaurlSrc: attr("notaurl", "src"),
};
});
assert.equal(state.lazy1Src, "og-image.png");
assert.equal(state.lazy1Loading, "eager");
assert.ok(state.lazy1Width > 1, "promoted image decoded to its real size");
assert.equal(state.lazy2Src, "seo-icon.png");
assert.equal(state.lazy2Sizes, null);
assert.equal(state.sourceSrcset, "og-image.png 1x");
assert.equal(state.lazy3Src, "og-image.png");
assert.ok(state.bg.includes("brand.svg"), `background promoted (got ${state.bg})`);
assert.equal(state.notaurlSrc, null, "non-URL data attr must not be promoted");
});
it("is idempotent and never clobbers an src that already equals the target", async () => {
const again = await page.evaluate(promoteLazyMediaInPage);
assert.equal(again, 0);
const alreadySrc = await page.evaluate(() => document.getElementById("already")?.getAttribute("src"));
assert.equal(alreadySrc, "seo-icon.png");
});
});
describe("stabilize: carousel settling", () => {
let browser: Browser;
let page: Page;
before(async () => {
browser = await chromium.launch();
page = await browser.newPage();
await page.addInitScript(ESBUILD_SHIM);
await page.goto(pathToFileURL(join(FIXTURES, "carousel-autoplay.html")).href);
});
after(async () => {
await browser.close();
});
const txOf = (sel: string) =>
page.evaluate((s) => {
const el = document.querySelector(s)!;
return new DOMMatrixReadOnly(getComputedStyle(el).transform).m41;
}, sel);
it("returns an autoplaying carousel to its home slide and pauses autoplay", async () => {
// let autoplay advance at least one slide so there is something to settle
await page.waitForFunction("window.__autoplayTicks >= 2", null, { timeout: 5000 });
assert.notEqual(await txOf(".splide__list"), 0);
const res = await page.evaluate(settleCarouselsInPage);
assert.equal(res.roots, 2);
assert.equal(res.normalized, 2);
assert.equal(await txOf(".splide__list"), 0, "track back at the home slide");
const bulletActive = await page.evaluate(() =>
document.querySelector(".splide__pagination__bullet")!.classList.contains("is-active"));
assert.ok(bulletActive, "first bullet re-activated");
// paused: no further autoplay ticks, and the track holds its home transform
const ticks = await page.evaluate("window.__autoplayTicks");
await page.waitForTimeout(700); // > 2 autoplay intervals
assert.equal(await page.evaluate("window.__autoplayTicks"), ticks, "autoplay latched paused");
assert.equal(await txOf(".splide__list"), 0, "track still at home after the autoplay window");
});
it("pins a control-less non-loop track left mid-offset back to translateX(0)", async () => {
assert.equal(await txOf("#stuck .swiper-wrapper"), 0);
});
it("cancels a mid-flight track transition instead of freezing it", async () => {
// Start a slide transition and settle while it is still running: pausing the
// CSSTransition would disassociate it from style and HOLD the frozen mid-flight
// transform over the home navigation — the settled track must still land at 0.
await page.evaluate(`(async () => {
const bullets = document.querySelectorAll(".splide__pagination__bullet");
bullets[2].click(); // transition toward slide 3 begins (0.2s)
return await (${settleCarouselsInPage.toString()})();
})()`);
assert.equal(await txOf(".splide__list"), 0, "track settled at home, not a frozen mid-flight offset");
});
});
describe("stabilize: force-reveal for element screenshots", () => {
let browser: Browser;
let page: Page;
before(async () => {
browser = await chromium.launch();
page = await browser.newPage();
await page.addInitScript(ESBUILD_SHIM);
await page.setContent(
'<div id="wrap" style="visibility:hidden">' +
'<video id="v" style="width:320px;height:180px;background:#000"></video>' +
"</div>",
);
});
after(async () => {
await browser.close();
});
it("reveals the hidden ancestor chain so the screenshot succeeds, then restores exactly", async () => {
const vis = () => page.evaluate(() => getComputedStyle(document.getElementById("v")!).visibility);
assert.equal(await vis(), "hidden");
const forced = await page.evaluate(forceRevealForShot, "#v");
assert.equal(forced, 2); // the video (inherited hidden) + the wrapping div
assert.equal(await vis(), "visible");
const buf = await page.locator("#v").screenshot({ type: "jpeg", quality: 82, timeout: 2000, animations: "disabled" });
assert.ok(buf.length > 0, "screenshot captured while revealed");
await page.evaluate(restoreRevealForShot);
assert.equal(await vis(), "hidden");
const after = await page.evaluate(() => ({
wrapInline: (document.getElementById("wrap") as HTMLElement).style.visibility,
videoInline: (document.getElementById("v") as HTMLElement).style.visibility,
markers: document.querySelectorAll("[data-clone-vis-restore]").length,
}));
assert.equal(after.wrapInline, "hidden", "original inline value restored");
assert.equal(after.videoInline, "", "no inline visibility left on the video");
assert.equal(after.markers, 0, "restore markers removed");
});
});
+66
View File
@@ -0,0 +1,66 @@
import { describe, it, before, after } from "node:test";
import assert from "node:assert/strict";
import { chromium, type Browser, type Page } from "playwright";
import { collectPage, type RawNode, type RawChild } from "../src/capture/walker.js";
function isText(c: RawChild): c is { text: string } {
return (c as { text?: string }).text !== undefined;
}
function findByTag(root: RawNode, tag: string): RawNode | null {
if (root.tag === tag) return root;
for (const c of root.children) {
if (isText(c)) continue;
const hit = findByTag(c, tag);
if (hit) return hit;
}
return null;
}
function textRun(node: RawNode): string {
let out = "";
for (const c of node.children) out += isText(c) ? c.text : textRun(c);
return out;
}
describe("walker whitespace-only text nodes", () => {
let browser: Browser;
let page: Page;
before(async () => {
browser = await chromium.launch();
page = await browser.newPage();
});
after(async () => {
await browser.close();
});
const capture = async (html: string) => {
await page.setContent(html);
// tsx/esbuild wraps functions with a __name() helper for stack traces; the
// serialized collectPage carries those calls, so shim it (same as capture.ts).
await page.evaluate("globalThis.__name = globalThis.__name || ((fn) => fn);");
return page.evaluate(collectPage);
};
it("keeps the lone space that is the only child of an inline element", async () => {
// Real case (ooni.com): the space between "of" and "the" lives alone inside
// <strong>; dropping it fuses the adjacent text runs ("ofthe").
const snap = await capture("<p>Creator of<strong> </strong><em><strong>the world's</strong></em></p>");
const p = findByTag(snap.root, "p")!;
const strong = p.children.find((c) => !isText(c) && c.tag === "strong") as RawNode;
assert.deepEqual(strong.children, [{ text: " " }]);
assert.equal(textRun(p), "Creator of the world's");
});
it("still keeps the single space between inline elements", async () => {
const snap = await capture("<p><em>a</em> <em>b</em></p>");
const p = findByTag(snap.root, "p")!;
assert.equal(textRun(p), "a b");
});
it("does not emit a space inside an empty block container", async () => {
const snap = await capture("<main><section>\n \n</section></main>");
const section = findByTag(snap.root, "section")!;
assert.deepEqual(section.children, []);
});
});