webcrawl 1.4.3 → 1.4.4

package/dist/core/local-engine.js

@@ -9,7 +9,7 @@
  * Every community improvement to extractors, parsers, or targets in `src/` immediately
  * powers both this local engine (`npx webcrawl`) and the hosted `webcrawl.app` API.
  */
-import { extractWebsiteData, fastHtmlToMarkdown, fetchWithTlsImpersonation, getCachedPageAsync, isChallengeMarkdown, isLocalChromeAvailable, mapWebsiteUrls, renderViaLocalChrome, setCachedPage, sha256, } from "./engine.js";
+import { extractWebsiteData, fastHtmlToMarkdown, fetchWithTlsImpersonation, getCachedPageAsync, isChallengeMarkdown, isLocalChromeAvailable, mapWebsiteUrls, renderViaLocalChromeDetailed, setCachedPage, sha256, } from "./engine.js";
 import { isDomainRenderNeeded, markDomainRenderNeeded, } from "./engine/session.js";
 import { isRenderProviderEnabled, renderViaProvider } from "./engine/render.js";
 import { fetchResource, isAntiBotChallenge, lastFetchTier, needsBrowserEscalation, } from "./engine/fetcher.js";
@@ -20,7 +20,7 @@ import { parseYouTubeCached, youtubeVideoId } from "./parse/youtube.js";
 import { rankDocuments } from "./ranker.js";
 import { getSearchCategoriesCatalog, searchWeb } from "./search.js";
 import { inspectTarget, listTargets, resolveTarget } from "./targets.js";
-import { hasCustomExtractRequest } from "./extract/structured.js";
+import { hasCustomExtractRequest, isEmptyExtractResult, } from "./extract/structured.js";
 /** Lowercased hostname of a URL (memo key for the Tier R render memo). */
 function safeDomainOf(url) {
     try {
@@ -53,6 +53,34 @@ export function blockedScrapeResult(url, statusCode, tier, latencyMs, reason) {
         evidence: { tier, blocked: true, statusCode, latencyMs, limitations: [error] },
     };
 }
+/**
+ * JS page whose renderer failed. `success: false` so callers never treat the
+ * unrendered nav shell as content (v1.4.3 smoke test: 114 chars of nav with
+ * success:true after Chrome hung on --dump-dom).
+ */
+export function jsRenderFailedResult(url, statusCode, tier, latencyMs, reason) {
+    return {
+        success: false,
+        mode: "local",
+        url,
+        blocked: false,
+        needsJs: true,
+        statusCode,
+        error: reason,
+        title: url,
+        markdown: "",
+        links: [],
+        data: { url, title: url, markdown: "", links: [] },
+        evidence: {
+            tier,
+            blocked: false,
+            needsJs: true,
+            statusCode,
+            latencyMs,
+            limitations: [reason],
+        },
+    };
+}
 /** Shells shorter than this are unrendered JS / empty pages and must not be cached as content. */
 export const MIN_CACHEABLE_MARKDOWN = 200;
 /** Attach source HTML when the caller asked for the `html` format. */
@@ -82,33 +110,55 @@ function emptyPageError(statusCode) {
         "The site may be blocking automated clients or serving a client-rendered shell. " +
         "Retry with `--mode cloud` (hosted stealth browser), or self-host a render provider via WEBCRAWL_RENDER_URL.");
 }
+function defaultJsRenderError() {
+    return ("Page needs JavaScript rendering; local Chrome was not available or produced no content. " +
+        "Install Chrome (see `webcrawl doctor`), point WEBCRAWL_CHROME_PATH at chrome-headless-shell " +
+        "(`npx playwright install chromium-headless-shell`), run `webcrawl login` and retry with `--mode cloud`, " +
+        "or self-host a render provider via WEBCRAWL_RENDER_URL.");
+}
 /**
- * JS / challenge shell → render provider, then local Chrome dump-dom, then stealth.
- * Returns null when no renderer produced real content.
+ * JS / challenge shell → render provider, then local Chrome (CDP, then dump-dom),
+ * then stealth. Returns null when no renderer produced real content.
  */
 async function escalateJsRender(url) {
+    const errors = [];
     if (isRenderProviderEnabled()) {
         const rendered = await renderViaProvider(url);
         if (rendered && rendered.markdown.trim().length >= 40) {
             return {
-                title: rendered.title,
-                markdown: rendered.markdown,
-                html: "",
-                links: [],
-                tier: "render",
+                page: {
+                    title: rendered.title,
+                    markdown: rendered.markdown,
+                    html: "",
+                    links: [],
+                    tier: "render",
+                },
             };
         }
+        errors.push("render provider produced no content");
     }
     if (isLocalChromeAvailable()) {
-        const html = await renderViaLocalChrome(url);
-        if (html && html.length > 100) {
-            const parsed = fastHtmlToMarkdown(html, url);
+        const chrome = await renderViaLocalChromeDetailed(url);
+        if (chrome.html && chrome.html.length > 100) {
+            const parsed = fastHtmlToMarkdown(chrome.html, url);
             if (parsed.markdown.trim().length > 40 &&
-                !isAntiBotChallenge(html, 200) &&
+                !isAntiBotChallenge(chrome.html, 200) &&
                 !isChallengeMarkdown(parsed.title, parsed.markdown)) {
-                return { ...parsed, html, tier: "browser" };
+                return { page: { ...parsed, html: chrome.html, tier: "browser" } };
             }
+            errors.push(chrome.method
+                ? `${chrome.method} rendered HTML but markdown was still a shell`
+                : "local Chrome rendered HTML but markdown was still a shell");
+        }
+        else if (chrome.error) {
+            errors.push(chrome.error);
         }
+        else {
+            errors.push("local Chrome produced no content");
+        }
+    }
+    else {
+        errors.push("no local Chrome/Chromium binary");
     }
     try {
         const { stealthAvailable } = await import("./engine/stealth.js");
@@ -119,19 +169,22 @@ async function escalateJsRender(url) {
                 esc.markdown.trim().length > 40 &&
                 !isChallengeResult(esc.html, esc.markdown, esc.statusCode)) {
                 return {
-                    title: esc.title,
-                    markdown: esc.markdown,
-                    html: esc.html,
-                    links: esc.links,
-                    tier: esc.tier,
+                    page: {
+                        title: esc.title,
+                        markdown: esc.markdown,
+                        html: esc.html,
+                        links: esc.links,
+                        tier: esc.tier,
+                    },
                 };
             }
+            errors.push("stealth browser produced no content");
         }
     }
     catch {
         /* stealth optional */
     }
-    return null;
+    return { page: null, error: errors.filter(Boolean).join("; ") || defaultJsRenderError() };
 }
 /**
  * Apply `map --filter` / `search` after discovery, then honour `-n` / `limit`.
@@ -329,12 +382,15 @@ export function createLocalEngineClient() {
             const needsRender = needsBrowserEscalation(fetched.status, html, markdown);
             const nativeTier = nativeScrapeTier();
             //    JS shell / challenge: memoize the domain and escalate to a renderer
-            //    (hosted provider, local Chrome dump-dom, then stealth). Never cache
-            //    the unrendered shell.
+            //    (hosted provider, local Chrome CDP, then stealth). Never cache
+            //    the unrendered shell, and never report the shell as success:true.
             if (needsRender) {
                 await markDomainRenderNeeded(renderDomain);
                 // 403/429/503 is already a refusal — don't spend a Chrome launch on the block page.
-                const rendered = blocked && fetched.status >= 400 ? null : await escalateJsRender(targetUrl);
+                const escalated = blocked && fetched.status >= 400
+                    ? { page: null, error: undefined }
+                    : await escalateJsRender(targetUrl);
+                const rendered = escalated.page;
                 if (rendered) {
                     const renderHash = sha256(rendered.markdown);
                     if (rendered.markdown.trim().length >= MIN_CACHEABLE_MARKDOWN) {
@@ -369,43 +425,10 @@ export function createLocalEngineClient() {
                         },
                     }, rendered.html, p.formats);
                 }
-                if (blocked || !markdown.trim()) {
-                    return blockedScrapeResult(targetUrl, fetched.status, nativeTier, Date.now() - start, blocked ? undefined : emptyPageError(fetched.status));
+                if (blocked) {
+                    return blockedScrapeResult(targetUrl, fetched.status, nativeTier, Date.now() - start);
                 }
-                const targetInspection = await inspectTarget({
-                    url: targetUrl,
-                    html,
-                    markdown,
-                });
-                return withOptionalHtml({
-                    success: true,
-                    mode: "local",
-                    url: targetUrl,
-                    title: title || targetInspection.title || targetUrl,
-                    markdown: targetInspection.markdown || markdown,
-                    links,
-                    data: {
-                        url: targetUrl,
-                        title: title || targetInspection.title || targetUrl,
-                        markdown: targetInspection.markdown || markdown,
-                        links,
-                        structured: targetInspection.structured,
-                    },
-                    evidence: {
-                        ...targetInspection.evidence,
-                        tier: nativeTier,
-                        contentHash: sha256(markdown || html),
-                        latencyMs: Date.now() - start,
-                        limitations: [
-                            ...(Array.isArray(targetInspection.evidence?.limitations)
-                                ? targetInspection.evidence.limitations
-                                : []),
-                            "Page needs JS rendering; local Chrome was not available or produced no content. " +
-                                "Install Chrome (see `webcrawl doctor`), run `webcrawl login` and retry with `--mode cloud`, " +
-                                "or self-host a render provider via WEBCRAWL_RENDER_URL.",
-                        ],
-                    },
-                }, html, p.formats);
+                return jsRenderFailedResult(targetUrl, fetched.status, nativeTier, Date.now() - start, escalated.error || (!markdown.trim() ? emptyPageError(fetched.status) : defaultJsRenderError()));
             }
             if (blocked || !markdown.trim()) {
                 return blockedScrapeResult(targetUrl, fetched.status, nativeTier, Date.now() - start, blocked ? undefined : emptyPageError(fetched.status));
@@ -557,22 +580,38 @@ export function createLocalEngineClient() {
                 formats: ["markdown", "html"],
                 noCache: true,
             }));
-            const markdown = (scraped.data && scraped.data.markdown) || scraped.markdown || "";
-            const html = (scraped.data && scraped.data.html) || scraped.html || "";
+            let markdown = (scraped.data && scraped.data.markdown) || scraped.markdown || "";
+            let html = (scraped.data && scraped.data.html) || scraped.html || "";
             if (scraped.blocked || scraped.success === false) {
-                return {
-                    success: false,
-                    blocked: true,
-                    mode: "local",
-                    url: targetUrl,
-                    data: {},
-                    error: scraped.error || emptyPageError(200),
-                    evidence: {
-                        tier: lastFetchTier(),
-                        sourceUrl: targetUrl,
-                        limitations: [String(scraped.error || emptyPageError(200))],
-                    },
-                };
+                // Bot-blocked pages stay failures. A JS shell whose Chrome render failed
+                // can still yield structured data from JSON islands once script tags are
+                // stripped for card matching — never from `document.write` templates.
+                if (scraped.needsJs && scraped.blocked !== true) {
+                    try {
+                        const fetched = await fetchWithTlsImpersonation(targetUrl);
+                        html = fetched.html || html;
+                        const parsedShell = fastHtmlToMarkdown(html, targetUrl);
+                        markdown = parsedShell.markdown || markdown;
+                    }
+                    catch {
+                        /* keep whatever scrape returned */
+                    }
+                }
+                else {
+                    return {
+                        success: false,
+                        blocked: true,
+                        mode: "local",
+                        url: targetUrl,
+                        data: {},
+                        error: scraped.error || emptyPageError(200),
+                        evidence: {
+                            tier: lastFetchTier(),
+                            sourceUrl: targetUrl,
+                            limitations: [String(scraped.error || emptyPageError(200))],
+                        },
+                    };
+                }
             }
             const targetInspection = await inspectTarget({
                 url: targetUrl,
@@ -587,15 +626,35 @@ export function createLocalEngineClient() {
                 bypassCache: true,
             });
             const { cached: _c, cacheLayer: _l, ...extracted } = siteData;
+            const data = extractOutputData(extracted, {
+                prompt: p.prompt,
+                schema: p.schema,
+                target: targetInspection.structured,
+            });
+            if (hasCustomExtractRequest(p.prompt, p.schema) && isEmptyExtractResult(data)) {
+                const error = "Extraction produced no data matching the prompt/schema. " +
+                    "If this is a JS-rendered page, install Chrome (see `webcrawl doctor`) and retry; " +
+                    "empty objects are not reported as success.";
+                return {
+                    success: false,
+                    mode: "local",
+                    url: targetUrl,
+                    data: {},
+                    error,
+                    evidence: {
+                        ...targetInspection.evidence,
+                        limitations: [
+                            ...(targetInspection.evidence?.limitations ?? []),
+                            error,
+                        ],
+                    },
+                };
+            }
             return {
                 success: true,
                 mode: "local",
                 url: targetUrl,
-                data: extractOutputData(extracted, {
-                    prompt: p.prompt,
-                    schema: p.schema,
-                    target: targetInspection.structured,
-                }),
+                data,
                 evidence: targetInspection.evidence,
             };
         },

package/dist/core/search/catalog.d.ts

@@ -52,7 +52,7 @@ export interface SearchRequest {
     bypassCache?: boolean;
     /**
      * Pick one search provider and skip the default multi-engine pipeline.
-     * `auto` (default) uses the free engines and falls back to Jina when results are thin.
+     * `auto` (default) uses the free engines and falls back to a keyed provider when results are thin.
      */
     provider?: SearchProviderChoice;
     rankOptions?: {

package/dist/core/search/catalog.js

@@ -11,7 +11,7 @@ export const SEARCH_PROVIDER_ENV_KEYS = {
 export const SEARCH_CATEGORY_CATALOG = {
     general: {
         description: "Broad web search across general-purpose engines.",
-        engines: ["brave-native", "ddg-lite", "wikipedia"],
+        engines: ["brave-native", "ddg-lite", "wikipedia", "hn-algolia", "bing-rss"],
         localizeBy: "none",
         preferLocalDomains: false,
         subcategories: {

package/dist/core/search/pipeline.js

@@ -115,7 +115,9 @@ export async function searchWeb(req) {
         }
     }
     const perEngineLimit = Math.min(Math.max(limit + 4, 8), 15);
-    const quorumTimeoutMs = req.rankOptions?.quorumTimeoutMs ?? 1500;
+    // Must outlast DDG/Brave impersonation (2200ms) so a slow SERP is not cut off
+    // after Wikipedia's 2 hits — the v1.4.3 keyless-search failure mode.
+    const quorumTimeoutMs = req.rankOptions?.quorumTimeoutMs ?? 2800;
     // Stage 1: Hedged Multi-Engine Native Federation
     const collectedOffers = [];
     const offerByUrl = new Map();
@@ -257,7 +259,7 @@ export async function searchWeb(req) {
         filterContext,
     };
     let ranked = await rankDocuments(localizedSearchQuery, rankCandidates, rankOpts);
-    // Thin result set after ranking and filtering: ask Jina once, then re-rank the combined pool.
+    // Thin result set after ranking and filtering: ask the keyed search provider once, then re-rank.
     const minResults = Math.min(3, limit);
     if ((!req.provider || req.provider === "auto") &&
         ranked.results.length < minResults &&
@@ -649,7 +651,7 @@ export async function searchWeb(req) {
             ...(matchedCves?.length ? { cves: matchedCves } : {}),
         };
     });
-    // Charge the fallback surcharge only when a Jina result actually reached the caller.
+    // Charge the fallback surcharge only when a keyed-provider result actually reached the caller.
     // An explicitly chosen provider is not a fallback.
     const usedJinaFallback = (!req.provider || req.provider === "auto") &&
         finalItems.some((item) => item.sources?.includes("jina"));

package/dist/core/search/providers.d.ts

@@ -14,18 +14,30 @@ export declare function isEntityQuery(q: string): boolean;
  * preventing any single slow engine from inflating tail latency.
  */
 export declare function fetchHedgedNativeFederation(rawQuery: string, localizedQuery: string, perEngineLimit: number, req: SearchRequest, userLocation: UserLocationContext, quorumTimeoutMs: number, targetSites: string[], onOffer: (offer: ProductOffer) => void, onCve?: (cve: CveRecord) => void): Promise<RawCandidateDocument[]>;
-/** Below this many unique URLs from the free engines, the Jina fallback runs. */
+/** Below this many unique URLs from the free engines, the keyed search fallback runs. */
 export declare const JINA_FALLBACK_MIN_URLS = 3;
 /** True when a scraped SERP is a bot-check / captcha rather than results. */
 export declare function isSerpChallengeHtml(html: string, status?: number): boolean;
 export declare const EMPTY_SEARCH_ERROR: string;
+/** Canonical URL for search dedupe (host lowercased, trailing slash / tracking params stripped). */
+export declare function canonicalSearchUrl(raw: string | undefined): string;
+/** Round-robin merge so one source (e.g. Wikipedia) cannot fill `-n` by itself. */
+export declare function mergeSearchCandidates(groups: RawCandidateDocument[][], limit: number): RawCandidateDocument[];
+export declare function withRetry<T>(fn: () => Promise<T>, opts?: {
+    retries?: number;
+    delayMs?: number;
+    isEmpty?: (v: T) => boolean;
+}): Promise<T>;
 /** Parse Bing's `format=rss` web-search feed into candidate docs. */
 export declare function parseBingWebRss(xml: string, limit: number): RawCandidateDocument[];
 /**
- * Last-resort keyless engines used when DDG/Brave HTML SERPs come back empty.
- * Wikipedia OpenSearch, HN Algolia, and Bing RSS do not require a JS challenge.
+ * Last-resort keyless engines used when DDG/Brave HTML SERPs come back empty
+ * or thinner than `limit`. Wikipedia OpenSearch, HN Algolia, Bing RSS, and
+ * DuckDuckGo Instant Answer do not require a JS challenge or API key.
  */
 export declare function fetchKeylessSearchFallbacks(query: string, limit: number): Promise<RawCandidateDocument[]>;
+/** DuckDuckGo Instant Answer API — keyless JSON, unrelated to the HTML SERP challenge. */
+export declare function fetchDuckDuckGoInstantAnswer(query: string, limit: number): Promise<RawCandidateDocument[]>;
 export declare function fetchJinaSearch(query: string, limit: number, req?: {
     country?: string;
     lang?: string;