@apmantza/greedysearch-pi 1.9.1 → 2.0.0

package/src/search/pdf.mjs

@@ -0,0 +1,68 @@
+// src/search/pdf.mjs — PDF text extraction helpers
+//
+// Adapted from pi-webaio's PDF pipeline. `pdf-parse` is loaded lazily so the
+// package remains importable even when optional native canvas bindings are not
+// available. PDF extraction is only attempted for actual PDF source fetches.
+
+function ensurePdfDomPolyfills() {
+	if (typeof globalThis.DOMMatrix === "undefined") {
+		globalThis.DOMMatrix = class DOMMatrix {
+			constructor(_init = undefined) {}
+			multiplySelf() {
+				return this;
+			}
+			preMultiplySelf() {
+				return this;
+			}
+			translateSelf() {
+				return this;
+			}
+			scaleSelf() {
+				return this;
+			}
+			rotateSelf() {
+				return this;
+			}
+		};
+	}
+	if (typeof globalThis.ImageData === "undefined") {
+		globalThis.ImageData = class ImageData {
+			constructor(data = undefined, width = 0, height = 0) {
+				this.data = data;
+				this.width = width;
+				this.height = height;
+			}
+		};
+	}
+	if (typeof globalThis.Path2D === "undefined") {
+		globalThis.Path2D = class Path2D {
+			constructor(_path = undefined) {}
+		};
+	}
+}
+
+async function loadPdfParseCtor() {
+	ensurePdfDomPolyfills();
+	const mod = await import("pdf-parse");
+	const ctor = mod.PDFParse ?? mod.default;
+	if (!ctor) throw new Error("pdf-parse did not export PDFParse");
+	return ctor;
+}
+
+export async function extractPdfMarkdown(buffer, url) {
+	try {
+		const PDFParseCtor = await loadPdfParseCtor();
+		const parser = new PDFParseCtor({ data: new Uint8Array(buffer) });
+		await parser.load();
+		const data = await parser.getText();
+		const text = data.text?.trim();
+		if (!text) return null;
+		return {
+			title: new URL(url).pathname.split("/").pop() || "Document.pdf",
+			content: `## PDF Content (${data.total} pages)\n\n${text}`,
+			pages: data.total,
+		};
+	} catch (error) {
+		return { error: error.message || String(error) };
+	}
+}

package/src/search/recovery.mjs

@@ -1,45 +1,51 @@
-// src/search/recovery.mjs — Headless-block detection and visible recovery policy
-
-// Only these engines use automatic headless → visible recovery. Google is
-// intentionally excluded for now; see issue #9 discussion / maintainer choice.
-export const HEADLESS_RECOVERY_ENGINES = ["perplexity", "bing"];
-
-const HEADLESS_BLOCKED_PATTERN =
-	/timed out|timeout|verification|captcha|cloudflare|turnstile|input not found|ask-input|clipboard|copy button hidden/i;
-
-const MANUAL_VERIFICATION_PATTERN =
-	/needs-human|verification required|please solve|captcha|cloudflare|turnstile|could not be completed automatically|manual intervention/i;
-
-export function isHeadlessBlockedError(error) {
-	return HEADLESS_BLOCKED_PATTERN.test(String(error || ""));
-}
-
-export function isManualVerificationError(error) {
-	return MANUAL_VERIFICATION_PATTERN.test(String(error || ""));
-}
-
-export function findHeadlessBlockedEngines(resultsByEngine) {
-	return HEADLESS_RECOVERY_ENGINES.filter((engine) => {
-		const result = resultsByEngine?.[engine];
-		if (!result) return false;
-		// Data-driven: check envelope first (zero regex cost)
-		if (result._envelope?.blockedBy) return true;
-		if (result._envelope?.verificationResult === "needs-human") return true;
-		// Fallback: legacy string matching for errors passed as plain strings
-		const error = result.error;
-		return error && isHeadlessBlockedError(error);
-	});
-}
-
-/**
- * Check if an extractor Error carries a structured envelope indicating
- * headless blocking. Used in single-engine recovery paths where the Error
- * object is caught directly rather than parsed from a result record.
- */
-export function isHeadlessBlockedResult(error) {
-	if (!error) return false;
-	const env = error.envelope;
-	if (env?.blockedBy) return true;
-	if (env?.verificationResult === "needs-human") return true;
-	return isHeadlessBlockedError(error.message);
-}
+// src/search/recovery.mjs — Headless-block detection and visible recovery policy
+
+// Only these engines use automatic headless → visible recovery. Google is
+// intentionally excluded for now; see issue #9 discussion / maintainer choice.
+export const HEADLESS_RECOVERY_ENGINES = [
+	"perplexity",
+	"bing",
+	"chatgpt",
+	"semantic-scholar",
+	"logically",
+];
+
+const HEADLESS_BLOCKED_PATTERN =
+	/timed out|timeout|verification|captcha|cloudflare|turnstile|input not found|ask-input|clipboard|copy button hidden|sign.in|login required/i;
+
+const MANUAL_VERIFICATION_PATTERN =
+	/needs-human|verification required|please solve|captcha|cloudflare|turnstile|could not be completed automatically|manual intervention|sign.in|login required/i;
+
+export function isHeadlessBlockedError(error) {
+	return HEADLESS_BLOCKED_PATTERN.test(String(error || ""));
+}
+
+export function isManualVerificationError(error) {
+	return MANUAL_VERIFICATION_PATTERN.test(String(error || ""));
+}
+
+export function findHeadlessBlockedEngines(resultsByEngine) {
+	return HEADLESS_RECOVERY_ENGINES.filter((engine) => {
+		const result = resultsByEngine?.[engine];
+		if (!result) return false;
+		// Data-driven: check envelope first (zero regex cost)
+		if (result._envelope?.blockedBy) return true;
+		if (result._envelope?.verificationResult === "needs-human") return true;
+		// Fallback: legacy string matching for errors passed as plain strings
+		const error = result.error;
+		return error && isHeadlessBlockedError(error);
+	});
+}
+
+/**
+ * Check if an extractor Error carries a structured envelope indicating
+ * headless blocking. Used in single-engine recovery paths where the Error
+ * object is caught directly rather than parsed from a result record.
+ */
+export function isHeadlessBlockedResult(error) {
+	if (!error) return false;
+	const env = error.envelope;
+	if (env?.blockedBy) return true;
+	if (env?.verificationResult === "needs-human") return true;
+	return isHeadlessBlockedError(error.message);
+}