/** @file Serialize markdown module. */
import TurndownService from "turndown";
import { gfm } from "turndown-plugin-gfm";

import { normalizeWhitespace } from "./text.ts";

export interface MarkdownOptions {
	removeImages?: boolean;
}

/**
 * Converts cleaned HTML to stable Markdown for model-facing output.
 *
 * @remarks
 *   Turndown service construction registers rule objects and plugins. Keeping one configured
 *   service per image policy avoids repeated setup in hot scrape paths while preserving
 *   deterministic output rules.
 */
export function htmlToMarkdown(html: string, options: MarkdownOptions = {}): string {
	const service = options.removeImages === false ? keepImagesService : removeImagesService;
	return normalizeWhitespace(service.turndown(stripLargeElements(html)));
}

/**
 * Strip large tables and very long lists before Turndown to avoid expensive conversion on
 * element-heavy pages where the output is likely to be truncated anyway. Only applies when HTML
 * exceeds 40 KB. Tables: > 20 rows. Lists: > 100 items.
 */
function stripLargeElements(html: string): string {
	if (html.length < 40_000) return html;
	// Quick check: does HTML contain tables or lists at all?
	const hasTable = html.includes("<table");
	const hasList = html.includes("<ul") || html.includes("<ol");
	if (!hasTable && !hasList) return html;
	// Count elements only if present - use efficient regex
	let trCount = 0;
	let liCount = 0;
	if (hasTable) {
		const trMatches = html.match(/<tr[\s>]/giu);
		trCount = trMatches ? trMatches.length : 0;
	}
	if (hasList) {
		const liMatches = html.match(/<li[\s>]/giu);
		liCount = liMatches ? liMatches.length : 0;
	}
	if (trCount < 20 && liCount < 100) return html;
	// Strip tables and/or lists if thresholds exceeded
	let result = html;
	if (trCount >= 20) {
		result = result.replaceAll(/<table[\s\S]*?<\/table>/giu, "\n\n");
	}
	if (liCount >= 100) {
		result = result.replaceAll(/<(ul|ol)[\s\S]*?<\/(ul|ol)>/giu, "\n\n[Long list]\n\n");
	}
	return result;
}

function createMarkdownService(removeImages: boolean): TurndownService {
	const turndown = new TurndownService({
		codeBlockStyle: "fenced",
		headingStyle: "atx",
		bulletListMarker: "-",
		emDelimiter: "_",
		strongDelimiter: "**",
	});
	turndown.use(gfm);
	turndown.remove(["script", "style", "noscript", "template"]);
	if (removeImages) {
		turndown.addRule("removeImages", { filter: "img", replacement: () => "" });
	}
	turndown.addRule("stableLinks", {
		filter: "a",
		replacement: (content, node) => {
			const href = node.getAttribute("href");
			// Fast path: no href, return content as-is
			if (!href) return content;
			// Simple trim instead of full normalizeWhitespace for link labels
			const label = content.trim().replaceAll(/\s+/gu, " ");
			return label ? `[${label}](${href})` : href;
		},
	});
	return turndown;
}

const removeImagesService = createMarkdownService(true);
const keepImagesService = createMarkdownService(false);