<?php
// SPDX-License-Identifier: GPL-3.0-or-later
declare(strict_types=1);

namespace Aiwend\Knowledge;

/**
 * Converts raw source content into a normalized Document.
 *
 * Supported types: "text" (TXT), "markdown" (MD), "html" (published WordPress
 * content), and text PDF through a resource-limited CLI worker (ADR-09).
 *
 * Bounded work: input above MAX_BYTES is rejected before parsing; binary
 * input (NUL bytes or invalid UTF-8) is rejected; no external resources are
 * fetched (HTML links and images are ignored, never retrieved).
 */
final class Extractor
{
    public const MAX_BYTES = 2_000_000;

    public function extract(string $raw, string $type, string $title): Document
    {
        if (!in_array($type, ['text', 'markdown', 'html', 'pdf'], true)) {
            throw new ExtractionException('unsupported_type', 'This file type is not supported yet. Supported: TXT, Markdown, published WordPress content.');
        }
        if (strlen($raw) > self::MAX_BYTES) {
            throw new ExtractionException('too_large', 'The content is larger than the 2 MB limit.');
        }
        if ($type === 'pdf') {
            if (!PdfSupport::available()) {
                throw new ExtractionException('pdf_unavailable', 'PDF files are not supported in this edition of Aiwend.');
            }
            return new Document($this->clean($title) ?: 'Untitled', (new PdfExtractor())->extract($raw));
        }
        if (str_contains($raw, "\0") || !mb_check_encoding($raw, 'UTF-8')) {
            throw new ExtractionException('binary', 'The content is not valid UTF-8 text.');
        }
        $raw = (string) preg_replace('/^\xEF\xBB\xBF/', '', $raw); // BOM

        $sections = match ($type) {
            'markdown' => $this->markdownSections($raw),
            'html' => $this->htmlSections($raw),
            default => $this->textSections($raw),
        };
        $sections = array_values(array_filter(
            array_map(fn(array $s): array => ['anchor' => $s['anchor'], 'heading' => $this->clean($s['heading']), 'text' => $this->clean($s['text'])], $sections),
            static fn(array $s): bool => $s['text'] !== '' && preg_match('/[\p{L}\p{N}]/u', $s['text']) !== 0 // a section with no letter or digit (a lone "#", a rule, a symbol) has nothing to retrieve; a regex error (false) keeps the text rather than losing it
        ));
        if ($sections === []) {
            throw new ExtractionException('empty', 'No readable text was found.');
        }
        return new Document($this->clean($title) ?: 'Untitled', $sections);
    }

    /** @return list<array{anchor:string, heading:string, text:string}> */
    private function textSections(string $raw): array
    {
        return [['anchor' => 'section-1', 'heading' => '', 'text' => $raw]];
    }

    /** @return list<array{anchor:string, heading:string, text:string}> */
    private function markdownSections(string $raw): array
    {
        $sections = [];
        $heading = '';
        $buffer = [];
        $used = [];
        $n = 0;
        $flush = function () use (&$sections, &$heading, &$buffer, &$used, &$n): void {
            $text = trim(implode("\n", $buffer));
            if ($text !== '' || $heading !== '') {
                $n++;
                $sections[] = ['anchor' => $this->uniqueSlug($heading, $n, $used), 'heading' => $heading, 'text' => $text];
            }
            $buffer = [];
        };
        $inFence = false;
        foreach (preg_split('/\R/u', $raw) ?: [] as $line) {
            if (preg_match('/^\s*(```|~~~)/', $line)) {
                $inFence = !$inFence;
                $buffer[] = $line;
                continue;
            }
            if (!$inFence && preg_match('/^\s{0,3}(#{1,6})\s+(.+?)\s*#*\s*$/u', $line, $m)) {
                $flush();
                $heading = $this->stripInlineMarkdown($m[2]);
                continue;
            }
            $buffer[] = $this->stripInlineMarkdown($line);
        }
        $flush();
        return $sections;
    }

    /** @return list<array{anchor:string, heading:string, text:string}> */
    private function htmlSections(string $raw): array
    {
        // Drop content that is never visible text.
        $raw = (string) preg_replace('#<(script|style|noscript|template|svg|iframe)\b[^>]*>.*?</\1>#is', ' ', $raw);
        $raw = (string) preg_replace('#<!--.*?-->#s', ' ', $raw);
        $parts = preg_split('#<h([1-6])\b([^>]*)>(.*?)</h\1>#is', $raw, -1, PREG_SPLIT_DELIM_CAPTURE) ?: [$raw];
        $sections = [];
        $used = [];
        $n = 0;
        $heading = '';
        $explicitId = '';
        for ($i = 0; $i < count($parts); $i++) {
            if ($i % 4 === 0) {
                $text = $this->htmlToText($parts[$i]);
                if (trim($text) !== '' || $heading !== '') {
                    $n++;
                    // Only an id that exists in the HTML can be linked as #fragment; otherwise use a
                    // non-linkable ordinal (headings in WordPress content usually have no id).
                    $anchor = $explicitId !== '' && !isset($used[$explicitId]) ? $explicitId : 'section-' . $n;
                    $used[$anchor] = true;
                    $sections[] = ['anchor' => $anchor, 'heading' => $heading, 'text' => $text];
                }
                continue;
            }
            if ($i % 4 === 2) {
                // Prefer an existing id="" on the heading: then the citation link really resolves.
                $explicitId = preg_match('/\bid\s*=\s*["\']([A-Za-z][\w\-:.]{0,100})["\']/', $parts[$i], $m) ? $m[1] : '';
            }
            if ($i % 4 === 3) {
                $heading = trim($this->htmlToText($parts[$i]));
            }
        }
        return $sections;
    }

    private function htmlToText(string $html): string
    {
        // Block-level boundaries become line breaks and table cells are separated by a space, so that "Plan" and "Price" in a table, or a
        // term and its definition, do not run together into one word that no question could match (Phase 115).
        $html = (string) preg_replace('#<(br|hr|/p|/div|/li|/tr|/h[1-6]|/blockquote|/section|/article|/header|/footer|/aside|/figure|/figcaption|/caption|/dt|/dd|/ul|/ol|/table|/pre|/details|/summary)\b[^>]*>#i', "\n", $html);
        $html = (string) preg_replace('#</t[dh]\s*>#i', ' ', $html);
        $html = (string) preg_replace('#<li\b[^>]*>#i', "\n- ", $html);
        // phpcs:ignore WordPress.WP.AlternativeFunctions.strip_tags_strip_tags -- pure PHP extractor without a WordPress dependency; the result is plain text escaped on output
        $text = strip_tags($html);
        return html_entity_decode($text, ENT_QUOTES | ENT_HTML5, 'UTF-8');
    }

    private function stripInlineMarkdown(string $line): string
    {
        $line = (string) preg_replace('/!\[([^\]]*)\]\([^)]*\)/u', '$1', $line);   // images → alt text
        $line = (string) preg_replace('/\[([^\]]+)\]\([^)]*\)/u', '$1', $line);    // links → label
        return (string) preg_replace('/(\*\*|__|`)/u', '', $line);
    }

    /** @param array<string, true> $used */
    private function uniqueSlug(string $heading, int $n, array &$used): string
    {
        $slug = trim((string) preg_replace('/[^a-z0-9]+/', '-', strtolower($this->asciiFold($heading))), '-');
        $slug = $slug === '' ? 'section-' . $n : substr($slug, 0, 80);
        $base = $slug;
        for ($k = 2; isset($used[$slug]); $k++) {
            $slug = $base . '-' . $k;
        }
        $used[$slug] = true;
        return $slug;
    }

    private function asciiFold(string $s): string
    {
        if (function_exists('transliterator_transliterate')) {
            $t = transliterator_transliterate('Any-Latin; Latin-ASCII', $s);
            if (is_string($t)) {
                return $t;
            }
        }
        return $s;
    }

    private function clean(string $text): string
    {
        if (class_exists(\Normalizer::class)) {
            $text = (string) \Normalizer::normalize($text, \Normalizer::FORM_C);
        }
        $text = (string) preg_replace('/[\x00-\x08\x0B\x0C\x0E-\x1F\x7F]/u', '', $text);
        $text = (string) preg_replace('/[ \t\x{00A0}]+/u', ' ', $text);
        $text = (string) preg_replace('/ *\n */u', "\n", $text);
        $text = (string) preg_replace('/\n{3,}/u', "\n\n", $text);
        return trim($text);
    }
}