//
//  PageContext.swift
//  Zyquo Atlas
//
//  Author: Simon-Pierre Boucher
//  Mail: contact@spboucher.ai
//
//  Normalized, model-ready representation of a web page: clean markdown, key
//  metadata, the user's current selection, and a heading index for chunking and
//  citations. Produced by ContentExtractor from the injected AtlasExtractor.js;
//  chunked by Chunker for long pages / multi-tab reasoning. This is the ONLY
//  page representation the AI layer sees — WKWebView/DOM guts never leak past it.
//

import Foundation

/// Extraction fidelity, so the UI can be honest when a page couldn't be read
/// cleanly (docs/AI-BROWSER-RESEARCH.md §2: extraction failure is AI failure).
enum ExtractionQuality: String, Codable {
    case reader   // Readability main-content
    case rawText  // visible structured-text fallback (app pages)
}

struct PageHeading: Codable, Hashable {
    let level: Int
    let title: String
    /// Character offset into `markdown`.
    let offset: Int
}

struct PageSelection: Codable, Hashable {
    let text: String
    /// Surrounding block text for grounding "explain this".
    let context: String?
}

struct PageContext: Codable, Hashable, Identifiable {
    var id: String { canonical.isEmpty ? url : canonical }

    let url: String
    let canonical: String
    let title: String
    let byline: String?
    let description: String?
    let siteName: String?
    let lang: String?
    let published: String?
    let favicon: String?
    /// Canonical model-facing representation.
    let markdown: String
    let quality: ExtractionQuality
    let headings: [PageHeading]
    let selection: PageSelection?
    let wordCount: Int
    let truncated: Bool

    /// Rough token estimate (chars/4) for the stuff-vs-map-reduce budget rule.
    var estimatedTokens: Int { markdown.count / 4 }

    private enum CodingKeys: String, CodingKey {
        case url, canonical, title, byline, description, siteName, lang
        case published, favicon, markdown, quality, headings, selection
        case wordCount, truncated
    }

    /// Full initializer (the Decodable init handles the JS bridge separately).
    init(url: String, canonical: String, title: String, byline: String? = nil,
         description: String? = nil, siteName: String? = nil, lang: String? = nil,
         published: String? = nil, favicon: String? = nil, markdown: String,
         quality: ExtractionQuality = .reader, headings: [PageHeading] = [],
         selection: PageSelection? = nil, wordCount: Int = 0, truncated: Bool = false) {
        self.url = url; self.canonical = canonical; self.title = title; self.byline = byline
        self.description = description; self.siteName = siteName; self.lang = lang
        self.published = published; self.favicon = favicon; self.markdown = markdown
        self.quality = quality; self.headings = headings; self.selection = selection
        self.wordCount = wordCount; self.truncated = truncated
    }

    /// Copies another context, replacing the markdown (used by the verifier to
    /// trim the sweep context so many model calls stay cheap but grounded).
    init(cloning other: PageContext, markdown: String) {
        self.init(url: other.url, canonical: other.canonical, title: other.title,
                  byline: other.byline, description: other.description, siteName: other.siteName,
                  lang: other.lang, published: other.published, favicon: other.favicon,
                  markdown: markdown, quality: other.quality, headings: other.headings,
                  selection: other.selection, wordCount: markdown.split(separator: " ").count,
                  truncated: other.truncated)
    }

    /// A minimal baked context (verifier fallback if extraction fails entirely).
    init(bakedTitle: String, url: String, markdown: String) {
        self.init(url: url, canonical: url, title: bakedTitle, markdown: markdown,
                  quality: .reader, wordCount: markdown.split(separator: " ").count)
    }

    // Tolerate a missing/unknown `quality` from JS by defaulting to rawText.
    init(from decoder: Decoder) throws {
        let c = try decoder.container(keyedBy: CodingKeys.self)
        url = try c.decode(String.self, forKey: .url)
        canonical = (try? c.decode(String.self, forKey: .canonical)) ?? url
        title = (try? c.decode(String.self, forKey: .title)) ?? ""
        byline = try? c.decodeIfPresent(String.self, forKey: .byline)
        description = try? c.decodeIfPresent(String.self, forKey: .description)
        siteName = try? c.decodeIfPresent(String.self, forKey: .siteName)
        lang = try? c.decodeIfPresent(String.self, forKey: .lang)
        published = try? c.decodeIfPresent(String.self, forKey: .published)
        favicon = try? c.decodeIfPresent(String.self, forKey: .favicon)
        markdown = (try? c.decode(String.self, forKey: .markdown)) ?? ""
        quality = ExtractionQuality(
            rawValue: (try? c.decode(String.self, forKey: .quality)) ?? "rawText"
        ) ?? .rawText
        headings = (try? c.decode([PageHeading].self, forKey: .headings)) ?? []
        selection = try? c.decodeIfPresent(PageSelection.self, forKey: .selection)
        wordCount = (try? c.decode(Int.self, forKey: .wordCount)) ?? 0
        truncated = (try? c.decode(Bool.self, forKey: .truncated)) ?? false
    }
}