src/lib/artifacts/ArtifactExtractor.ts

/**
 * ArtifactExtractor.ts — Parse agent LLM responses to extract workspace artifacts (FASE 7)
 *
 * Detects code blocks, HTML pages, markdown headers, JSON blobs, CSV tables
 * in raw LLM text output and converts them to Artifact objects ready for storage.
 *
 * Heuristics (in priority order):
 *   1. Fenced code blocks with explicit language tag: ```python … ```
 *   2. ```html … ``` → html kind
 *   3. ```json … ``` → json kind
 *   4. ```csv … ``` → csv kind
 *   5. ```svg … ``` → svg kind
 *   6. ```markdown … ``` or ``` … ``` that looks like markdown → markdown kind
 *   7. Long code block without tag (≥10 lines) → code / text kind
 */

import type { Artifact, ArtifactKind } from "./ArtifactStore";

// ── Language → kind mapping ───────────────────────────────────────────────────

const CODE_LANGS = new Set([
  "python","py","javascript","js","typescript","ts","jsx","tsx",
  "rust","go","java","c","cpp","c++","c#","csharp","bash","sh","shell",
  "ruby","rb","php","swift","kotlin","r","scala","dart","lua",
  "yaml","toml","sql","graphql","regex",
]);

function langToKind(lang: string): ArtifactKind {
  const l = lang.toLowerCase().trim();
  if (l === "html")            return "html";
  if (l === "json")            return "json";
  if (l === "csv")             return "csv";
  if (l === "svg")             return "svg";
  if (l === "markdown" || l === "md") return "markdown";
  if (CODE_LANGS.has(l))      return "code";
  return "code"; // default
}

function inferKindFromContent(content: string): ArtifactKind {
  const trimmed = content.trim();
  if (trimmed.startsWith("<html") || (trimmed.startsWith("<!") && /doctype/i.test(trimmed))) return "html";
  if (trimmed.startsWith("{") || trimmed.startsWith("[")) {
    try { JSON.parse(trimmed); return "json"; } catch { /* not json */ }
  }
  if (trimmed.startsWith("<svg")) return "svg";
  const lines = trimmed.split("\n");
  const hasCommas = lines.slice(0, 5).every(l => l.includes(","));
  if (hasCommas && lines.length >= 2) return "csv";
  if (/^#{1,6} /m.test(trimmed)) return "markdown";
  return "code";
}

// ── Title inference ───────────────────────────────────────────────────────────

const TITLE_MAX = 48;

function inferTitle(kind: ArtifactKind, language: string | undefined, content: string, idx: number): string {
  // Try to extract a meaningful name from the content
  const lines = content.trim().split("\n");

  if (kind === "html") {
    const m = content.match(/<title>([^<]+)<\/title>/i);
    if (m) return m[1].trim().slice(0, TITLE_MAX);
    const h1 = content.match(/<h1[^>]*>([^<]+)<\/h1>/i);
    if (h1) return h1[1].trim().slice(0, TITLE_MAX);
    return "Pagina HTML";
  }

  if (kind === "json") {
    try {
      const j = JSON.parse(content.trim());
      if (j.name)  return String(j.name).slice(0, TITLE_MAX);
      if (j.title) return String(j.title).slice(0, TITLE_MAX);
    } catch { /* ok */ }
    return "Dati JSON";
  }

  if (kind === "csv") {
    const header = lines[0].replace(/,/g, " / ").trim();
    return header.slice(0, TITLE_MAX) || "Tabella CSV";
  }

  if (kind === "svg") return "Immagine SVG";

  if (kind === "markdown") {
    const h1 = content.match(/^#{1,3} (.+)/m);
    if (h1) return h1[1].trim().slice(0, TITLE_MAX);
    return "Documento";
  }

  // code — look for function name, class name, filename comment
  const commentFile = content.match(/^#\s+(?:file:\s*)?([a-zA-Z0-9_.\-/]+\.[a-z]{1,5})/m);
  if (commentFile) return commentFile[1].slice(0, TITLE_MAX);
  const fnName = content.match(/^(?:def|function|fn|class|func)\s+(\w+)/m);
  if (fnName) return `${language ?? "codice"}: ${fnName[1]}`.slice(0, TITLE_MAX);

  return `${language ?? kind} snippet ${idx + 1}`;
}

// ── Unique IDs ────────────────────────────────────────────────────────────────

let _counter = 0;
function shortId(): string {
  return `${Date.now().toString(36)}_${(++_counter).toString(36)}`;
}

// ── Fenced code block regex ───────────────────────────────────────────────────

const FENCE_RE = /```([a-zA-Z0-9_+#\-.]*)\n([\s\S]*?)```/g;

// ── Main extraction function ──────────────────────────────────────────────────

export interface ExtractOptions {
  sessionId?:  string;
  messageId?:  string;
  minLines?:   number;  // minimum lines to consider an artifact (default 4)
}

export function extractArtifacts(
  text: string,
  opts: ExtractOptions = {},
): Artifact[] {
  const { sessionId, messageId, minLines = 4 } = opts;
  const now = Date.now();
  const found: Artifact[] = [];
  let idx = 0;

  let match: RegExpExecArray | null;
  FENCE_RE.lastIndex = 0;
  while ((match = FENCE_RE.exec(text)) !== null) {
    const rawLang = (match[1] ?? "").toLowerCase().trim();
    const content = match[2].trim();

    if (!content || content.split("\n").length < minLines) continue;

    const language = rawLang || undefined;
    const kind     = rawLang ? langToKind(rawLang) : inferKindFromContent(content);
    const title    = inferTitle(kind, language, content, idx);

    found.push({
      id:        shortId(),
      kind,
      language,
      title,
      content,
      sessionId,
      messageId,
      createdAt: now,
      updatedAt: now,
      version:   1,
      pinned:    false,
      tags:      language ? [language] : [],
    });
    idx++;
  }

  return found;
}

/**
 * Quick check — does this text contain anything worth extracting?
 * Cheap scan before running full extraction.
 */
export function hasExtractableContent(text: string): boolean {
  return FENCE_RE.test(text);
}