# File: main/app.py
# Purpose: One Space that offers four tools/tabs:
#   1) Fetch — extract relevant page content (title, metadata, clean text, hyperlinks)
#   2) Websearch — structured DuckDuckGo search via LangChain tool (JSON)
#   3) Unstructured DDG — raw DuckDuckGo list[dict] rendered into a Textbox
#   4) DDG (Concise) — ultra-succinct DuckDuckGo search that emits JSONL with short keys to minimize tokens

from __future__ import annotations

import re
import json
from typing import List, Dict, Literal, Tuple

import gradio as gr
import requests
from bs4 import BeautifulSoup
from readability import Document
from urllib.parse import urljoin, urldefrag, urlparse
from langchain_community.tools import DuckDuckGoSearchResults
from duckduckgo_search import DDGS


# ==============================
# Fetch: HTTP + extraction utils
# ==============================

def _http_get(url: str) -> requests.Response:
    """
    Download the page politely with a short timeout and realistic headers.
    """
    headers = {
        "User-Agent": "Mozilla/5.0 (compatible; WebMCP/1.0; +https://example.com)",
        "Accept-Language": "en-US,en;q=0.9",
        "Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8",
    }
    return requests.get(url, headers=headers, timeout=15)


def _normalize_whitespace(text: str) -> str:
    """
    Squeeze extra spaces and blank lines to keep things compact.
    """
    text = re.sub(r"[ \t\u00A0]+", " ", text)
    text = re.sub(r"\n\s*\n\s*\n+", "\n\n", text.strip())
    return text.strip()


def _truncate(text: str, max_chars: int) -> Tuple[str, bool]:
    """
    Cut text if it gets too long; return the text and whether we trimmed.
    """
    if max_chars is None or max_chars <= 0 or len(text) <= max_chars:
        return text, False
    return text[:max_chars].rstrip() + " …", True


def _shorten(text: str, limit: int) -> str:
    """
    Hard cap a string with an ellipsis to keep tokens small.
    """
    if limit <= 0 or len(text) <= limit:
        return text
    return text[: max(0, limit - 1)].rstrip() + "…"


def _domain_of(url: str) -> str:
    """
    Show a friendly site name like "example.com".
    """
    try:
        return urlparse(url).netloc or ""
    except Exception:
        return ""


def _meta(soup: BeautifulSoup, name: str) -> str | None:
    tag = soup.find("meta", attrs={"name": name})
    return tag.get("content") if tag and tag.has_attr("content") else None


def _og(soup: BeautifulSoup, prop: str) -> str | None:
    tag = soup.find("meta", attrs={"property": prop})
    return tag.get("content") if tag and tag.has_attr("content") else None


def _extract_metadata(soup: BeautifulSoup, final_url: str) -> Dict[str, str]:
    """
    Pull the useful bits: title, description, site name, canonical URL, language, etc.
    """
    meta: Dict[str, str] = {}

    # Title preference: <title> > og:title > twitter:title
    title_candidates = [
        (soup.title.string if soup.title and soup.title.string else None),
        _og(soup, "og:title"),
        _meta(soup, "twitter:title"),
    ]
    meta["title"] = next((t.strip() for t in title_candidates if t and t.strip()), "")

    # Description preference: description > og:description > twitter:description
    desc_candidates = [
        _meta(soup, "description"),
        _og(soup, "og:description"),
        _meta(soup, "twitter:description"),
    ]
    meta["description"] = next((d.strip() for d in desc_candidates if d and d.strip()), "")

    # Canonical link (helps dedupe)
    link_canonical = soup.find("link", rel=lambda v: v and "canonical" in v)
    meta["canonical"] = (link_canonical.get("href") or "").strip() if link_canonical else ""

    # Site name + language info if present
    meta["site_name"] = (_og(soup, "og:site_name") or "").strip()
    html_tag = soup.find("html")
    meta["lang"] = (html_tag.get("lang") or "").strip() if html_tag else ""

    # Final URL + domain
    meta["fetched_url"] = final_url
    meta["domain"] = _domain_of(final_url)

    return meta


def _extract_main_text(html: str) -> Tuple[str, BeautifulSoup]:
    """
    Use Readability to isolate the main article and turn it into clean text.
    Returns (clean_text, soup_of_readable_html).
    """
    # Simplified article HTML from Readability
    doc = Document(html)
    readable_html = doc.summary(html_partial=True)

    # Parse simplified HTML
    s = BeautifulSoup(readable_html, "lxml")

    # Remove noisy tags
    for sel in ["script", "style", "noscript", "iframe", "svg"]:
        for tag in s.select(sel):
            tag.decompose()

    # Keep paragraphs, list items, and subheadings for structure without bloat
    text_parts: List[str] = []
    for p in s.find_all(["p", "li", "h2", "h3", "h4", "blockquote"]):
        chunk = p.get_text(" ", strip=True)
        if chunk:
            text_parts.append(chunk)

    clean_text = _normalize_whitespace("\n\n".join(text_parts))
    return clean_text, s


def _extract_links(readable_soup: BeautifulSoup, base_url: str, max_links: int) -> List[Tuple[str, str]]:
    """
    Collect clean, unique, absolute links from the readable section only.
    """
    seen = set()
    links: List[Tuple[str, str]] = []

    for a in readable_soup.find_all("a", href=True):
        href = a.get("href").strip()
        # Skip junk links we can't use
        if not href or href.startswith("#") or href.startswith("mailto:") or href.startswith("javascript:"):
            continue

        # Resolve relative URLs, strip fragments (#…)
        absolute = urljoin(base_url, href)
        absolute, _ = urldefrag(absolute)

        if absolute in seen:
            continue
        seen.add(absolute)

        text = a.get_text(" ", strip=True)
        if len(text) > 120:
            text = text[:117] + "…"

        links.append((text or absolute, absolute))

        if len(links) >= max_links > 0:
            break

    return links


def _format_markdown(
    meta: Dict[str, str],
    body: str,
    body_truncated: bool,
    links: List[Tuple[str, str]],
    include_text: bool,
    include_metadata: bool,
    include_links: bool,
    verbosity: str,
) -> str:
    """
    Assemble a compact Markdown summary with optional sections.
    """
    lines: List[str] = []

    # Title header
    title = meta.get("title") or meta.get("domain") or "Untitled"
    lines.append(f"# {title}")

    # Metadata section (only show what exists)
    if include_metadata:
        md: List[str] = []
        if meta.get("description"):
            md.append(f"- **Description:** {meta['description']}")
        if meta.get("site_name"):
            md.append(f"- **Site:** {meta['site_name']}")
        if meta.get("canonical"):
            md.append(f"- **Canonical:** {meta['canonical']}")
        if meta.get("lang"):
            md.append(f"- **Language:** {meta['lang']}")
        if meta.get("fetched_url"):
            md.append(f"- **Fetched From:** {meta['fetched_url']}")
        if md:
            lines.append("## Metadata")
            lines.extend(md)

    # Body text
    if include_text and body:
        if verbosity == "Brief":
            brief, was_more = _truncate(body, 800)
            lines.append("## Text")
            lines.append(brief)
            if was_more or body_truncated:
                lines.append("\n> (Trimmed for brevity)")
        else:
            lines.append("## Text")
            lines.append(body)
            if body_truncated:
                lines.append("\n> (Trimmed for brevity)")

    # Links section
    if include_links and links:
        lines.append(f"## Links ({len(links)})")
        for text, url in links:
            lines.append(f"- [{text}]({url})")

    return "\n\n".join(lines).strip()


def Fetch_Webpage(  # <-- MCP tool #1 (Fetch)
    url: str,
    verbosity: str = "Standard",
    include_metadata: bool = True,
    include_text: bool = True,
    include_links: bool = True,
    max_chars: int = 3000,
    max_links: int = 20,
) -> str:
    """
    Fetch a web page and return a compact Markdown summary that includes title, key
    metadata, readable main text, and outbound links.

    Args:
        url (str): The HTTP/HTTPS URL to fetch. Must be publicly reachable.
        verbosity (str): Controls body length. One of: "Brief", "Standard", or "Full".
            - Brief ≈ up to 1,200 chars
            - Standard ≈ up to 3,000 chars
            - Full = no cap (still limited by `max_chars` if smaller)
        include_metadata (bool): If True, include a Metadata section with description,
            site name, canonical URL, language, and fetched URL.
        include_text (bool): If True, include the extracted readable body text.
        include_links (bool): If True, include a list of outbound links found in the
            readable section only (deduped and fragment-stripped).
        max_chars (int): Hard cap for body text length. Numeric value between 400 and
            12000. The effective cap is the smaller of this value and the preset based
            on `verbosity`.
        max_links (int): Maximum number of links to include. Numeric value between 0 and 100.

    Returns:
        str: Markdown string containing the extracted summary. If the page cannot be
        fetched or parsed, a short error message is returned instead.
    """
    if not url or not url.strip():
        return "Please enter a valid URL."

    try:
        resp = _http_get(url)
        resp.raise_for_status()
    except requests.exceptions.RequestException as e:
        return f"An error occurred: {e}"

    final_url = str(resp.url)
    ctype = resp.headers.get("Content-Type", "")
    if "html" not in ctype.lower():
        return f"Unsupported content type for extraction: {ctype or 'unknown'}"

    # Decode to text
    resp.encoding = resp.encoding or resp.apparent_encoding
    html = resp.text

    # Full-page soup for metadata
    full_soup = BeautifulSoup(html, "lxml")
    meta = _extract_metadata(full_soup, final_url)

    # Readable content
    body_text, readable_soup = _extract_main_text(html)
    if not body_text:
        # Fallback to "whole-page text" if Readability found nothing
        fallback_text = full_soup.get_text(" ", strip=True)
        body_text = _normalize_whitespace(fallback_text)

    # Verbosity presets (we keep the smaller of preset vs. user cap)
    preset_caps = {"Brief": 1200, "Standard": 3000, "Full": 999_999}
    target_cap = preset_caps.get(verbosity, 3000)
    cap = min(max_chars if max_chars > 0 else target_cap, target_cap)
    body_text, truncated = _truncate(body_text, cap) if include_text else ("", False)

    # Extract links from the simplified content only
    links = _extract_links(readable_soup, final_url, max_links=max_links if include_links else 0)

    # Final compact Markdown
    md = _format_markdown(
        meta=meta,
        body=body_text,
        body_truncated=truncated,
        links=links,
        include_text=include_text,
        include_metadata=include_metadata,
        include_links=include_links,
        verbosity=verbosity,
    )
    return md or "No content could be extracted."


# ==========================
# Websearch: DuckDuckGo tool
# ==========================

def Search_Structured(  # <-- MCP tool #2 (Structured DDG)
    input_query: str,
    max_results: int = 5,
) -> List[Dict[Literal["snippet", "title", "link"], str]]:
    """
    Run a DuckDuckGo search and return structured results as a list of dictionaries.

    Args:
        input_query (str): The search query. Supports operators like site:, quotes,
            and boolean keywords.
        max_results (int): Number of results to return (1–20).

    Returns:
        List[Dict[Literal["snippet","title","link"], str]]: Each item contains:
            - snippet: Short text snippet
            - title: Result title
            - link: Result URL
    """
    if not input_query or not input_query.strip():
        return []

    # Create the search tool (LangChain community wrapper)
    search = DuckDuckGoSearchResults(output_format="list", num_results=max_results)

    # Run the search and return results as a list of dicts
    results = search.invoke(input_query)
    return results


# ========================================
# Unstructured DDG: raw list into Textbox
# ========================================

def Search_Raw(  # <-- MCP tool #3 (Unstructured DDG)
    query: str,
) -> list[dict]:
    """
    Run a DuckDuckGo search using the native `duckduckgo_search` client and return the
    raw Python list of dictionaries from the library.

    Args:
        query (str): The search query string.

    Returns:
        list[dict]: The unmodified objects returned by `DDGS().text(...)`, typically
        containing keys like: title, href/link, body/snippet, source, etc.
    """
    if not query or not query.strip():
        return []
    with DDGS() as ddgs:
        results = ddgs.text(query, max_results=5)
    return results


# ============================================
# Concise DDG: ultra-succinct JSONL for tokens
# ============================================

def Search_Concise(  # <-- MCP tool #4 (Concise DDG)
    query: str,
    max_results: int = 5,
    include_snippets: bool = False,
    max_snippet_chars: int = 80,
    dedupe_domains: bool = True,
    title_chars: int = 80,
) -> str:
    """
    Run a DuckDuckGo search and return ultra-compact JSONL lines with short keys to
    minimize tokens.

    Args:
        query (str): The search query string.
        max_results (int): Maximum number of results to retrieve (1–20).
        include_snippets (bool): If True, include a shortened snippet per result under
            key "s".
        max_snippet_chars (int): Hard cap for snippet length when `include_snippets`
            is True. Range 20–200.
        dedupe_domains (bool): If True, only keep the first result per domain.
        title_chars (int): Hard cap for the title length. Range 20–120.

    Returns:
        str: Newline-delimited JSON (JSONL). Each line is a compact JSON object with
        short keys: "t" (title), "u" (URL), and optionally "s" (snippet).

        Example lines:
            {"t":"Example","u":"https://example.com/x"}
            {"t":"Another…","u":"https://a.com/y","s":"Short snippet…"}
    """
    
    if not query or not query.strip():
        return ""

    try:
        with DDGS() as ddgs:
            raw = ddgs.text(query, max_results=max_results)
    except Exception as e:
        
        return json.dumps({"error": str(e)[:120]}, ensure_ascii=False, separators=(",", ":"))

    seen_domains = set()
    lines: List[str] = []

    for r in raw or []:
        title = _shorten((r.get("title") or "").strip(), title_chars)
        url = (r.get("href") or r.get("link") or "").strip()
        body = (r.get("body") or r.get("snippet") or "").strip()

        if not url:
            continue

        if dedupe_domains:
            dom = _domain_of(url)
            if dom in seen_domains:
                continue
            seen_domains.add(dom)

        obj = {"t": title or _domain_of(url), "u": url}

        if include_snippets and body:
            obj["s"] = _shorten(body, max_snippet_chars)

        # Emit most compact JSON possible (no spaces)
        lines.append(json.dumps(obj, ensure_ascii=False, separators=(",", ":")))

    # Join as JSONL (each result on its own line)
    return "\n".join(lines)


# ======================
# UI: four-tab interface
# ======================

# --- Fetch tab (compact controllable extraction) ---
fetch_interface = gr.Interface(
    fn=Fetch_Webpage,  # connect the function to the UI
    inputs=[
        gr.Textbox(label="URL", placeholder="https://example.com/article"),
        gr.Dropdown(label="Verbosity", choices=["Brief", "Standard", "Full"], value="Standard"),
        gr.Checkbox(value=True, label="Include Metadata"),
        gr.Checkbox(value=True, label="Include Main Text"),
        gr.Checkbox(value=True, label="Include Links"),
        gr.Slider(400, 12000, value=3000, step=100, label="Max Characters (body text)"),
        gr.Slider(0, 100, value=20, step=1, label="Max Links"),
    ],
    outputs=gr.Markdown(label="Extracted Summary"),
    title="Fetch Webpage",
    description="Extract title, key metadata, readable text, and links. No noisy HTML.",
    api_description=(
        "Fetch a web page and return a compact Markdown summary with title, key "
        "metadata, readable body text, and outbound links. Parameters let you "
        "control verbosity, whether to include metadata/text/links, and limits "
        "for characters and number of links."
    ),
    allow_flagging="never",
    theme="Nymbo/Nymbo_Theme",
)

# --- Websearch tab (structured DDG via LangChain) ---
websearch_interface = gr.Interface(
    fn=Search_Structured,  # connect the function to the UI
    inputs=[
        gr.Textbox(value="", label="Search query", placeholder="site:example.com interesting topic"),
        gr.Slider(minimum=1, maximum=20, value=5, step=1, label="Max results"),
    ],
    outputs=gr.JSON(label="Search results"),
    title="DuckDuckGo Search (Structured)",
    description="Search the web using DuckDuckGo; returns snippet, title, and link.",
    api_description=(
        "Run a DuckDuckGo web search and return a list of objects with keys: "
        "snippet, title, and link. Configure the number of results."
    ),
    allow_flagging="never",
    theme="Nymbo/Nymbo_Theme",
)

# --- Unstructured DDG tab (matches your separate app’s output) ---
unstructured_interface = gr.Interface(
    fn=Search_Raw,
    inputs=gr.Textbox(label="Enter Search Query"),
    outputs=gr.Textbox(label="Results", interactive=False),
    title="DuckDuckGo Search (Raw)",
    description="Returns the raw list of results (list[dict]) shown as text.",
    api_description=(
        "Run DuckDuckGo via the native client and return the raw list[dict] as "
        "provided by duckduckgo_search (fields like title, href/link, body/snippet)."
    ),
    allow_flagging="never",
    theme="Nymbo/Nymbo_Theme",
    submit_btn="Search",
)

# --- Concise DDG tab (JSONL with short keys, minimal tokens) ---
concise_interface = gr.Interface(
    fn=Search_Concise,
    inputs=[
        gr.Textbox(label="Query", placeholder="topic OR site:example.com"),
        gr.Slider(minimum=1, maximum=20, value=5, step=1, label="Max results"),
        gr.Checkbox(value=False, label="Include snippets (adds tokens)"),
        gr.Slider(minimum=20, maximum=200, value=80, step=5, label="Max snippet chars"),
        gr.Checkbox(value=True, label="Dedupe by domain"),
        gr.Slider(minimum=20, maximum=120, value=80, step=5, label="Max title chars"),
    ],
    outputs=gr.Textbox(label="Results (JSONL)", interactive=False),
    title="DuckDuckGo Search (Concise)",
    description="Emits JSONL with short keys (t,u[,s]). Defaults avoid snippets and duplicate domains.",
    api_description=(
        "Run a DuckDuckGo search and return newline-delimited JSON with short keys: "
        "t=title, u=url, optional s=snippet. Options control result count, "
        "snippet inclusion and length, domain deduping, and title length."
    ),
    allow_flagging="never",
    theme="Nymbo/Nymbo_Theme",
    submit_btn="Search",
)

# --- Combine all into a single app with tabs ---
demo = gr.TabbedInterface(
    interface_list=[fetch_interface, websearch_interface, unstructured_interface, concise_interface],
    tab_names=[
        "Fetch Webpage",
        "DuckDuckGo Search (Structured)",
        "DuckDuckGo Search (Raw)",
        "DuckDuckGo Search (Concise)",
    ],
    title="Web MCP — Fetch & DuckDuckGo search with customizable output modes.",
    theme="Nymbo/Nymbo_Theme",
)

# Launch the UI and expose all functions as MCP tools in one server
if __name__ == "__main__":
    demo.launch(mcp_server=True)