use anyhow::{anyhow, Context, Result};
use scraper::{Html, Selector};
use serde_json::Value;
use std::collections::HashMap;
use std::sync::RwLock;
use std::time::{Duration, Instant};

#[derive(Clone, Debug, Eq, Hash, PartialEq)]
struct TgLink {
    username: String,
    message_id: usize,
    translated: bool,
}

impl TgLink {
    fn human_friendly_url(&self) -> String {
        format!("t.me/{}/{}", self.username, self.message_id)
    }

    fn to_url(&self, embed: bool) -> String {
        format!(
            "https://t.me/{}/{}{}",
            self.username,
            self.message_id,
            if embed { "?embed=1" } else { "" }
        )
    }

    fn parse(url: &str, translated: bool) -> Option<Self> {
        let url = url.strip_prefix("/")?;
        let parsed = url::Url::parse(url).ok()?;

        if parsed.host()? != url::Host::Domain("t.me") {
            // only t.me links are supported
            return None;
        }

        let parts = parsed.path_segments()?.collect::<Vec<&str>>();
        if parts.len() != 2 {
            // only message links are supported
            return None;
        }

        Some(TgLink {
            username: parts[0].into(),
            message_id: parts[1].parse().ok()?,
            translated,
        })
    }
}

fn fetch_post(link: &TgLink, embed: bool) -> Result<String> {
    println!("fetching {}#{}", link.username, link.message_id);
    let response = crimp::Request::get(&link.to_url(embed))
        .send()
        .context("failed to fetch embed data")?
        .as_string()
        .context("failed to decode embed data")?
        .error_for_status(|resp| {
            anyhow!("telegram request failed: {} ({})", resp.body, resp.status)
        })?;

    Ok(response.body)
}

fn fetch_translation(message: &str) -> Result<String> {
    let request = serde_json::json!({
        "model": "gpt-3.5-turbo",
        "messages": [
            {"role": "user", "content": "Please translate the following message from a Telegram channel into English. If the post is already partially in English, please leave those bits intact as they are. Please respond only with the translation."},
            {"role": "user", "content": message}
        ]
    });

    let response: Value = crimp::Request::post("https://api.openai.com/v1/chat/completions")
        .bearer_auth(&std::env::var("OPENAPI_KEY").context("no openapi key set")?)?
        .json(&request)?
        .send()
        .context("failed to fetch translation from openai")?
        .as_json::<Value>()?
        .error_for_status(|resp| {
            anyhow!(
                "translation request failed: {} ({})",
                resp.body,
                resp.status
            )
        })?
        .body;

    // we want choices[0].message.content, and inshallah it's the right thing.
    let translation = response
        .get("choices")
        .ok_or_else(|| anyhow!("missing 'choices' key"))?
        .get(0)
        .ok_or_else(|| anyhow!("empty 'choices' or something"))?
        .get("message")
        .ok_or_else(|| anyhow!("missing 'message' key"))?
        .get("content")
        .ok_or_else(|| anyhow!("missing 'content' key"))?
        .as_str()
        .ok_or_else(|| anyhow!("'content' was not a string"))?;

    Ok(translation.to_string())
}

// in some cases, posts can not be embedded, but telegram still
// includes their content in metadata tags for content previews.
//
// we skip images in this case, as they are scaled down to thumbnail
// size and not useful.
fn fetch_fallback(link: &TgLink) -> Result<Option<String>> {
    let post = fetch_post(link, false)?;
    let doc = Html::parse_document(&post);
    let desc_sel = Selector::parse("meta[property=\"og:description\"]").unwrap();
    let desc_elem = match doc.select(&desc_sel).next() {
        None => return Ok(None),
        Some(elem) => elem,
    };

    let content = match desc_elem.value().attr("content") {
        None => return Ok(None),
        Some(content) => content.to_string(),
    };

    return Ok(Some(content));
}

#[derive(Debug)]
struct TgMessage {
    author: String,
    message: Option<String>,
    photos: Vec<String>,
    videos: Vec<String>,
    has_audio: bool,
}

fn extract_photo_url(style: &str) -> Option<&str> {
    let url_start = style.find("url('")? + 5;
    let url_end = style.find("')")?;

    Some(&style[url_start..url_end])
}

fn parse_tgmessage(embed: &str) -> Result<TgMessage> {
    let doc = Html::parse_document(embed);

    let author_sel = Selector::parse("a.tgme_widget_message_owner_name").unwrap();
    let author = doc
        .select(&author_sel)
        .next()
        .ok_or_else(|| anyhow!("failed to find message author"))?
        .text()
        .collect::<Vec<&str>>()
        .concat();

    let msg_sel = Selector::parse("div.tgme_widget_message_text.js-message_text").unwrap();

    // The ElementRef::text() iterator does not yield newlines present
    // in the message, so it is partially reimplemented here.
    let message = if let Some(msg_elem) = doc.select(&msg_sel).next() {
        use ego_tree::iter::Edge;
        use scraper::node::Node;

        let mut out = String::new();

        for edge in &mut msg_elem.traverse() {
            if let Edge::Open(node) = edge {
                match node.value() {
                    Node::Text(ref text) => out.push_str(&*text),
                    Node::Element(elem) if elem.name() == "br" => out.push_str("\n"),
                    _ => {}
                }
            }
        }

        Some(out)
    } else {
        // Not all Telegram messages have a textual message.
        None
    };

    let photo_sel = Selector::parse("a.tgme_widget_message_photo_wrap").unwrap();
    let mut photos = vec![];

    for photo in doc.select(&photo_sel) {
        if let Some(style) = photo.value().attr("style") {
            if let Some(url) = extract_photo_url(style) {
                photos.push(url.to_string())
            }
        }
    }

    let video_sel = Selector::parse("i.tgme_widget_message_video_thumb").unwrap();
    let mut videos = vec![];

    for video in doc.select(&video_sel) {
        if let Some(style) = video.value().attr("style") {
            if let Some(url) = extract_photo_url(style) {
                videos.push(url.to_string())
            }
        }
    }

    let audio_sel = Selector::parse("audio.tgme_widget_message_voice.js-message_voice").unwrap();
    let mut has_audio = false;
    if doc.select(&audio_sel).next().is_some() {
        has_audio = true;
    }

    Ok(TgMessage {
        author,
        message,
        photos,
        videos,
        has_audio,
    })
}

// create a permanent media url that tgsa can redirect if telegram
// changes its upstream links.
//
// assumes that tgsa lives at tgsa.tazj.in (which it does)
fn media_url(link: &TgLink, idx: usize) -> String {
    format!(
        "https://tgsa.tazj.in/img/{}/{}/{}",
        link.username, link.message_id, idx
    )
}

fn to_bbcode(link: &TgLink, msg: &TgMessage) -> String {
    let mut out = String::new();

    out.push_str(&format!("[quote=\"{}\"]\n", msg.author));

    for video in 0..msg.videos.len() {
        out.push_str(&format!("[url=\"{}\"]", link.to_url(true)));

        // video thumbnail links are appended to the photos, hence the
        // addition here
        out.push_str(&format!(
            "[img]{}[/img]",
            media_url(link, video + msg.photos.len())
        ));

        out.push_str("[/url]\n");
        out.push_str("[sub](Click thumbnail to open video)[/sub]\n")
    }

    for photo in 0..msg.photos.len() {
        out.push_str(&format!("[timg]{}[/timg]\n", media_url(link, photo)));
    }

    if msg.has_audio {
        out.push_str(&format!(
            "[i]This message has audio attached. Go [url=\"{}\"]to Telegram[/url] to listen.[/i]",
            link.to_url(true),
        ));
    }

    if let Some(message) = &msg.message {
        out.push_str(message);
    }

    out.push_str("\n[/quote]\n");

    out.push_str(&format!(
        "[sub](from [url=\"{}\"]{}[/url], via [url=\"https://tgsa.tazj.in\"]tgsa[/url])[/sub]\n",
        link.to_url(true),
        link.human_friendly_url(),
    ));

    out
}

// cache everything for one hour
const CACHE_EXPIRY: Duration = Duration::from_secs(60 * 60);

#[derive(Clone)]
struct TgPost {
    bbcode: String,
    at: Instant,
    media: Vec<String>,
}

type Cache = RwLock<HashMap<TgLink, TgPost>>;

fn fetch_with_cache(cache: &Cache, link: &TgLink) -> Result<TgPost> {
    if let Some(entry) = cache.read().unwrap().get(&link) {
        if Instant::now() - entry.at < CACHE_EXPIRY {
            println!("serving {}#{} from cache", link.username, link.message_id);
            return Ok(entry.clone());
        }
    }

    // limit concurrent fetching
    // TODO(tazjin): per link?
    let mut writer = cache.write().unwrap();

    let post = fetch_post(&link, true)?;
    let mut msg = parse_tgmessage(&post)?;

    if msg.message.is_none() {
        msg.message = fetch_fallback(&link)?;
    }

    if let Some(message) = &msg.message {
        if link.translated {
            msg.message = Some(fetch_translation(message)?);
        }
    }

    let bbcode = to_bbcode(&link, &msg);

    let mut media = vec![];
    media.append(&mut msg.photos);
    media.append(&mut msg.videos);

    let post = TgPost {
        bbcode,
        media,
        at: Instant::now(),
    };

    writer.insert(link.clone(), post.clone());

    Ok(post)
}

fn handle_img_redirect(cache: &Cache, img_path: &str) -> Result<rouille::Response> {
    // img_path:
    //
    // RWApodcast/113/1
    // ^          ^   ^
    // |          |   |
    // |          |   image (0-indexed)
    // |          post ID
    // username

    let img_parts: Vec<&str> = img_path.split("/").collect();

    if img_parts.len() != 3 {
        println!("invalid image link: {}", img_path);
        return Err(anyhow!("not a valid image link: {}", img_path));
    }

    let link = TgLink {
        username: img_parts[0].into(),
        message_id: img_parts[1].parse().context("failed to parse message_id")?,
        translated: false,
    };

    let img_idx: usize = img_parts[2].parse().context("failed to parse img_idx")?;
    let post = fetch_with_cache(cache, &link)?;

    if img_idx >= post.media.len() {
        return Err(anyhow!(
            "there is no {}. image in {}/{}",
            img_idx,
            link.username,
            link.message_id
        ));
    }

    Ok(rouille::Response::redirect_303(post.media[img_idx].clone()))
}

fn handle_tg_link(cache: &Cache, link: &TgLink) -> Result<rouille::Response> {
    let post = fetch_with_cache(cache, link)?;
    Ok(rouille::Response::text(post.bbcode))
}

fn main() {
    crimp::init();

    let cache: Cache = RwLock::new(HashMap::new());

    rouille::start_server("0.0.0.0:8472", move |request| {
        let mut raw_url = request.raw_url();
        let mut translate = false;

        let response = loop {
            if raw_url.starts_with("/img/") {
                break handle_img_redirect(&cache, &raw_url[5..]);
            }

            if raw_url.starts_with("/translate/") {
                translate = true;
                raw_url = &raw_url[10..];
            }

            break match TgLink::parse(raw_url, translate) {
                None => Ok(rouille::Response::text(
                    r#"tgsa
----

this is a stupid program that lets you turn telegram message links
into BBcode suitable for pasting on somethingawful dot com

you can use it by putting a valid telegram message link in the url and
waiting for some bbcode to show up.

for example:

  https://tgsa.tazj.in/https://t.me/RWApodcast/113

yes, that looks stupid, but it works

if you see this message and think you did the above correctly, you
didn't. try again. idiot.

it can also translate posts from russian, ukrainian or whatever other
dumb language you speak into english, by adding `/translate/`, for
example:

  https://tgsa.tazj.in/translate/https://t.me/strelkovii/4329

expect this to be slow though. that's the price to pay for translating
shitty slang.

pm me on the forums if any of this makes you mad or something.
"#,
                )),
                Some(link) => handle_tg_link(&cache, &link),
            };
        };

        match response {
            Ok(resp) => resp,
            Err(err) => {
                println!("something failed: {}", err);
                rouille::Response::text(format!(
                    r#"ugh, something broke: {}

nobody has been alerted about this and it has probably not been
logged. pm me on the forums if you think it's important."#,
                    err
                ))
            }
        }
    });
}