forks/sorcery

git clone https://git.gae.moe/forks/sorcery

Charlotte Somdaemon: write catalog pages to disk743f5022026-11-02

use std::collections::HashSet;
use std::fs;
use std::path::{Component, Path, PathBuf};

use anyhow::{Context, Result};
use axum::extract::{OriginalUri, Path as AxumPath, State};
use axum::http::{HeaderValue, StatusCode, header};
use axum::response::{IntoResponse, Redirect, Response};
use include_dir::{Dir, include_dir};

use crate::catalog::Repository;
use crate::generate::{MANIFEST_FILE, STATE_FILE, STYLESHEET};
use crate::render::{catalog_index, encode_path, user_index};

use super::{AppState, WebResult, x_accel};

/// Bundle first: `cd web && deno task build`.
static WEB_DIST: Dir<'_> = include_dir!("$CARGO_MANIFEST_DIR/web/dist");
static FONTS: Dir<'_> = include_dir!("$CARGO_MANIFEST_DIR/assets/fonts");

/// Dot-prefixed so that discovery can never yield a colliding user or repo.
const CATALOG_DIR: &str = ".catalog";

pub(super) fn stage_assets(cache: &Path) -> Result<()> {
    // Embedded assets are served via X-Accel like everything else, so nginx
    // provides ETag/Last-Modified/304s.
    write_asset(&cache.join("css/style.css"), STYLESHEET.as_bytes())?;
    write_embedded_dir(&WEB_DIST, &cache.join("js"))?;
    write_embedded_dir(&FONTS, &cache.join("fonts"))
}

fn write_embedded_dir(dir: &Dir<'_>, out: &Path) -> Result<()> {
    for file in dir.files() {
        write_asset(&out.join(file.path()), file.contents())?;
    }
    for child in dir.dirs() {
        write_embedded_dir(child, out)?;
    }
    Ok(())
}

/// Rewrites the catalog index and per-user pages, removing pages for users
/// who no longer have any repositories.
pub(super) fn write_catalog(
    cache: &Path,
    instance_name: &str,
    repositories: &[Repository],
) -> Result<()> {
    let catalog = cache.join(CATALOG_DIR);
    write_asset(
        &catalog.join("index.html"),
        catalog_index(instance_name, repositories).as_bytes(),
    )?;
    let users = catalog.join("users");
    let mut live = HashSet::new();
    for repos in repositories.chunk_by(|a, b| a.user == b.user) {
        let file = format!("{}.html", repos[0].user);
        write_asset(
            &users.join(&file),
            user_index(instance_name, &repos[0].user, repos).as_bytes(),
        )?;
        live.insert(file);
    }
    for entry in fs::read_dir(&users).into_iter().flatten() {
        let entry = entry?;
        if !live.contains(entry.file_name().to_string_lossy().as_ref()) {
            fs::remove_file(entry.path())?;
        }
    }
    Ok(())
}

fn write_asset(path: &Path, content: &[u8]) -> Result<()> {
    if fs::read(path).ok().as_deref() == Some(content) {
        // bail if the file is unchanged so that we don't bump mtime
        // (which would upset mtime-derived ETags)
        return Ok(());
    }
    let parent = path.parent().expect("asset path has a parent");
    fs::create_dir_all(parent)?;
    let name = path.file_name().expect("asset path has a name");
    // write to staging and then rename over so nginx never caches a partially-written file
    let staging = parent.join(format!(".{}.new", name.to_string_lossy()));
    fs::write(&staging, content)?;
    fs::rename(&staging, path).with_context(|| format!("installing {}", path.display()))
}

pub(super) async fn index(State(state): State<AppState>) -> WebResult<Response> {
    serve_cache_path(&state, &format!("{CATALOG_DIR}/index.html"))
}

pub(super) async fn user_root(
    State(state): State<AppState>,
    AxumPath(user): AxumPath<String>,
) -> WebResult<Response> {
    if !state.snapshot.load().users.contains(&user) {
        return Err((StatusCode::NOT_FOUND, "not found".into()));
    }
    serve_cache_path(
        &state,
        &format!("{CATALOG_DIR}/users/{}.html", encode_path(&user)),
    )
}

/// Unlike repos, only redirect known users: this route also catches
/// top-level strays like `/favicon.ico`.
pub(super) async fn user_redirect(
    State(state): State<AppState>,
    OriginalUri(uri): OriginalUri,
    AxumPath(user): AxumPath<String>,
) -> WebResult<Redirect> {
    if !state.snapshot.load().users.contains(&user) {
        return Err((StatusCode::NOT_FOUND, "not found".into()));
    }
    Ok(add_trailing_slash(OriginalUri(uri)).await)
}

pub(super) async fn stylesheet(State(state): State<AppState>) -> WebResult<Response> {
    serve_cache_path(&state, "css/style.css")
}

pub(super) async fn js_asset(
    State(state): State<AppState>,
    AxumPath(path): AxumPath<String>,
) -> WebResult<Response> {
    embedded_asset(&state, &WEB_DIST, "js", &path)
}

pub(super) async fn font_asset(
    State(state): State<AppState>,
    AxumPath(path): AxumPath<String>,
) -> WebResult<Response> {
    embedded_asset(&state, &FONTS, "fonts", &path)
}

fn embedded_asset(
    state: &AppState,
    dir: &Dir<'_>,
    prefix: &str,
    path: &str,
) -> WebResult<Response> {
    if dir.get_file(path).is_none() {
        return Err((StatusCode::NOT_FOUND, "not found".into()));
    }
    serve_cache_path(state, &format!("{prefix}/{}", encode_path(path)))
}

pub(super) async fn add_trailing_slash(OriginalUri(uri): OriginalUri) -> Redirect {
    Redirect::permanent(&format!("{}/", uri.path()))
}

pub(super) async fn repo_root(
    State(state): State<AppState>,
    OriginalUri(uri): OriginalUri,
    AxumPath((user, repo)): AxumPath<(String, String)>,
) -> WebResult<Response> {
    serve_repo(state, user, repo, String::new(), uri.path().to_owned()).await
}

pub(super) async fn repo_path(
    State(state): State<AppState>,
    OriginalUri(uri): OriginalUri,
    AxumPath((user, repo, path)): AxumPath<(String, String, String)>,
) -> WebResult<Response> {
    serve_repo(state, user, repo, path, uri.path().to_owned()).await
}

async fn serve_repo(
    state: AppState,
    user: String,
    name: String,
    path: String,
    request_path: String,
) -> WebResult<Response> {
    let entry = state.resolve(&user, &name)?;
    let relative = requested_file(&path)?;
    let output = state.config.repo_cache(&entry.repository);
    if !output.join(STATE_FILE).is_file() {
        return Ok((
            StatusCode::SERVICE_UNAVAILABLE,
            [(header::RETRY_AFTER, HeaderValue::from_static("10"))],
            "this repository's site is still being generated",
        )
            .into_response());
    }

    let target = output.join(&relative);
    if target.is_dir() {
        return Ok(Redirect::permanent(&format!("{request_path}/")).into_response());
    }
    if !target.is_file() {
        return Err((StatusCode::NOT_FOUND, "page not found".into()));
    }

    serve_cache_path(
        &state,
        &format!(
            "{}/{}/{}",
            encode_path(&entry.repository.user),
            encode_path(&entry.repository.name),
            encode_path(&relative.to_string_lossy()),
        ),
    )
}

fn serve_cache_path(state: &AppState, encoded_path: &str) -> WebResult<Response> {
    x_accel(&state.config.internal_prefix, encoded_path)
}

fn requested_file(path: &str) -> WebResult<PathBuf> {
    if path.is_empty() {
        return Ok(PathBuf::from("index.html"));
    }
    if path == STATE_FILE || path == MANIFEST_FILE {
        return Err((StatusCode::NOT_FOUND, "page not found".into()));
    }
    let directory = path.ends_with('/');
    let path = Path::new(path);
    if path
        .components()
        .any(|component| !matches!(component, Component::Normal(_)))
    {
        return Err((StatusCode::BAD_REQUEST, "invalid path".into()));
    }
    Ok(if directory {
        path.join("index.html")
    } else {
        path.to_owned()
    })
}

#[cfg(test)]
mod tests {
    use crate::generate::{MANIFEST_FILE, STATE_FILE};

    use super::requested_file;

    #[test]
    fn generator_metadata_is_private() {
        assert!(requested_file(STATE_FILE).is_err());
        assert!(requested_file(MANIFEST_FILE).is_err());
    }
}