tower-serve-embedded-build 0.2.0

//! Build-time helper for [`tower-serve-embedded`](https://docs.rs/tower-serve-embedded).
//!
//! Call this from your crate's `build.rs`. It walks an asset directory, content-hashes every file
//! with BLAKE3, and writes a generated Rust file to `OUT_DIR` containing the embedded manifest
//! (`ASSETS`) and a compile-time `asset!` macro. Pull that into your crate with
//! [`tower_serve_embedded::embed!()`](https://docs.rs/tower-serve-embedded).
//!
//! ```ignore
//! // build.rs
//! fn main() {
//!     tower_serve_embedded_build::Builder::new("assets").emit().unwrap();
//! }
//! ```
//!
//! Asset paths are expressed **relative to the crate root**, so the embedded URL mirrors the
//! file's location in your project. Ordinary files are served at a cache-busted URL
//! (`/assets/css/style.<hash>.css`). Files under [`Builder::immutable_dir`] are treated as already
//! versioned and served only at their plain URL (`/assets/lib/htmx-1.9.10.min.js`). Hidden files
//! and directories (names starting with `.`) and symlinks are ignored.

use std::env;
use std::fs;
use std::io;
use std::path::{Path, PathBuf};

/// Configures and runs asset embedding from a `build.rs`.
pub struct Builder {
    dir: PathBuf,
    hash_len: usize,
    immutable: Vec<String>,
}

impl Builder {
    /// Embed every file under `dir`, resolved relative to `CARGO_MANIFEST_DIR` (your crate root).
    ///
    /// Files keep their crate-root-relative paths, so embedding `assets/` makes
    /// `assets/css/style.css` available as `asset!("assets/css/style.css")`. Ordinary files are
    /// served at a cache-busted URL (`/assets/css/style.<hash>.css`). Files in an
    /// [`immutable_dir`] are served only at their plain URL (`/assets/lib/htmx-1.9.10.min.js`).
    /// The hash defaults to 16 hex chars.
    ///
    /// [`immutable_dir`]: Builder::immutable_dir
    pub fn new(dir: impl Into<PathBuf>) -> Self {
        Self {
            dir: dir.into(),
            hash_len: 16,
            immutable: Vec::new(),
        }
    }

    /// Mark a directory (and everything under it) as **already immutable**, given as a path
    /// **relative to the embedded directory** passed to [`Builder::new`].
    ///
    /// Files under it are still embedded and served, but at their original, non-hashed URL with a
    /// one-year `immutable` cache — they are *not* given an extra content hash. Use this for assets
    /// whose names already encode a version (vendored libraries, CDN-style
    /// `lib/htmx-1.9.10.min.js`), so `asset!("assets/lib/htmx-1.9.10.min.js")` resolves to
    /// `/assets/lib/htmx-1.9.10.min.js`. A leading or trailing slash is accepted (`"/lib"`,
    /// `"lib/"`); matching is by path prefix, not a glob. Call it multiple times for several
    /// directories.
    pub fn immutable_dir(mut self, dir: impl AsRef<str>) -> Self {
        let normalized = dir.as_ref().trim_matches('/').to_string();
        if !normalized.is_empty() {
            self.immutable.push(normalized);
        }
        self
    }

    /// Number of hex characters of the BLAKE3 hash to embed in filenames and ETags (default 16,
    /// i.e. 64 bits — clamped to 1..=64).
    pub fn hash_length(mut self, len: usize) -> Self {
        self.hash_len = len.clamp(1, 64);
        self
    }

    /// Walk the directory, hash the files, and write the generated code to `OUT_DIR`.
    ///
    /// Also emits `cargo:rerun-if-changed` lines so a content change refreshes the hashes and
    /// added/removed files are picked up — pair it with `cargo watch` for hot reload.
    pub fn emit(self) -> io::Result<()> {
        let manifest_dir = PathBuf::from(env_var("CARGO_MANIFEST_DIR")?);
        let root = manifest_dir.join(&self.dir);
        let out_dir = PathBuf::from(env_var("OUT_DIR")?);

        // Re-run if the build script itself changes (emitting any rerun-if-changed line opts out
        // of cargo's default "rerun if any package file changed" behaviour).
        println!("cargo:rerun-if-changed=build.rs");

        let mut files = Vec::new();
        let mut dirs = Vec::new();
        if root.is_dir() {
            collect(&root, &mut files, &mut dirs)?;
        } else {
            println!(
                "cargo:warning=tower-serve-embedded: asset directory {} not found",
                root.display()
            );
        }

        // rerun-if-changed for every directory (catches added/removed files) and every file
        // (catches content edits, which change the hash).
        for dir in &dirs {
            println!("cargo:rerun-if-changed={}", dir.display());
        }

        let mut assets: Vec<Asset> = Vec::with_capacity(files.len());
        for abs in &files {
            println!("cargo:rerun-if-changed={}", abs.display());
            let bytes = fs::read(abs)?;
            // Relative to the crate root, so the URL mirrors the project layout.
            let logical = logical_path(&manifest_dir, abs);
            // Relative to the embedded dir, for matching against `immutable_dir` entries.
            let rel = logical_path(&root, abs);
            let immutable = is_immutable(&rel, &self.immutable);
            let hash = hash_hex(&bytes, self.hash_len);
            let url = format!("/{logical}");
            // Already-immutable assets keep their plain URL; everything else gets a hashed alias.
            let hashed_url = if immutable {
                None
            } else {
                Some(hashed_path(&logical, &hash))
            };
            let content_type = mime_guess::from_path(abs)
                .first_or_octet_stream()
                .to_string();
            assets.push(Asset {
                abs: abs.clone(),
                logical,
                url,
                hashed_url,
                hash,
                content_type,
            });
        }

        // Stable, deterministic codegen order.
        assets.sort_by(|a, b| a.logical.cmp(&b.logical));

        let code = generate(&assets);
        fs::write(out_dir.join("embed_assets.rs"), code)?;
        Ok(())
    }
}

struct Asset {
    abs: PathBuf,
    logical: String,
    /// The plain, non-hashed URL, e.g. `/assets/css/style.css`. This is served only for assets in
    /// an `immutable_dir`.
    url: String,
    /// The cache-busted served URL, e.g. `/assets/css/style.<hash>.css`. `None` for assets in an
    /// `immutable_dir`, which are served only at `url`.
    hashed_url: Option<String>,
    hash: String,
    content_type: String,
}

/// Whether `rel` (a path relative to the embedded dir) is inside one of the `immutable` dirs —
/// either the dir itself or a descendant (prefix match on path components).
fn is_immutable(rel: &str, immutable: &[String]) -> bool {
    immutable
        .iter()
        .any(|i| rel == i || rel.starts_with(&format!("{i}/")))
}

fn generate(assets: &[Asset]) -> String {
    let mut out = String::new();
    out.push_str("// @generated by tower-serve-embedded-build. Do not edit.\n");

    out.push_str("#[doc(hidden)]\n");
    out.push_str("static __TSE_FILES: &[::tower_serve_embedded::EmbeddedFile] = &[\n");
    for a in assets {
        let etag = format!("\"{}\"", a.hash);
        out.push_str("    ::tower_serve_embedded::EmbeddedFile {\n");
        out.push_str(&format!("        url: {},\n", lit(&a.url)));
        out.push_str(&format!(
            "        hashed_url: {},\n",
            opt_lit(&a.hashed_url)
        ));
        out.push_str(&format!("        logical_path: {},\n", lit(&a.logical)));
        out.push_str(&format!(
            "        bytes: ::core::include_bytes!({}),\n",
            lit(&a.abs.to_string_lossy())
        ));
        out.push_str(&format!(
            "        content_type: {},\n",
            lit(&a.content_type)
        ));
        out.push_str(&format!("        etag: {},\n", lit(&etag)));
        out.push_str(&format!("        hash: {},\n", lit(&a.hash)));
        out.push_str("    },\n");
    }
    out.push_str("];\n\n");

    // Every served URL → (file index, Cache-Control), sorted by URL for binary search. Hashable
    // assets serve only their cache-busted alias; already-immutable assets serve only their plain
    // URL.
    let immutable = "::core::option::Option::Some(::tower_serve_embedded::IMMUTABLE_CACHE_CONTROL)";
    let mut routes: Vec<(String, usize, &str)> = Vec::with_capacity(assets.len());
    for (i, a) in assets.iter().enumerate() {
        match &a.hashed_url {
            // Hashable asset: only the hashed URL is served, immutable.
            Some(hashed) => routes.push((hashed.clone(), i, immutable)),
            // Already-immutable asset: served only at its plain URL, immutable.
            None => routes.push((a.url.clone(), i, immutable)),
        }
    }
    routes.sort_by(|a, b| a.0.cmp(&b.0));

    out.push_str("#[doc(hidden)]\n");
    out.push_str("static __TSE_ROUTES: &[::tower_serve_embedded::Route] = &[\n");
    for (url, index, cache_control) in &routes {
        out.push_str("    ::tower_serve_embedded::Route {\n");
        out.push_str(&format!("        url: {},\n", lit(url)));
        out.push_str(&format!("        file: {index}usize,\n"));
        out.push_str(&format!("        cache_control: {cache_control},\n"));
        out.push_str("    },\n");
    }
    out.push_str("];\n\n");

    out.push_str(
        "/// Assets embedded at build time by `tower-serve-embedded`.\n\
         pub static ASSETS: ::tower_serve_embedded::Assets =\n    \
         ::tower_serve_embedded::Assets::new(__TSE_FILES, __TSE_ROUTES);\n\n",
    );

    // A compile-time map from crate-root-relative path to the URL to reference it by (the
    // cache-busted alias when present, otherwise the plain URL). Unknown names are a compile error.
    out.push_str(
        "/// Resolve a crate-root-relative asset path to its served URL at compile time.\n",
    );
    out.push_str("#[doc(hidden)]\n");
    out.push_str("macro_rules! __tower_serve_embedded_asset {\n");
    for a in assets {
        let referenced = a.hashed_url.as_deref().unwrap_or(&a.url);
        out.push_str(&format!(
            "    ({}) => {{ {} }};\n",
            lit(&a.logical),
            lit(referenced)
        ));
    }
    out.push_str(
        "    ($other:literal) => {\n        \
         ::core::compile_error!(::core::concat!(\"tower-serve-embedded: unknown asset `\", $other, \"`\"))\n    \
         };\n",
    );
    out.push_str("}\n");
    out.push_str("#[doc(hidden)]\n");
    out.push_str("pub(crate) use __tower_serve_embedded_asset as asset;\n");

    out
}

/// Recursively collect files (into `files`) and directories (into `dirs`), skipping dotfiles and
/// symlinks. Entries are visited in sorted order for deterministic output.
fn collect(dir: &Path, files: &mut Vec<PathBuf>, dirs: &mut Vec<PathBuf>) -> io::Result<()> {
    dirs.push(dir.to_path_buf());
    let mut entries: Vec<_> = fs::read_dir(dir)?.collect::<Result<_, _>>()?;
    entries.sort_by_key(|e| e.file_name());
    for entry in entries {
        if entry.file_name().to_string_lossy().starts_with('.') {
            continue;
        }
        let file_type = entry.file_type()?;
        let path = entry.path();
        if file_type.is_dir() {
            collect(&path, files, dirs)?;
        } else if file_type.is_file() {
            files.push(path);
        }
    }
    Ok(())
}

/// The path of `file` relative to `base`, using `/` separators (e.g. `assets/css/style.css`).
fn logical_path(base: &Path, file: &Path) -> String {
    file.strip_prefix(base)
        .unwrap_or(file)
        .components()
        .map(|c| c.as_os_str().to_string_lossy())
        .collect::<Vec<_>>()
        .join("/")
}

/// Insert `hash` before the extension and prepend a leading slash:
/// `assets/css/style.css` + `9f3a1c2b` → `/assets/css/style.9f3a1c2b.css`.
fn hashed_path(logical: &str, hash: &str) -> String {
    let (dir, file) = match logical.rsplit_once('/') {
        Some((d, f)) => (Some(d), f),
        None => (None, logical),
    };
    let hashed_file = match file.rsplit_once('.') {
        Some((stem, ext)) if !stem.is_empty() => format!("{stem}.{hash}.{ext}"),
        _ => format!("{file}.{hash}"),
    };
    match dir {
        Some(d) => format!("/{d}/{hashed_file}"),
        None => format!("/{hashed_file}"),
    }
}

fn hash_hex(bytes: &[u8], len: usize) -> String {
    let full = blake3::hash(bytes).to_hex();
    full[..len.min(full.len())].to_string()
}

/// Render `s` as a valid Rust string literal (handles quotes, backslashes, etc.).
fn lit(s: &str) -> String {
    format!("{s:?}")
}

/// Render an optional string as a Rust `Option<&'static str>` expression.
fn opt_lit(s: &Option<String>) -> String {
    match s {
        Some(s) => format!("::core::option::Option::Some({})", lit(s)),
        None => "::core::option::Option::None".to_string(),
    }
}

fn env_var(key: &str) -> io::Result<String> {
    env::var(key).map_err(|_| {
        io::Error::new(
            io::ErrorKind::NotFound,
            format!("environment variable {key} is not set (is this running from build.rs?)"),
        )
    })
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn hashed_path_inserts_hash_before_extension() {
        assert_eq!(
            hashed_path("assets/css/style.css", "abcd"),
            "/assets/css/style.abcd.css"
        );
        assert_eq!(hashed_path("static/app.js", "abcd"), "/static/app.abcd.js");
        assert_eq!(hashed_path("a/b/c.png", "ff"), "/a/b/c.ff.png");
    }

    #[test]
    fn hashed_path_handles_no_extension_and_multi_dot() {
        assert_eq!(
            hashed_path("assets/LICENSE", "abcd"),
            "/assets/LICENSE.abcd"
        );
        assert_eq!(hashed_path("a.tar.gz", "ff"), "/a.tar.ff.gz");
    }

    #[test]
    fn hash_is_deterministic_and_truncated() {
        let a = hash_hex(b"hello world", 16);
        let b = hash_hex(b"hello world", 16);
        assert_eq!(a, b);
        assert_eq!(a.len(), 16);
        assert_ne!(hash_hex(b"hello world", 16), hash_hex(b"goodbye world", 16));
    }

    #[test]
    fn lit_escapes() {
        assert_eq!(lit("a\"b"), "\"a\\\"b\"");
    }

    #[test]
    fn collect_walks_every_dir_including_what_will_be_immutable() {
        let base =
            std::env::temp_dir().join(format!("tse_collect_{}_{}", std::process::id(), line!()));
        let _ = fs::remove_dir_all(&base);
        fs::create_dir_all(base.join("css")).unwrap();
        fs::create_dir_all(base.join("lib/sub")).unwrap();
        fs::write(base.join("css/a.css"), "a").unwrap();
        fs::write(base.join("root.txt"), "r").unwrap();
        fs::write(base.join("lib/b.js"), "b").unwrap();
        fs::write(base.join("lib/sub/c.js"), "c").unwrap();

        let mut files = Vec::new();
        let mut dirs = Vec::new();
        collect(&base, &mut files, &mut dirs).unwrap();

        // Everything is collected — nothing is skipped anymore.
        let mut logicals: Vec<String> = files.iter().map(|f| logical_path(&base, f)).collect();
        logicals.sort();
        assert_eq!(
            logicals,
            vec!["css/a.css", "lib/b.js", "lib/sub/c.js", "root.txt"]
        );
        // The "immutable" directory is walked and watched like any other.
        assert!(dirs.iter().any(|d| logical_path(&base, d) == "lib"));
        assert!(dirs.iter().any(|d| logical_path(&base, d) == "lib/sub"));

        fs::remove_dir_all(&base).unwrap();
    }

    #[test]
    fn is_immutable_matches_dir_and_descendants_only() {
        let immutable = vec!["lib".to_string(), "vendor/pkg".to_string()];
        assert!(is_immutable("lib/htmx.js", &immutable));
        assert!(is_immutable("lib/sub/a.js", &immutable));
        assert!(is_immutable("vendor/pkg/x.css", &immutable));
        // Not under an immutable dir.
        assert!(!is_immutable("css/a.css", &immutable));
        assert!(!is_immutable("vendor/other.js", &immutable));
        // A prefix that isn't a path boundary must not match.
        assert!(!is_immutable("library/a.js", &immutable));
    }

    #[test]
    fn generated_routes_only_include_plain_urls_for_immutable_assets() {
        let assets = vec![
            Asset {
                abs: PathBuf::from("/tmp/assets/css/style.css"),
                logical: "assets/css/style.css".to_string(),
                url: "/assets/css/style.css".to_string(),
                hashed_url: Some("/assets/css/style.abcd.css".to_string()),
                hash: "abcd".to_string(),
                content_type: "text/css".to_string(),
            },
            Asset {
                abs: PathBuf::from("/tmp/assets/lib/htmx-1.9.10.min.js"),
                logical: "assets/lib/htmx-1.9.10.min.js".to_string(),
                url: "/assets/lib/htmx-1.9.10.min.js".to_string(),
                hashed_url: None,
                hash: "beef".to_string(),
                content_type: "text/javascript".to_string(),
            },
        ];

        let code = generate(&assets);
        let routes = code
            .split("static __TSE_ROUTES")
            .nth(1)
            .unwrap()
            .split("];")
            .next()
            .unwrap();

        assert!(routes.contains("url: \"/assets/css/style.abcd.css\""));
        assert!(!routes.contains("url: \"/assets/css/style.css\""));
        assert!(routes.contains("url: \"/assets/lib/htmx-1.9.10.min.js\""));
    }

    #[test]
    fn generated_asset_macro_is_not_macro_exported() {
        let assets = vec![Asset {
            abs: PathBuf::from("/tmp/assets/css/style.css"),
            logical: "assets/css/style.css".to_string(),
            url: "/assets/css/style.css".to_string(),
            hashed_url: Some("/assets/css/style.abcd.css".to_string()),
            hash: "abcd".to_string(),
            content_type: "text/css".to_string(),
        }];

        let code = generate(&assets);

        assert!(!code.contains("#[macro_export]"));
        assert!(code.contains("macro_rules! __tower_serve_embedded_asset"));
        assert!(code.contains("pub(crate) use __tower_serve_embedded_asset as asset;"));
    }
}