semora-cli 0.8.0

//! `.semora`: marks a folder as a Semora project, and where a push came from.
//!
//! The file is TOML meant to be committed:
//!
//! ```toml
//! server = "https://semora.example.com"
//! project = "semora"
//! ```
//!
//! `server` is only compared with the configured server, never used: a cloned
//! repository must not be able to send your API key to another host.

use std::path::{Component, Path, PathBuf};

use anyhow::{Context, bail};
use serde::Deserialize;

use crate::client::Client;
use crate::config::{origin, parse_server};

pub const FILE: &str = ".semora";

/// A path for messages: without the `\\?\` prefix that `canonicalize` adds on Windows.
pub fn shown(path: &Path) -> String {
    let s = path.display().to_string();
    match s.strip_prefix(r"\\?\") {
        Some(rest) => match rest.strip_prefix(r"UNC\") {
            Some(share) => format!(r"\\{share}"),
            None => rest.to_string(),
        },
        None => s,
    }
}

/// A `.semora` names another server than the one this command uses. Commands add their own
/// way out (push: `--id` or `--new`).
#[derive(Debug)]
pub struct OtherServer(pub String);

impl std::fmt::Display for OtherServer {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.write_str(&self.0)
    }
}

impl std::error::Error for OtherServer {}

#[derive(Debug, Clone, PartialEq, Eq)]
pub struct Link {
    /// The folder that holds `.semora` (canonical).
    pub root: PathBuf,
    /// The server's origin, normalized.
    pub server: String,
    pub slug: String,
}

#[derive(Deserialize)]
struct Raw {
    server: Option<String>,
    project: Option<String>,
}

/// `[a-z0-9-]`, 1 to 40 characters, not starting with a dash.
pub fn valid_slug(s: &str) -> bool {
    (1..=40).contains(&s.len())
        && s.bytes()
            .all(|b| b.is_ascii_lowercase() || b.is_ascii_digit() || b == b'-')
        && !s.starts_with('-')
}

/// A project id from a directory name: lowercase it, turn every run of other characters into
/// one `-`, trim `-` from both ends, cut to 40 characters, trim again.
pub fn slug_from(dir_name: &str) -> Option<String> {
    let mut out = String::new();
    for c in dir_name.to_lowercase().chars() {
        if c.is_ascii_lowercase() || c.is_ascii_digit() {
            out.push(c);
        } else if !out.ends_with('-') {
            out.push('-');
        }
    }
    let cut: String = out.trim_matches('-').chars().take(40).collect();
    let slug = cut.trim_matches('-').to_string();
    valid_slug(&slug).then_some(slug)
}

/// The `.semora` in `dir` itself, if there is one.
pub fn read_at(dir: &Path) -> anyhow::Result<Option<Link>> {
    let file = dir.join(FILE);
    if !file.is_file() {
        return Ok(None);
    }
    let text =
        std::fs::read_to_string(&file).with_context(|| format!("reading {}", shown(&file)))?;
    let raw: Raw = toml::from_str(&text).with_context(|| format!("parsing {}", shown(&file)))?;
    let server = raw
        .server
        .filter(|s| !s.trim().is_empty())
        .with_context(|| format!("{} needs a `server` line", shown(&file)))?;
    let slug = raw
        .project
        .filter(|s| !s.trim().is_empty())
        .with_context(|| format!("{} needs a `project` line", shown(&file)))?;
    if !valid_slug(&slug) {
        bail!(
            "{}: `{slug}` is not a valid project id (lowercase letters, digits and dashes, up to 40, starting with a letter or digit)",
            shown(&file)
        );
    }
    let server =
        parse_server(&server).with_context(|| format!("{}: bad `server`", shown(&file)))?;
    Ok(Some(Link {
        root: dir.to_path_buf(),
        server: origin(&server),
        slug,
    }))
}

/// The folder to start looking from: the path's folder, or the path itself for a folder.
fn start_dir(full: &Path) -> Option<&Path> {
    if full.is_dir() {
        Some(full)
    } else {
        full.parent()
    }
}

/// Looks for `.semora` in the path's folder and each parent. The nearest file wins. The
/// current directory plays no part. `Err` means a file was found but is unusable.
pub fn find(path: &Path) -> anyhow::Result<Option<Link>> {
    let full =
        std::fs::canonicalize(path).with_context(|| format!("cannot read {}", shown(path)))?;
    let mut dir = start_dir(&full);
    while let Some(d) = dir {
        if let Some(link) = read_at(d)? {
            return Ok(Some(link));
        }
        dir = d.parent();
    }
    Ok(None)
}

/// Writes `dir/.semora`. Returns its path.
pub fn write(dir: &Path, server: &str, slug: &str) -> anyhow::Result<PathBuf> {
    let file = dir.join(FILE);
    let text = format!(
        "# Semora project. Commit this file so every checkout publishes to the same place.\nserver = \"{server}\"\nproject = \"{slug}\"\n"
    );
    std::fs::write(&file, text).with_context(|| format!("writing {}", shown(&file)))?;
    Ok(file)
}

impl Link {
    /// The server named in `.semora` must be the one this command uses.
    pub fn check_origin(&self, have: &str) -> anyhow::Result<()> {
        if self.server == have {
            return Ok(());
        }
        Err(OtherServer(format!(
            "{} is for {}, but this command uses {have}. To use that server, set SEMORA_URL or pass --server {}.",
            shown(&self.root.join(FILE)),
            self.server,
            self.server
        ))
        .into())
    }

    pub fn check_server(&self, client: &Client) -> anyhow::Result<()> {
        self.check_origin(&client.origin())
    }

    /// The pushed path relative to the project root, joined with `/`.
    pub fn relative_path(&self, target: &Path) -> anyhow::Result<String> {
        let full = std::fs::canonicalize(target)
            .with_context(|| format!("cannot read {}", shown(target)))?;
        let rel = full.strip_prefix(&self.root).with_context(|| {
            format!(
                "{} is not inside the project at {}",
                shown(&full),
                shown(&self.root)
            )
        })?;
        let mut parts = Vec::new();
        for c in rel.components() {
            let s = c.as_os_str().to_str().with_context(|| {
                format!(
                    "{} isn't valid Unicode, so it can't be a project path",
                    shown(&full)
                )
            })?;
            parts.push(s.to_string());
        }
        if parts.is_empty() {
            bail!(
                "{} is here: push a file or a folder inside this project, not the project root.",
                shown(&self.root.join(FILE))
            );
        }
        Ok(parts.join("/"))
    }
}

/// The branch of the git checkout the path is in, read from `HEAD` (no `git` process).
/// A worktree's `.git` is a file that names its git folder; a relative name is relative to
/// the folder that holds that file. The walk starts from `std::path::absolute`, not
/// `canonicalize`: on Windows a canonical path is a `\\?\` path, and those do not resolve `..`
/// or `/`, so the relative name is joined and cleaned up lexically instead.
pub fn git_branch(path: &Path) -> Option<String> {
    let full = std::path::absolute(path).ok()?;
    let mut dir = start_dir(&full);
    while let Some(d) = dir {
        let dot = d.join(".git");
        if dot.is_dir() {
            return branch_in(&dot);
        }
        if dot.is_file() {
            let text = std::fs::read_to_string(&dot).ok()?;
            let target = text
                .lines()
                .find_map(|l| l.trim().strip_prefix("gitdir:"))?
                .trim();
            let gitdir = if Path::new(target).is_absolute() {
                PathBuf::from(target)
            } else {
                lexical(&d.join(target))
            };
            return branch_in(&gitdir);
        }
        dir = d.parent();
    }
    None
}

/// `.` and `..` resolved without touching the file system.
fn lexical(path: &Path) -> PathBuf {
    let mut out = PathBuf::new();
    for c in path.components() {
        match c {
            Component::CurDir => {}
            Component::ParentDir => {
                if !out.pop() {
                    out.push("..");
                }
            }
            other => out.push(other.as_os_str()),
        }
    }
    out
}

fn branch_in(gitdir: &Path) -> Option<String> {
    let head = std::fs::read_to_string(gitdir.join("HEAD")).ok()?;
    head.trim()
        .strip_prefix("ref: refs/heads/")
        .map(str::to_string)
        .filter(|b| !b.is_empty())
}

/// Where a push came from: `machine · branch`, or just the machine.
pub fn origin_for(path: &Path) -> String {
    let machine = crate::commands::login::hostname();
    match git_branch(path) {
        Some(b) => format!("{machine} \u{b7} {b}"),
        None => machine,
    }
}

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

    fn dir() -> tempfile::TempDir {
        tempfile::tempdir().unwrap()
    }

    fn put(root: &Path, rel: &str, text: &str) -> PathBuf {
        let p = root.join(rel);
        std::fs::create_dir_all(p.parent().unwrap()).unwrap();
        std::fs::write(&p, text).unwrap();
        p
    }

    #[test]
    fn slugs_follow_the_rule() {
        let (x40, x41) = ("x".repeat(40), "x".repeat(41));
        for ok in ["semora", "a", "9-a", x40.as_str()] {
            assert!(valid_slug(ok), "{ok}");
        }
        for bad in ["", "Semora", "-x", "a b", x41.as_str()] {
            assert!(!valid_slug(bad), "{bad}");
        }
    }

    #[test]
    fn a_slug_is_made_from_a_directory_name() {
        assert_eq!(slug_from("My Docs_v2").as_deref(), Some("my-docs-v2"));
        assert_eq!(slug_from(".tmpAbC123").as_deref(), Some("tmpabc123"));
        assert_eq!(slug_from("--a--b--").as_deref(), Some("a-b"));
        assert_eq!(slug_from("semora").as_deref(), Some("semora"));
        assert_eq!(slug_from(&"a".repeat(50)).unwrap().len(), 40);
        let a39 = "a".repeat(39);
        assert_eq!(
            slug_from(&format!("{a39}-b")).as_deref(),
            Some(a39.as_str()),
            "no dash left at the cut"
        );
        assert_eq!(slug_from("___"), None);
        assert_eq!(slug_from(""), None);
    }

    #[test]
    fn dot_semora_is_parsed_and_checked() {
        let d = dir();
        put(
            d.path(),
            ".semora",
            "# note\nserver = \"https://Semora.example.com/\"\nproject = \"semora\"\nextra = 1\n",
        );
        let link = read_at(d.path()).unwrap().unwrap();
        assert_eq!(link.slug, "semora");
        assert_eq!(
            link.server, "https://semora.example.com",
            "the origin, whatever the spelling"
        );

        for (text, why) in [
            ("project = \"x\"\n", "server"),
            ("server = \"https://a.example.com\"\n", "project"),
            (
                "server = \"https://a.example.com\"\nproject = \"Bad Id\"\n",
                "valid project id",
            ),
            ("server = \n", "parsing"),
        ] {
            put(d.path(), ".semora", text);
            let err = read_at(d.path()).unwrap_err().to_string();
            assert!(
                err.contains(why) && err.contains(".semora"),
                "{text:?} -> {err}"
            );
        }
        std::fs::remove_file(d.path().join(".semora")).unwrap();
        assert_eq!(read_at(d.path()).unwrap(), None);
    }

    #[test]
    fn find_walks_up_from_the_path_and_the_nearest_file_wins() {
        let d = dir();
        put(
            d.path(),
            ".semora",
            "server = \"https://a.example.com\"\nproject = \"outer\"\n",
        );
        put(
            d.path(),
            "inner/.semora",
            "server = \"https://a.example.com\"\nproject = \"inner\"\n",
        );
        let file = put(d.path(), "docs/deep/x.md", "# x");
        put(d.path(), "inner/y.md", "# y");
        assert_eq!(find(&file).unwrap().unwrap().slug, "outer");
        assert_eq!(
            find(&d.path().join("docs")).unwrap().unwrap().slug,
            "outer",
            "a folder"
        );
        assert_eq!(
            find(&d.path().join("inner/y.md")).unwrap().unwrap().slug,
            "inner"
        );
        assert_eq!(
            find(&d.path().join("inner")).unwrap().unwrap().slug,
            "inner",
            "the folder's own file"
        );
        let lonely = tempfile::tempdir().unwrap();
        let f = put(lonely.path(), "z.md", "# z");
        assert_eq!(find(&f).unwrap(), None);
    }

    #[test]
    fn the_key_path_is_relative_with_forward_slashes() {
        let d = dir();
        put(
            d.path(),
            ".semora",
            "server = \"https://a.example.com\"\nproject = \"p\"\n",
        );
        let file = put(d.path(), "docs/deep/x.md", "# x");
        let link = find(&file).unwrap().unwrap();
        assert_eq!(link.relative_path(&file).unwrap(), "docs/deep/x.md");
        assert_eq!(link.relative_path(&d.path().join("docs")).unwrap(), "docs");
        let err = link.relative_path(d.path()).unwrap_err().to_string();
        assert!(err.contains("not the project root"), "{err}");
    }

    #[test]
    fn a_link_for_another_server_names_both() {
        let d = dir();
        put(
            d.path(),
            ".semora",
            "server = \"https://a.example.com\"\nproject = \"p\"\n",
        );
        let link = read_at(d.path()).unwrap().unwrap();
        assert!(link.check_origin("https://a.example.com").is_ok());
        let err = link
            .check_origin("https://b.example.com")
            .unwrap_err()
            .to_string();
        assert!(
            err.contains("https://a.example.com") && err.contains("https://b.example.com"),
            "{err}"
        );
        assert!(
            err.contains("--server") && err.contains("SEMORA_URL"),
            "says how to use the other server: {err}"
        );
        assert!(
            !err.contains("--id") && !err.contains("--new"),
            "push flags are push's to mention, not every command's: {err}"
        );
    }

    #[test]
    fn write_makes_a_file_that_reads_back() {
        let d = dir();
        let path = write(d.path(), "https://a.example.com", "semora").unwrap();
        let text = std::fs::read_to_string(&path).unwrap();
        assert!(text.starts_with("# Semora project."), "{text}");
        let link = read_at(d.path()).unwrap().unwrap();
        assert_eq!(
            (link.server.as_str(), link.slug.as_str()),
            ("https://a.example.com", "semora")
        );
    }

    #[test]
    fn paths_in_messages_lose_the_windows_verbatim_prefix() {
        assert_eq!(
            shown(Path::new(r"\\?\C:\Development\x\.semora")),
            r"C:\Development\x\.semora"
        );
        assert_eq!(shown(Path::new(r"\\?\UNC\nas\share\x")), r"\\nas\share\x");
        assert_eq!(shown(Path::new("/home/a/x")), "/home/a/x");
    }

    #[test]
    fn dots_are_resolved_without_the_file_system() {
        assert_eq!(lexical(Path::new("/a/b/../c/./d")), PathBuf::from("/a/c/d"));
        assert_eq!(lexical(Path::new("a/../../b")), PathBuf::from("../b"));
    }

    #[test]
    fn the_branch_comes_from_head() {
        let d = dir();
        put(d.path(), ".git/HEAD", "ref: refs/heads/feat/projects\n");
        let file = put(d.path(), "docs/x.md", "# x");
        assert_eq!(git_branch(&file).as_deref(), Some("feat/projects"));
        put(
            d.path(),
            ".git/HEAD",
            "0123456789abcdef0123456789abcdef01234567\n",
        );
        assert_eq!(git_branch(&file), None, "a detached HEAD has no branch");
    }

    #[test]
    fn a_worktree_points_at_its_git_folder_absolute_or_relative() {
        let d = dir();
        let main = put(
            d.path(),
            "main/.git/worktrees/wt/HEAD",
            "ref: refs/heads/feat/wt\n",
        );
        let abs = main.parent().unwrap().to_string_lossy().replace('\\', "/");
        put(d.path(), "abs/.git", &format!("gitdir: {abs}\n"));
        let file = put(d.path(), "abs/x.md", "# x");
        assert_eq!(
            git_branch(&file).as_deref(),
            Some("feat/wt"),
            "absolute gitdir"
        );
        // Relative to the folder that holds the `.git` file, not to the current directory.
        put(d.path(), "rel/.git", "gitdir: ../main/.git/worktrees/wt\n");
        let file = put(d.path(), "rel/x.md", "# x");
        assert_eq!(
            git_branch(&file).as_deref(),
            Some("feat/wt"),
            "relative gitdir"
        );
    }

    #[test]
    fn a_verbatim_path_still_reads_a_relative_gitdir() {
        // `watch` passes a canonical path; on Windows that is a `\\?\` path, where `/` is not a
        // separator, so a relative `gitdir` must not be joined onto it as it is.
        let d = dir();
        put(
            d.path(),
            "main/.git/worktrees/wt/HEAD",
            "ref: refs/heads/feat/wt\n",
        );
        put(d.path(), "rel/.git", "gitdir: ../main/.git/worktrees/wt\n");
        let file = put(d.path(), "rel/x.md", "# x");
        let canonical = std::fs::canonicalize(&file).unwrap();
        if cfg!(windows) {
            assert!(
                canonical.to_string_lossy().starts_with(r"\\?\"),
                "{canonical:?}"
            );
        }
        assert_eq!(git_branch(&canonical).as_deref(), Some("feat/wt"));
    }

    #[test]
    fn origin_is_the_machine_and_the_branch_when_there_is_one() {
        let d = dir();
        let file = put(d.path(), "x.md", "# x");
        let machine = origin_for(&file);
        assert!(!machine.contains('\u{b7}'), "no branch: {machine}");
        put(d.path(), ".git/HEAD", "ref: refs/heads/main\n");
        let with = origin_for(&file);
        assert_eq!(with, format!("{machine} \u{b7} main"));
    }
}