amont_runtime/hooks/

common.rs

//! Shared plumbing for the linter-orchestration hooks.
//!
//! Nine of them do the same four things: collect staged files of some kind,
//! bail out if there are none, resolve a tool, run it. In shell that was ~65
//! lines apiece, mostly duplicated; here it is a handful of helpers and each
//! hook keeps only what is actually specific to it.

use crate::git;
use crate::ui::{error_sign, valid_sign, warning_sign};
use std::path::Path;
use std::process::{Command, Stdio};
use std::sync::OnceLock;

/// Staged files, deletions excluded, whose name ends with one of `exts`.
/// The file set every check asks about, when it is not the staged one.
///
/// Set at most once, before any check runs, by `amont run --all-files`. A
/// process-level override rather than a parameter because a check's signature
/// is `(&[OsString])` — it never sees a `Ctx` — and threading a file set
/// through twenty of them to serve one mode would be a worse trade than a
/// value that is written once and read many times.
///
/// Same shape as `PushRefs`: read once, lent to every check that asks.
static OVERRIDE: OnceLock<Vec<String>> = OnceLock::new();

/// Set once the file set stops being the index.
///
/// `restage`'s own doc says what makes re-staging safe: the pre-commit stage
/// holds the unstaged changes aside, so the tree contains the staged content
/// and nothing else, and anything a formatter touched is by definition part of
/// this commit. `amont run --all-files` replaces the file set with every
/// tracked path — which is that precondition being FALSE.
///
/// With `amont.fix true`, every fixer's `restage(&files)` would then `git
/// add` everything in the working tree that differs from the index, turning a
/// read-only "does my tree pass" query into `git add .`. That is the hazard §2
/// of docs/index-fidelity-and-run-modes.md names.
///
/// The gate hangs off the OVERRIDE rather than off a flag threaded through
/// twenty check signatures, because the override IS the fact that matters. It
/// therefore covers built-ins and `manifest::External::run` (which consults
/// `fixing_enabled` in two places) in one change, and a future check cannot
/// forget it.
static NOT_THE_INDEX: std::sync::atomic::AtomicBool = std::sync::atomic::AtomicBool::new(false);

/// Make every subsequent `staged_files` answer from `files` instead of the
/// index. Only the first call counts.
pub fn override_file_set(files: Vec<String>) {
    // Set unconditionally, even if a set already won the `OnceLock`: the
    // statement "the file set is not the index" is true from the first call
    // onwards regardless of which one supplied the paths.
    NOT_THE_INDEX.store(true, std::sync::atomic::Ordering::SeqCst);
    let _ = OVERRIDE.set(files);
}

/// Whether the file set every check sees is something other than the index.
pub fn not_the_index() -> bool {
    NOT_THE_INDEX.load(std::sync::atomic::Ordering::SeqCst)
}

/// An empty `exts` returns them all.
///
/// The UNFILTERED list is read from git ONCE per process and lent to every
/// caller — the per-stage snapshot. Eleven of the pre-commit checks ask this
/// question, concurrently, and each used to pay its own `git diff` spawn for
/// an answer that cannot change while the stage runs: the index-fidelity
/// hold pins the tree, and a fixer's `restage()` re-adds only paths already
/// on this list. `PushRefs` ("read once and lent") and `Overrides` ("ONE
/// subprocess for the whole stage") are the same pattern; this was the last
/// hot question still answered per asker.
pub fn staged_files(exts: &[&str]) -> Vec<String> {
    if let Some(all) = OVERRIDE.get() {
        return all
            .iter()
            .filter(|f| exts.is_empty() || exts.iter().any(|e| f.ends_with(e)))
            .cloned()
            .collect();
    }
    static INDEX: OnceLock<Vec<String>> = OnceLock::new();
    INDEX
        .get_or_init(|| {
            match git::stdout_paths(&["diff", "--diff-filter=d", "--cached", "--name-only"]) {
                Some(files) => files,
                // The third member of a bug family (`repo_hooks`, the push
                // gates): git FAILING is not git answering "empty", and a
                // stage that judges an empty set on a git failure reports
                // clean having verified nothing. Say so — once, this cache
                // being the once — and still fail open: pre-commit's job is
                // never to block a commit over its own plumbing.
                None => {
                    warn(
                        "git would not list the staged files — the checks are judging \
                         an EMPTY set, not a verified one",
                    );
                    Vec::new()
                }
            }
        })
        .iter()
        .filter(|f| exts.is_empty() || exts.iter().any(|e| f.ends_with(e)))
        .cloned()
        .collect()
}

/// Every path the index holds — what the repository CARRIES, as opposed to
/// what the current change touches.
///
/// The one honest source for a [`Scope`](crate::check::Scope) opt-in marker.
/// [`staged_files`] answers a different question, and answering the opt-in one
/// with it makes a `+marker` row fire only when the marker itself is in the
/// change — which is never, in ordinary work.
///
/// `ls-files` reads the INDEX, not `HEAD`, so a marker being added by this very
/// commit already counts. A marker sitting untracked on disk does not, which is
/// the same rule the manifest itself lives by: commit it, or it is not real.
///
/// Fails OPEN, unlike `staged_files`, and the asymmetry is deliberate. There,
/// an empty list means the checks judge nothing and say so. Here, an empty list
/// would silently switch every gated check OFF — a check that has quietly never
/// run is the one failure this design is arranged against — so a git failure
/// reports the check as opted in and lets the command itself be the judge. A
/// command that then finds no project fails to spawn, which is `Unavailable`:
/// a warning, never a block.
/// `None` when git would not answer — which is NOT the same as an empty
/// repository, and the caller must not flatten the two. An empty `Vec` opts
/// every gated check OUT; `None` means "unverified", and the gate opts them IN.
pub fn tracked_files() -> Option<Vec<String>> {
    static TRACKED: OnceLock<Option<Vec<String>>> = OnceLock::new();
    TRACKED
        .get_or_init(|| match git::stdout_paths(&["ls-files"]) {
            Some(files) => Some(files),
            None => {
                warn(
                    "git would not list the repository's files — opt-in gated checks \
                     will run rather than be skipped on an unverified answer",
                );
                None
            }
        })
        .clone()
}

/// Repo root, or "." when git cannot say.
///
/// **For CHECK BODIES ONLY.** The fallback is safe there and nowhere else: git
/// invokes a hook with the working tree as the current directory, so a check
/// that reaches this line is already standing in the repository, and "." is the
/// right answer rather than a guess.
///
/// Anything a user types — `amont agents-md`, `install`, `trust`, `restore`
/// — can be typed from any directory on the machine, and there the fallback is
/// not a fallback but a wrong answer that reads as a right one. Use
/// [`repo_root_checked`] at every command entry point.
pub fn repo_root() -> String {
    // Cached: the answer is a property of the process's repository, and
    // every check asked it through its own subprocess.
    static ROOT: OnceLock<String> = OnceLock::new();
    ROOT.get_or_init(|| {
        git::stdout(&["rev-parse", "--show-toplevel"]).unwrap_or_else(|| ".".into())
    })
    .clone()
}

/// Repo root, or an error naming the problem.
///
/// The same question as [`repo_root`] without the "." — because "." is a
/// PLAUSIBLE root, and that is what made it dangerous. `amont agents-md`
/// run outside a repository did not fail; it resolved the root to the current
/// directory and wrote `./AGENTS.md` into whatever directory the user happened
/// to be standing in, then printed `wrote ./AGENTS.md` as if that were the
/// answer. Same shape in `install`'s two prompts, in `trust` (which then
/// looked for a manifest, and would have recorded trust, under `.`) and in
/// `restore`.
///
/// Every one of those is a command somebody types, and a command somebody
/// types is a command they can type from `~`. There is no correct behaviour
/// available to this function when git cannot answer, so it does not invent
/// one.
pub fn repo_root_checked() -> Result<String, String> {
    git::stdout(&["rev-parse", "--show-toplevel"])
        .filter(|s| !s.is_empty())
        .ok_or_else(|| "not inside a git repository".to_string())
}

/// Resolve a tool, preferring the repo's PINNED copy so the hook matches CI.
///
///
/// Order: `<root>/node_modules/.bin/<tool>`, then the MAIN worktree's (a linked
/// worktree has no node_modules of its own — this is why the shell version
/// consulted the git common dir), then PATH.
pub fn resolve_tool(root: &str, tool: &str) -> Option<Vec<String>> {
    // Same extension problem as `which`: an npm-installed binary is `eslint.cmd`
    // on Windows, so the bare name misses the repo's PINNED copy and the hook
    // silently falls through to an ambient one.
    if let Some(p) = in_bin_dir(&format!("{root}/node_modules/.bin"), tool) {
        return Some(vec![p]);
    }
    if let Some(common) = git::stdout(&["rev-parse", "--path-format=absolute", "--git-common-dir"])
    {
        if let Some(main) = Path::new(&common).parent() {
            if let Some(p) = in_bin_dir(&main.join("node_modules/.bin").to_string_lossy(), tool) {
                return Some(vec![p]);
            }
        }
    }
    if let Some(full) = which(tool) {
        return Some(vec![full]);
    }
    // `npx --no-install`: never silently download a random latest version — a
    // hook that quietly pulls a different linter than CI uses is worse than one
    // that skips.
    if which("npx").is_some()
        && Command::new(program("npx"))
            .args(["--no-install", tool, "--version"])
            .current_dir(root)
            .stdin(Stdio::null())
            .stdout(Stdio::null())
            .stderr(Stdio::null())
            .status()
            .map(|s| s.success())
            .unwrap_or(false)
    {
        return Some(vec![
            program("npx"),
            "--no-install".to_string(),
            tool.to_string(),
        ]);
    }
    None
}

/// First match for `tool` on PATH.
///
/// Windows executables carry an extension — `git` is `git.exe`, an npm-installed
/// `eslint` is `eslint.cmd` — so the bare name finds nothing there. PATHEXT is
/// the OS's own list of what counts as executable; fall back to the usual set
/// when it is unset. Found by the Windows CI job on its first run, where
/// `which("git")` returned None on a machine that plainly has git.
pub fn which(tool: &str) -> Option<String> {
    which_on(&std::env::var_os("PATH")?, tool)
}

/// [`which`] against an EXPLICIT path list — the seam its own test needs.
///
/// The test that pins the Windows extension order used to `set_var("PATH")`
/// around the call, which is process-global: for the length of that call
/// every OTHER test in the binary — 340 of them, running in parallel, many
/// spawning git — had a PATH containing one fake tool and nothing else. A
/// git spawned in that window fails with "not found", which is not a
/// transient `git::retrying` may retry (correctly: it is a hard error), so
/// the caller reads it as git's ANSWER. In `gate_stamp` that answer is
/// "nothing is stamped". Passing the path in deletes the shared state
/// rather than guarding it — a lock only protects the callers who remember
/// to take it, and every future test here would have to remember.
pub fn which_on(path: &std::ffi::OsStr, tool: &str) -> Option<String> {
    let exts: Vec<String> = if cfg!(windows) {
        std::env::var("PATHEXT")
            .unwrap_or_else(|_| ".COM;.EXE;.BAT;.CMD".into())
            .split(';')
            .filter(|e| !e.is_empty())
            .map(|e| e.to_lowercase())
            .collect()
    } else {
        Vec::new()
    };
    for dir in std::env::split_paths(path) {
        // On Windows the EXTENSION forms come first. A node install ships both
        // `npm` (an extensionless shell script, for MSYS) and `npm.cmd` in the
        // same directory; preferring the bare name hands CreateProcess a shell
        // script it cannot execute — "%1 is not a valid Win32 application" —
        // and the hook reports an installed tool as broken.
        for e in &exts {
            let c = dir.join(format!("{tool}{e}"));
            if c.is_file() {
                return Some(c.to_string_lossy().into_owned());
            }
        }
        let bare = dir.join(tool);
        if bare.is_file() {
            return Some(bare.to_string_lossy().into_owned());
        }
    }
    None
}

/// `<dir>/<tool>`, trying the Windows executable extensions too.
fn in_bin_dir(dir: &str, tool: &str) -> Option<String> {
    let bare = Path::new(dir).join(tool);
    if bare.is_file() {
        return Some(bare.to_string_lossy().into_owned());
    }
    if cfg!(windows) {
        for e in [".cmd", ".exe", ".bat", ".ps1"] {
            let c = Path::new(dir).join(format!("{tool}{e}"));
            if c.is_file() {
                return Some(c.to_string_lossy().into_owned());
            }
        }
    }
    None
}

/// Resolve a tool name to a full path for spawning.
///
/// `Command::new("npm")` cannot execute `npm.cmd`: Rust does no PATHEXT
/// resolution, so on Windows every bare-name spawn fails with "program not
/// found" and the hook reports the tool as broken rather than absent. Found by
/// the Windows job on its first FULL-suite run — the smoke never spawned a
/// tool, so it could not have surfaced this.
///
/// Falls back to the name unchanged, so a caller still gets a sensible error.
pub fn program(name: &str) -> String {
    which(name).unwrap_or_else(|| name.to_string())
}

/// The first of `names` that exists at the repo root — how these hooks decide
/// a repo has opted into a tool.
pub fn first_existing(root: &str, names: &[&str]) -> Option<String> {
    names
        .iter()
        .find(|n| Path::new(root).join(n).exists())
        .map(|n| (*n).to_string())
}

/// Strip git's own environment before handing a Command to another tool.
///
/// git exports GIT_DIR, GIT_INDEX_FILE, GIT_WORK_TREE and friends to every
/// hook. Those OVERRIDE the working directory, so any tool that shells out to
/// git operates on the hook's repository no matter where it was launched.
///
/// That is not hypothetical: `pre-push-cargo-test` runs a project's test suite,
/// and this repo's own suite creates throwaway repos and commits to them. With
/// GIT_DIR inherited, `git commit` in a test wrote into the REAL repository —
/// an actual stray commit, authored by the test fixture, pushed to a branch.
///
/// A test suite should behave exactly as it does when run by hand, which means
/// seeing no git environment at all.
pub fn strip_git_env(cmd: &mut Command) {
    for (k, _) in std::env::vars_os() {
        let key = k.to_string_lossy();
        if key.starts_with("GIT_") {
            cmd.env_remove(&k);
        }
    }
}

/// The wall-clock CEILING for one check's spawned command, in seconds.
///
/// `amont.timeout`, default 3600. This used to be 600 and to be the only
/// clock, which made it answer two different questions with one number: "is
/// this tool stuck?" and "is this suite slow?". A stuck tool is silent, and
/// [`idle_timeout`] catches it in minutes; what is left for the ceiling is
/// the tool that keeps printing and never finishes, which is rare enough to
/// afford an hour. `0` disables. Read once per process: twenty concurrent
/// checks must not each spawn a `git config` to learn the same number.
pub fn check_timeout(settings: &crate::config::Settings) -> u64 {
    *settings.timeout.get_or_init(|| {
        crate::config::integer_or(settings, "amont.timeout", 3600, 0..=86_400) as u64
    })
}

/// The SILENCE budget: how long a spawned command may go without writing a
/// byte before it is judged stuck, in seconds.
///
/// `amont.idleTimeout`, default 120. A hang is silent; a slow test suite
/// talks — `cargo test` prints a line per test. Killing on silence catches
/// the captive portal, the deadlocked lock file and the tool waiting on a
/// prompt nobody will answer FASTER than a ten-minute wall clock did, while
/// letting a chatty twenty-five-minute suite finish. Only applies where the
/// output is observed (the captured runners); a command inheriting the
/// terminal directly answers to the ceiling alone. `0` disables.
pub fn idle_timeout(settings: &crate::config::Settings) -> u64 {
    *settings.idle.get_or_init(|| {
        crate::config::integer_or(settings, "amont.idleTimeout", 120, 0..=86_400) as u64
    })
}

/// `secs` as people read it: `12s`, `8m12s`, `1h02m`.
pub fn human_secs(secs: u64) -> String {
    match secs {
        s if s < 60 => format!("{s}s"),
        s if s < 3600 => format!("{}m{:02}s", s / 60, s % 60),
        s => format!("{}h{:02}m", s / 3600, (s % 3600) / 60),
    }
}

/// When a spawned command last wrote a byte, shared between the reader
/// threads that see the bytes and the wait loop that judges the silence.
pub struct Activity {
    last: std::sync::Mutex<std::time::Instant>,
}

impl Activity {
    pub fn new() -> std::sync::Arc<Activity> {
        std::sync::Arc::new(Activity {
            last: std::sync::Mutex::new(std::time::Instant::now()),
        })
    }
    pub fn touch(&self) {
        *self.last.lock().unwrap_or_else(|p| p.into_inner()) = std::time::Instant::now();
    }
    pub fn quiet_for(&self) -> std::time::Duration {
        self.last
            .lock()
            .unwrap_or_else(|p| p.into_inner())
            .elapsed()
    }
}

/// The deadline for a network PROBE — an `ls-remote` asked before the real
/// work, not the work itself. Capped at 30s below [`check_timeout`]: a
/// probe answers in a second or two when the network is there at all, and
/// a healthy `amont.timeout` of ten minutes is sized for a test suite, not
/// for deciding whether the remote is reachable. Shrinking `amont.timeout`
/// below the cap shrinks this too, and `0` keeps meaning no deadline —
/// somebody who disabled the clock disabled all of it.
pub fn network_probe_budget(settings: &crate::config::Settings) -> u64 {
    match check_timeout(settings) {
        0 => 0,
        t => t.min(30),
    }
}

/// Which clock killed a command.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum Why {
    /// The wall-clock ceiling, `amont.timeout`, in seconds.
    Ceiling(u64),
    /// The silence budget, `amont.idleTimeout`, in seconds.
    Silence(u64),
}

/// A command killed by a clock — what happened, said with enough to tell
/// "slow" from "stuck", which is the whole reason there are two clocks.
#[derive(Debug, Clone, Copy)]
pub struct Killed {
    pub why: Why,
    /// How long it had been running.
    pub ran_secs: u64,
    /// How long since its last output; `None` when the output was not ours
    /// to observe (inherited stdio).
    pub quiet_secs: Option<u64>,
}

/// What became of a command run under the deadline.
pub enum Ran {
    Status(std::process::ExitStatus),
    /// Killed by a clock; see [`Killed`].
    TimedOut(Killed),
}

/// `cmd.status()`, bounded by [`check_timeout`].
///
/// Without a bound, one hung tool — a linter deadlocked on a lock file, a
/// plugin doing network I/O — blocked the commit FOREVER, and it hung inside
/// the index-fidelity hold: the user's unstaged changes parked in `$GIT_DIR`,
/// their tree showing staged content only, for as long as they were willing
/// to wait. The learned response to that is `--no-verify`, permanently —
/// which disarms every check to escape one.
///
/// The kill reaches the direct child only. A grandchild that detached
/// survives, orphaned — but the COMMIT is no longer hostage to it, which is
/// the property that matters.
pub fn status_within(
    settings: &crate::config::Settings,
    cmd: &mut Command,
) -> std::io::Result<Ran> {
    status_within_secs(cmd, check_timeout(settings))
}

/// [`status_within`] with an explicit ceiling — the testable seam. The
/// output is inherited, so nobody sees the bytes and the silence budget
/// cannot apply; the ceiling is the only clock.
pub fn status_within_secs(cmd: &mut Command, budget_secs: u64) -> std::io::Result<Ran> {
    if budget_secs == 0 {
        return cmd.status().map(Ran::Status);
    }
    let mut child = cmd.spawn()?;
    wait_within(&mut child, budget_secs, 0, None)
}

/// Spawn `cmd` with both streams piped, hand every chunk to `on_output` as
/// it arrives, and wait under BOTH clocks — the reader threads are what
/// make the silence budget observable. The shared runner behind the
/// streamed, captured and discarded variants.
fn run_observed(
    settings: &crate::config::Settings,
    cmd: &mut Command,
    on_output: impl Fn(&[u8]) + Send + Sync + 'static,
) -> std::io::Result<Ran> {
    cmd.stdout(Stdio::piped()).stderr(Stdio::piped());
    let activity = Activity::new();
    let on_output = std::sync::Arc::new(on_output);
    let mut child = cmd.spawn()?;
    // The silence is the CHILD's, so its clock starts when the child does:
    // a spawn that itself took a second on a loaded machine is not a
    // second the tool spent saying nothing.
    activity.touch();
    let mut readers = Vec::new();
    for pipe in [
        child
            .stdout
            .take()
            .map(|p| Box::new(p) as Box<dyn std::io::Read + Send>),
        child
            .stderr
            .take()
            .map(|p| Box::new(p) as Box<dyn std::io::Read + Send>),
    ]
    .into_iter()
    .flatten()
    {
        let activity = std::sync::Arc::clone(&activity);
        let on_output = std::sync::Arc::clone(&on_output);
        readers.push(std::thread::spawn(move || {
            let mut pipe = pipe;
            let mut chunk = [0u8; 4096];
            loop {
                match std::io::Read::read(&mut pipe, &mut chunk) {
                    Ok(0) | Err(_) => break,
                    Ok(n) => {
                        activity.touch();
                        on_output(&chunk[..n]);
                    }
                }
            }
        }));
    }
    let ran = wait_within(
        &mut child,
        check_timeout(settings),
        idle_timeout(settings),
        Some(&activity),
    );
    for r in readers {
        let _ = r.join();
    }
    ran
}

/// [`status_within`], with the child's stdout and stderr CAPTURED into the
/// calling check's slot instead of inherited — the other half of one-check-
/// one-block: a linter's twelve lines used to land on the shared terminal
/// between two other checks' lines. Falls back to plain [`status_within`]
/// when no slot is installed on this thread (`amont.progress false`, or a
/// spawn outside a stage), which is byte-for-byte the old behaviour.
///
/// stdout and stderr merge in ARRIVAL order inside the block, which is what
/// the terminal showed before. The readers are threads, not processes, and
/// they are joined before the status is returned so a block can never grow
/// after its check finished.
pub fn status_streamed(
    settings: &crate::config::Settings,
    cmd: &mut Command,
) -> std::io::Result<Ran> {
    let Some((stage, idx)) = crate::live::current_sink() else {
        return status_within(settings, cmd);
    };
    if crate::live::watching() {
        // The block lands on a real terminal but the tool sees a pipe and
        // would strip its colors; the big three opt-in knobs put them back.
        cmd.env("FORCE_COLOR", "1")
            .env("CLICOLOR_FORCE", "1")
            .env("CARGO_TERM_COLOR", "always");
    }
    run_observed(settings, cmd, move |bytes| stage.append_raw(idx, bytes))
}

/// Run to completion under the `amont.timeout` deadline with stdout and
/// stderr CAPTURED into a string the caller can parse — what the audit
/// checks need: their verdict lives in the tool's output, not its exit
/// code alone. Arrival-ordered merge of both streams, like
/// [`status_streamed`]'s blocks. `None` when the child cannot be spawned.
pub fn capture_within(
    settings: &crate::config::Settings,
    cmd: &mut Command,
) -> Option<(Ran, String)> {
    let text = std::sync::Arc::new(std::sync::Mutex::new(String::new()));
    let sink = std::sync::Arc::clone(&text);
    let ran = run_observed(settings, cmd, move |bytes| {
        sink.lock()
            .unwrap_or_else(|p| p.into_inner())
            .push_str(&String::from_utf8_lossy(bytes));
    })
    .ok()?;
    let text = std::sync::Arc::try_unwrap(text)
        .map(|m| m.into_inner().unwrap_or_else(|p| p.into_inner()))
        .unwrap_or_default();
    Some((ran, text))
}

/// The two-clock wait over an already-spawned child — shared by every
/// runner. `wall_secs` is the ceiling, `idle_secs` the silence budget; each
/// `0` means that clock is off, and the silence budget is also off when
/// there is no [`Activity`] to consult (inherited stdio).
pub(crate) fn wait_within(
    child: &mut std::process::Child,
    wall_secs: u64,
    idle_secs: u64,
    activity: Option<&Activity>,
) -> std::io::Result<Ran> {
    let started = std::time::Instant::now();
    let ceiling = (wall_secs > 0).then(|| started + std::time::Duration::from_secs(wall_secs));
    let silence = match activity {
        Some(_) if idle_secs > 0 => Some(std::time::Duration::from_secs(idle_secs)),
        _ => None,
    };
    if ceiling.is_none() && silence.is_none() {
        return child.wait().map(Ran::Status);
    }
    loop {
        if let Some(status) = child.try_wait()? {
            return Ok(Ran::Status(status));
        }
        let now = std::time::Instant::now();
        let quiet = activity.map(|a| a.quiet_for());
        let why = judge(
            now.duration_since(started),
            quiet,
            ceiling.map(|_| wall_secs),
            silence.map(|_| idle_secs),
        );
        if let Some(why) = why {
            let _ = child.kill();
            let _ = child.wait();
            return Ok(Ran::TimedOut(Killed {
                why,
                ran_secs: now.duration_since(started).as_secs(),
                quiet_secs: quiet.map(|q| q.as_secs()),
            }));
        }
        std::thread::sleep(std::time::Duration::from_millis(25));
    }
}

/// Which clock, if any, has fired — the decision, with no process or
/// clock of its own so it can be tested to the second.
///
/// `ran` is how long the command has been running; `quiet` how long since
/// it last wrote, `None` when nobody is watching its output. `ceiling` and
/// `silence` are the two budgets in seconds, `None` when that clock is off.
/// The ceiling wins when both have fired: it is the larger claim, and the
/// message for it carries the silence figure anyway.
pub fn judge(
    ran: std::time::Duration,
    quiet: Option<std::time::Duration>,
    ceiling: Option<u64>,
    silence: Option<u64>,
) -> Option<Why> {
    if let Some(wall) = ceiling {
        if ran >= std::time::Duration::from_secs(wall) {
            return Some(Why::Ceiling(wall));
        }
    }
    if let (Some(idle), Some(q)) = (silence, quiet) {
        if q >= std::time::Duration::from_secs(idle) {
            return Some(Why::Silence(idle));
        }
    }
    None
}

/// Say a command was killed, by which clock, and what that tells you.
///
/// The two clocks exist to answer two different questions, so the message
/// answers the one that was asked: silence means stuck — look at the tool;
/// the ceiling with recent output means slow — raise the ceiling.
pub fn say_timed_out(what: &str, k: Killed) {
    match k.why {
        Why::Silence(budget) => fail(&format!(
            "{} printed nothing for {} and was killed after {} — a tool this quiet is \
             usually stuck, not slow. {} raises the silence budget (0 disables)",
            hl(what),
            human_secs(budget),
            human_secs(k.ran_secs),
            hl("git config amont.idleTimeout <secs>")
        )),
        Why::Ceiling(budget) => {
            let verdict = match k.quiet_secs {
                Some(q) if q < 30 => format!(
                    " It was still printing ({} since its last line): slow, not stuck.",
                    human_secs(q)
                ),
                Some(q) => format!(" Its last output was {} ago.", human_secs(q)),
                None => String::new(),
            };
            fail(&format!(
                "{} timed out: ran for {} and was killed at the ceiling. {} raises it \
                 (0 disables).{verdict}",
                hl(what),
                human_secs(budget),
                hl("git config amont.timeout <secs>")
            ))
        }
    }
}

/// [`status_within`], collapsed to "did it exit 0" — the shape the one-shot
/// tool spawns want. A timeout says so, names `what`, and reads as failure.
pub fn bounded_success(settings: &crate::config::Settings, cmd: &mut Command, what: &str) -> bool {
    match status_streamed(settings, cmd) {
        Ok(Ran::Status(s)) => s.success(),
        Ok(Ran::TimedOut(b)) => {
            say_timed_out(what, b);
            false
        }
        Err(_) => false,
    }
}

/// Run `argv` from `root`, inheriting stdio. True when it exits 0.
pub fn run(
    settings: &crate::config::Settings,
    root: &str,
    argv: &[String],
    extra: &[String],
) -> bool {
    let Some((program, rest)) = argv.split_first() else {
        return true;
    };
    let mut cmd = Command::new(program);
    cmd.args(rest)
        .args(extra)
        .current_dir(root)
        .stdin(Stdio::null());
    strip_git_env(&mut cmd);
    bounded_success(settings, &mut cmd, program)
}

/// As [`run`], but with the tool's own output discarded.
///
/// For a pass whose only job is to decide something — prettier's `--check`,
/// ruff's `--fix` sweep — where the offenders are printed once, by the pass
/// that reports them, rather than twice.
pub fn run_quiet(
    settings: &crate::config::Settings,
    root: &str,
    argv: &[String],
    extra: &[String],
) -> bool {
    let Some((program, rest)) = argv.split_first() else {
        return true;
    };
    let mut cmd = Command::new(program);
    cmd.args(rest)
        .args(extra)
        .current_dir(root)
        .stdin(Stdio::null());
    strip_git_env(&mut cmd);
    // Deliberately NOT the streamed runner: this helper's contract is that
    // the output is discarded, and capture would resurrect it into the
    // block. Observed and dropped instead of `/dev/null`, so the silence
    // clock still sees whether the tool is alive.
    match run_observed(settings, &mut cmd, |_| {}) {
        Ok(Ran::Status(s)) => s.success(),
        Ok(Ran::TimedOut(b)) => {
            say_timed_out(program, b);
            false
        }
        Err(_) => false,
    }
}

/// Whether the user asked for checks to repair what they find.
///
/// OFF by default. `git config amont.fix true` turns it on, per repository,
/// because a hook that edits your files without being asked is a larger
/// surprise than one that complains — and because with index fidelity in place
/// the repair lands in the commit you are making, which is a bigger claim to
/// make on somebody's behalf than printing an error.
pub fn fixing_enabled(settings: &crate::config::Settings) -> bool {
    // Never while the file set is not the index — see `NOT_THE_INDEX`.
    !not_the_index() && fixing_requested(settings)
}

/// What the CONFIG says, ignoring whether the current run may act on it.
///
/// Split out so `run_all` can tell the difference between "fixing is off" and
/// "you asked for fixing and this mode will not do it", and say the second out
/// loud instead of silently ignoring the key.
pub fn fixing_requested(settings: &crate::config::Settings) -> bool {
    *settings
        .fixing
        .get_or_init(|| crate::config::boolean_or(settings, "amont.fix", false))
}

/// What a re-stage actually did. THREE answers, because the old `bool`
/// conflated two of them and the conflation shipped unformatted code.
///
/// `prettier.rs` read `if run_quiet(write) && restage(&files) { … Fixed }`. When
/// `git add` FAILED, `restage` returned `false` — indistinguishable from
/// "nothing needed staging" — so control fell through to a second `--check`
/// pass, which inspected the NOW-FORMATTED WORKING TREE, passed, printed
/// "Prettier passed" and returned `Outcome::Passed`. The index still held the
/// unformatted content, so the commit contained unformatted code and the hook
/// said it had passed. `manifest.rs` had the same shape.
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum Restaged {
    /// No path differed from the index — nothing to do, and nothing wrong.
    Nothing,
    /// `git add` succeeded; the index now holds the repair.
    Staged,
    /// `git add` failed, carrying the paths it could not stage. The index
    /// holds content the fixer has already replaced on disk, so this MUST be
    /// loud at every call site — and naming the files is the difference
    /// between a message somebody can act on and one they cannot.
    Failed(Vec<String>),
}

/// Serialises this process's own `git add` calls.
///
/// pre-commit runs its checks concurrently (`dispatch.rs`), and up to three of
/// them can re-stage. git takes `$GIT_DIR/index.lock` exclusively, so two
/// concurrent `git add`s in the same repository make one of them fail — which,
/// before `Restaged`, was silently read as "nothing moved". Holding this across
/// the `git add` removes self-contention entirely; the retry below is only for
/// OTHER processes.
static INDEX_LOCK: std::sync::Mutex<()> = std::sync::Mutex::new(());

/// Re-stage exactly the paths a fixer rewrote, and say what happened.
///
/// Safe ONLY because the pre-commit stage holds unstaged changes aside: the
/// tree contains the staged content and nothing else, so anything a formatter
/// touched is by definition part of this commit. Without that, re-staging would
/// sweep in work the author deliberately kept back.
pub fn restage(paths: &[String]) -> Restaged {
    // Belt and braces alongside `fixing_enabled`: a future fixer that forgets
    // the gate still cannot turn `amont run --all-files` into `git add .`.
    if not_the_index() {
        return Restaged::Nothing;
    }
    let changed: Vec<String> = paths
        .iter()
        .filter(|p| !git::succeeds(&["diff", "--quiet", "--", p]))
        .cloned()
        .collect();
    if changed.is_empty() {
        return Restaged::Nothing;
    }
    let mut args = vec!["add", "--"];
    args.extend(changed.iter().map(String::as_str));

    let _serialised = INDEX_LOCK.lock().unwrap_or_else(|e| e.into_inner());
    // Another PROCESS can hold `index.lock` — a `git status` from an editor, a
    // second hook in a linked worktree. Back off and retry rather than
    // reporting a transient collision as a failed repair. `git add` of the same
    // paths is idempotent: it records the paths' current worktree content, so
    // running it twice records the same thing twice and cannot double-stage.
    const BACKOFF_MS: [u64; 3] = [50, 150, 400];
    if git::succeeds(&args) {
        return Restaged::Staged;
    }
    for wait in BACKOFF_MS {
        std::thread::sleep(std::time::Duration::from_millis(wait));
        if git::succeeds(&args) {
            return Restaged::Staged;
        }
    }
    Restaged::Failed(changed)
}

/// A check passed. THE funnel for every success line, which is what lets
/// `amont.quiet` swallow them in one place — see [`crate::live::quiet`].
pub fn ok(settings: &crate::config::Settings, msg: &str) {
    if crate::live::quiet(settings) {
        return;
    }
    crate::live::say(&format!("{} {msg}", valid_sign()));
}
pub fn fail(msg: &str) {
    crate::live::say(&format!("{} {msg}", error_sign()));
}
pub fn warn(msg: &str) {
    crate::live::say(&format!("{} {msg}", warning_sign()));
}
/// A line with no sign of its own — what a check's direct `println!` becomes,
/// so it lands in the check's block instead of interleaving. See `live::say`.
pub fn say(msg: &str) {
    crate::live::say(msg);
}

/// Orange, for the fragments these hooks highlight.
pub fn hl(s: &str) -> String {
    crate::ui::highlight(s)
}

#[cfg(test)]
mod tests {

    /// The two clocks, decided to the second. A chatty command outlives any
    /// silence budget however long it runs; a silent one dies at the budget
    /// however short; a command nobody watches answers to the ceiling only.
    #[test]
    fn the_clocks_judge_silence_and_ceiling_separately() {
        use super::{judge, Why};
        use std::time::Duration as D;
        let s = D::from_secs;
        // Chatty and long: past a five-second silence budget, still fine.
        assert_eq!(judge(s(900), Some(s(0)), Some(3600), Some(5)), None);
        assert_eq!(judge(s(900), Some(s(4)), Some(3600), Some(5)), None);
        // Silent for the budget: killed, and the silence is blamed.
        assert_eq!(
            judge(s(30), Some(s(5)), Some(3600), Some(5)),
            Some(Why::Silence(5))
        );
        // Unobserved output: the silence clock cannot run at all.
        assert_eq!(judge(s(900), None, Some(3600), Some(5)), None);
        // The ceiling fires on elapsed time whatever the output is doing.
        assert_eq!(
            judge(s(3600), Some(s(0)), Some(3600), Some(120)),
            Some(Why::Ceiling(3600))
        );
        // Both fired at once: the ceiling is the answer.
        assert_eq!(
            judge(s(3600), Some(s(600)), Some(3600), Some(120)),
            Some(Why::Ceiling(3600))
        );
        // Both off: nothing ever fires.
        assert_eq!(judge(s(86_400), Some(s(86_400)), None, None), None);
        // Only silence on: no ceiling, however long it runs.
        assert_eq!(judge(s(86_400), Some(s(1)), None, Some(120)), None);
    }

    /// The deadline kills what outlives it and reports what finished.
    #[cfg(unix)]
    #[test]
    fn the_deadline_kills_a_sleeper_and_spares_a_finisher() {
        let started = std::time::Instant::now();
        let mut slow = Command::new(program("sleep"));
        slow.arg("300").stdin(Stdio::null());
        match status_within_secs(&mut slow, 1) {
            Ok(Ran::TimedOut(Killed {
                why: Why::Ceiling(1),
                quiet_secs: None,
                ..
            })) => {}
            other => panic!("expected TimedOut(1), got {:?}", other.map(|_| "ran")),
        }
        assert!(
            started.elapsed() < std::time::Duration::from_secs(60),
            "the kill did not happen at the deadline"
        );

        let mut quick = Command::new(program("true"));
        quick.stdin(Stdio::null());
        match status_within_secs(&mut quick, 60) {
            Ok(Ran::Status(s)) => assert!(s.success()),
            other => panic!("expected a clean exit, got {:?}", other.map(|_| "?")),
        }
    }

    use super::*;

    #[test]
    fn which_finds_a_real_binary_and_not_a_fake_one() {
        assert!(which("git").is_some());
        assert!(which("definitely-not-a-real-binary-xyz").is_none());
    }

    /// On Windows a tool can exist BOTH as an extensionless shell script and as
    /// a .cmd/.exe in the same directory; only the latter is executable by
    /// CreateProcess, so the extension forms must win.
    #[test]
    #[cfg(windows)]
    fn windows_prefers_an_executable_extension_over_a_bare_file() {
        let dir = std::env::temp_dir().join("amont-which-order");
        let _ = std::fs::create_dir_all(&dir);
        std::fs::write(dir.join("faketool"), "#!/bin/sh\n").unwrap();
        std::fs::write(dir.join("faketool.cmd"), "@echo off\n").unwrap();
        // The path is PASSED, never installed into this process: see
        // `which_on`. The old spelling swapped the real PATH out from under
        // every other test in this binary for the length of the call.
        let found = which_on(dir.as_os_str(), "faketool").unwrap();
        assert!(found.ends_with(".cmd"), "got {found}");
        let _ = std::fs::remove_dir_all(&dir);
    }

    /// "Nothing moved" and "`git add` FAILED" are different answers, and the
    /// old `bool` gave the same one for both.
    ///
    /// That conflation is what shipped unformatted code: `prettier.rs` read
    /// `if wrote && restage(&files)`, so a failed `git add` fell through to a
    /// second `--check` against the now-formatted WORKING TREE, which passed —
    /// while the INDEX still held the unformatted content the commit would
    /// carry.
    ///
    /// An absolute path outside any repository is a `git add` git will always
    /// refuse, which is the only way to reach the failing branch without
    /// sabotaging a real index.
    #[test]
    fn restage_distinguishes_nothing_from_failure() {
        // `restage` runs `git add` in the PROCESS cwd, so this test depends
        // on that cwd as surely as one that moves it — see `crate::TEST_CWD`.
        // Without the lock it ran inside whatever fixture `gate_stamp` had
        // moved into, and took that repository's index.lock out from under
        // its own commit.
        let _cwd = crate::TEST_CWD.lock().unwrap_or_else(|p| p.into_inner());
        let outside = std::env::temp_dir()
            .join("amont-restage-outside-any-repo")
            .to_string_lossy()
            .into_owned();
        assert_eq!(
            restage(std::slice::from_ref(&outside)),
            Restaged::Failed(vec![outside]),
            "a `git add` git refuses must report Failed, never Nothing"
        );
        assert_eq!(
            restage(&[]),
            Restaged::Nothing,
            "no paths is nothing to do, and nothing wrong"
        );
    }

    /// No check may hand `Command` a bare program name.
    ///
    /// `Command::new` does NO PATHEXT resolution, so `Command::new("npm")`
    /// cannot execute `npm.cmd` and `Command::new("uvx")` cannot execute
    /// `uvx.exe`: the spawn fails with "program not found" and a
    /// `Severity::Block` check reports an installed tool as broken. That is the
    /// incident `program()` exists for, and it kept recurring — `yamllint` and
    /// three sites in `python_tools` were still doing it, THREE OF THEM after
    /// `which()` had already succeeded and discarded the answer.
    ///
    /// A source scan rather than a runtime assertion because the failure only
    /// reproduces on Windows, and the whole point is to catch the next one on
    /// every platform. Comment lines are skipped: `program()`'s own doc quotes
    /// the offending call. The needle is assembled from two pieces so this
    /// module — which the scan also reads — does not match itself.
    #[test]
    fn no_hook_spawns_a_bare_program_name() {
        let needle = concat!("Command", "::new(");
        let dir = concat!(env!("CARGO_MANIFEST_DIR"), "/src/hooks");
        let mut scanned = 0usize;
        for entry in std::fs::read_dir(dir).expect("hooks dir").flatten() {
            let path = entry.path();
            if path.extension().and_then(|e| e.to_str()) != Some("rs") {
                continue;
            }
            scanned += 1;
            let src = std::fs::read_to_string(&path).expect("read a hook module");
            for (n, line) in src.lines().enumerate() {
                if line.trim_start().starts_with("//") {
                    continue;
                }
                let Some(after) = line.split_once(needle) else {
                    continue;
                };
                assert!(
                    !after.1.starts_with('"'),
                    "{}:{} spawns a bare name — route it through `program()` or \
                     the path `which()` already resolved: {}",
                    path.display(),
                    n + 1,
                    line.trim()
                );
            }
        }
        assert!(
            scanned > 10,
            "the scan found almost nothing: {scanned} files"
        );
    }

    #[test]
    fn first_existing_picks_the_earliest_present_name() {
        let dir = std::env::temp_dir().join("amont-first-existing-test");
        let _ = std::fs::create_dir_all(&dir);
        let root = dir.to_string_lossy().into_owned();
        let _ = std::fs::write(dir.join("second"), "x");
        assert_eq!(
            first_existing(&root, &["first", "second", "third"]).as_deref(),
            Some("second")
        );
        assert_eq!(first_existing(&root, &["nope"]), None);
        let _ = std::fs::remove_dir_all(&dir);
    }
}