git_policy_cli/markdown/

markdown_linter_guard.rs

//! What:
//!  Run one job on a thread of its own and wait for it until it ends, or until the thread has shown no sign of work for a
//!  window, so a job that panics or never ends costs the caller one failed result, not the process.
//! Why:
//!  The Markdown linter used to run as a child process,
//!  and the operating system gave the wrapper a free boundary:
//!  a child that hung could be killed and a child that crashed could not take the wrapper down.
//!  Linked into the wrapper,
//!  the linter shares its address space,
//!  so a boundary is rebuilt here from what the language offers,
//!  and it is a smaller one than a process gave:
//!  a hang and a panic are contained, a stack overflow, an out-of-memory abort or any other abort is not,
//!  and ends the wrapper like any other crash.
//!  The size of a finished result is bounded by the caller after the fact.
//!  The long-term answer is to run the linter in the resident service, outside the client.
//!  A panic is caught on the job's thread and returned as text.
//!  A job that shows no sign of work for the window cannot be stopped from outside,
//!  so the wrapper stops waiting for it and leaves the thread to end with the process,
//!  which is short-lived;
//!  the thread owns everything it uses and shares nothing with the caller,
//!  so abandoning it leaves nothing locked.
//!  The thread has the linter's own stack size,
//!  because the main thread's stack is set by the platform and a stack overflow cannot be caught;
//!  the size makes an overflow as unlikely as in the executable, it does not contain one.
//!  Elapsed time never ends the wait (owner, 2026-11-03: "Stop only on no progress"; pick 9 of
//!  `doc/planning/cli-git-shipped-limits-audit.md`, the coordinator's under the owner's delegation and open to the owner's
//!  veto). The signs of work are those the push reads, taken from the job thread's own files under `/proc/self/task/`
//!  (`thread_quiet.rs`): the bytes it read and wrote, its processor time, its state. The longest quiet gap of every run is kept
//!  in the ledger under `monochromatic-lint` (`quiet_ledger.rs`). Where `/proc` cannot be read the wait has no end but the
//!  job's.
//!
//! In TS you'd write (pseudocode):
//! ```ts
//! // const result = await Promise.race([worker.run(job), rejectWhenQuiet(worker, windowMs)]); // the worker is left running when it is quiet
//! ```

/// When a long wait prints its notices.
use crate::wait_progress::{Cadence, WaitReporter};
/// The longest quiet gap of each tool, kept for the timing records.
use crate::quiet_ledger;
/// The thread's own number, and the clock that says when a thread has been quiet for a window.
use crate::thread_quiet::{ThreadClock, current_thread_id};
/// The thread builder the linter's own executable uses for everything that lints.
use monochromatic_lint::run_workers::lint_thread;
/// Panic containment and the payload type of a caught panic.
use std::panic::{AssertUnwindSafe, catch_unwind};
/// The sink of notices.
use std::io::Write;
/// A one-shot channel from the job's thread to the waiting caller.
use std::sync::mpsc::{Receiver, RecvTimeoutError, Sender, channel};
/// Spans of time and the clock.
use std::time::Duration;

/// What:
///  Why a guarded job produced no result.
/// Why:
///  Each cause names a different way forward for the person.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// type GuardFailure = { kind: 'start'; reason } | { kind: 'panicked'; reason } | { kind: 'quiet'; windowMs } | { kind: 'vanished' };
/// ```
#[derive(Clone, Debug, Eq, PartialEq)]
pub enum GuardFailure {
    /// The operating system would not start the thread.
    Start(String),
    /// The job panicked; the text is the panic's message.
    Panicked(String),
    /// The job showed no sign of work for the window and was abandoned.
    Quiet(Duration),
    /// The thread ended without sending a result or a panic,
    ///  which no code path here does.
    Vanished,
}

/// What:
///  The text of a caught panic's payload.
/// Why:
///  A panic carries a `&str` or a `String`; anything else has no message to show.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// const panicText = (payload: unknown) => typeof payload === 'string' ? payload : 'a panic payload that is not text';
/// ```
fn panic_text(payload: &(dyn std::any::Any + Send)) -> String {
    if let Some(message) = payload.downcast_ref::<String>() {
        return message.clone();
    }
    if let Some(message) = payload.downcast_ref::<&str>() {
        return String::from(*message);
    }
    return String::from("a panic payload that is not text");
}

/// What:
///  The notice a long wait prints.
/// Why:
///  The linter hashes every image a file links, which takes seconds for a large one,
///  and a silent wait looks like a hang;
///  the notice names the file and where the wait stops.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// const noticeLine = (waitingFor, waitedMs, windowMs) => `cli-git: still waiting for ${waitingFor} after ...`;
/// ```
fn notice_line(waiting_for: &str, waited: Duration, window: Duration, repeat: bool) -> String {
    let hint: &str = if repeat {
        ""
    } else {
        " Hint: it is still working; if it looks stuck, wait for the window or set the policy markdown/autofix to off for this command."
    };
    return format!(
        "cli-git: still waiting for {waiting_for} after {} seconds; giving up on it only after {} seconds with no sign of work.{hint}",
        waited.as_secs(),
        window.as_secs()
    );
}

/// What:
///  Wait for the job's thread to report,
///  printing notices at the cadence,
///  until the thread has shown no sign of work for the window.
/// Why:
///  The wait sleeps on the channel,
///  so a job that finishes wakes it at once,
///  and wakes by itself only when a notice or a look at the thread is due.
///  The thread's number arrives first on its own channel; until it does the thread is starting, which is not quiet.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// async function waitFor<T>(receiver, ids, windowMs, reporter): Promise<T>; // throws GuardFailure
/// ```
fn wait_for<T>(
    receiver: &Receiver<Result<T, String>>,
    thread_numbers: &Receiver<i32>,
    window: Duration,
    waiting_for: &str,
    reporter: &mut WaitReporter<'_>,
) -> (Result<T, GuardFailure>, Duration) {
    let mut clock: Option<ThreadClock> = None;
    let mut longest: Duration = Duration::ZERO;
    loop {
        // A notice that is due goes out before the thread is judged, so an abandoned job was announced.
        reporter.poll(|waited, repeat| {
            return notice_line(waiting_for, waited, window, repeat);
        });
        if clock.is_none()
            && let Ok(number) = thread_numbers.try_recv()
        {
            clock = Some(ThreadClock::start(number, window, false));
        }
        if let Some(watching) = clock.as_mut() {
            let quiet: bool = watching.quiet();
            longest = watching.longest_quiet();
            if quiet {
                return (Err(GuardFailure::Quiet(window)), longest);
            }
        }
        match receiver.recv_timeout(NOTICE_POLL) {
            Ok(Ok(value)) => return (Ok(value), longest),
            Ok(Err(reason)) => return (Err(GuardFailure::Panicked(reason)), longest),
            Err(RecvTimeoutError::Timeout) => {}
            Err(RecvTimeoutError::Disconnected) => return (Err(GuardFailure::Vanished), longest),
        }
    }
}

/// The longest the wait sleeps before it looks at the notice schedule again.
///  Exempt from a citation under `LQM` (class: poll): it sets how often a condition is looked at again, so it adds at
///  most one poll interval to a wait that already exists. Classed once in `doc/planning/cli-git-shipped-limits-audit.md`,
///  "Exempt limits"; the coordinator's pick under the owner's delegation, open to the owner's veto.
const NOTICE_POLL: Duration = Duration::from_millis(50);

/// What:
///  Run `job` on a new thread and return its result,
///  or why there is none.
///  `T` is whatever the job returns;
///  both the job and its result move between threads.
/// Why:
///  See the module comment.
///  The caller names what it waits for ("monochromatic-lint to check docs/a.md"),
///  which the notices print.
///
/// In TS you'd write (pseudocode):
/// ```ts
/// async function runGuarded<T>(job: () => T, windowMs, cadence, waitingFor, notices): Promise<T>; // throws GuardFailure
/// ```
pub fn run_guarded<T: Send + 'static>(
    job: impl FnOnce() -> T + Send + 'static,
    window: Duration,
    cadence: Cadence,
    waiting_for: &str,
    notices: &mut dyn Write,
) -> Result<T, GuardFailure> {
    let (sender, receiver): (Sender<Result<T, String>>, Receiver<Result<T, String>>) = channel();
    let (numbers, thread_numbers): (Sender<i32>, Receiver<i32>) = channel();
    let spawned = lint_thread().name(String::from("cli-git-markdown-lint")).spawn(move || {
        // The waiting side judges this thread by its number; a closed channel is a waiter that gave up.
        if let Some(number) = current_thread_id() {
            let _told = numbers.send(number);
        }
        let outcome: Result<T, String> = match catch_unwind(AssertUnwindSafe(job)) {
            Ok(value) => Ok(value),
            Err(payload) => Err(panic_text(payload.as_ref())),
        };
        // The caller may have given up on this job; a closed channel is that case, and there is nobody to tell.
        let _abandoned = sender.send(outcome);
    });
    if let Err(error) = spawned {
        return Err(GuardFailure::Start(error.to_string()));
    }
    let mut reporter: WaitReporter<'_> = WaitReporter::start(cadence, notices);
    let (result, longest): (Result<T, GuardFailure>, Duration) = wait_for(&receiver, &thread_numbers, window, waiting_for, &mut reporter);
    quiet_ledger::note("monochromatic-lint", longest, matches!(result, Err(GuardFailure::Quiet(_))));
    return result;
}

/// Panic, quiet-window and notice controls stay out of the release executable.
#[cfg(test)]
#[path = "markdown_linter_guard_tests.rs"]
mod tests;