ic_testkit/pic/

startup.rs

use std::{
    fmt::Write as _,
    fs::{self, File, OpenOptions},
    future::Future as _,
    io::{self, Read as _},
    path::{Path, PathBuf},
    process::{Command, ExitStatus, Stdio},
    sync::mpsc::{self, RecvTimeoutError},
    thread,
    time::{Duration, Instant},
};

#[cfg(unix)]
use std::os::unix::fs::OpenOptionsExt as _;

use ic_host_fs::{directory::create_unique_directory, owned_directory::OwnedDirectory};
use ic_host_process::child::{CleanupError, CleanupPolicy, OwnedChild};
use ic_platform::pocket_ic::{PocketIc, PocketIcBuilder};

use super::transport;

const STARTUP_POLL_INTERVAL: Duration = Duration::from_millis(20);
const SERVER_OUTPUT_LIMIT: usize = 16 * 1024;
// PocketIC publishes a decimal u16 and a newline. Leave whitespace room
// without allowing readiness polling to read an arbitrary-size file.
const SERVER_PORT_FILE_LIMIT: usize = 64;

/// Explicit bounded source and policy for one `PocketIC` startup.
///
/// Spawn-only settings cannot be attached to a borrowed connection. Callers own
/// binary admission, output retention and namespace custody.
#[derive(Clone, Debug, Eq, PartialEq)]
pub enum PocketIcStartupConfig {
    /// Spawn and own one exact server executable.
    Spawn {
        /// Caller-selected, already prepared server binary.
        server_binary: PathBuf,
        /// Complete startup deadline; must be positive.
        timeout: Duration,
        /// Optional hard lifetime, in whole seconds; `None` disables it.
        server_hard_ttl: Option<Duration>,
        /// Optional operation-idle lifetime; `None` uses `PocketIC`'s default.
        server_idle_ttl: Option<Duration>,
        /// New caller-owned stdout/stderr files, or temporary output when absent.
        /// Existing entries are refused. Created files survive failure and teardown.
        server_output_files: Option<(PathBuf, PathBuf)>,
    },
    /// Borrow an existing server without claiming its process lifetime.
    Connect {
        /// Existing server URL, admitted by the upstream builder.
        server_url: String,
        /// Complete startup deadline; must be positive.
        timeout: Duration,
    },
}

/// Caller-owned `PocketIC` server process with bounded startup and output capture.
///
/// Dropping the handle terminates and waits for the managed child. On Unix,
/// teardown also terminates descendants remaining in its owned process group.
/// Callers may create several instances through [`Self::url`] and
/// [`PocketIcStartupConfig::Connect`] while retaining explicit server ownership.
/// The handle is process-local and does not coordinate ownership across Cargo
/// or test-runner processes; use an externally owned server with bounded
/// connect mode for that topology.
/// The handle owns no binary discovery, download, cache, or compatibility policy.
pub struct PocketIcManagedServer {
    server: ManagedServer,
    url: String,
}

/// An upstream instance together with the server owned by its construction.
///
/// Fields are ordered so a synchronous [`PocketIc`] is deleted before the server
/// is terminated. Connect mode borrows the caller's server and returns `None`.
/// Native `PocketIC` operations remain available through [`Self::instance`].
/// [`Self::into_parts`] explicitly transfers teardown ordering to the caller.
///
/// `PocketIC` 16 synchronous deletion has no request deadline and can block drop.
/// Its nonblocking instance has no automatic deletion: call
/// [`Self::into_parts`], then `instance.drop().await` while retaining `server`,
/// when deletion matters, especially with a borrowed server. Dropping an owned
/// server stops all its instances.
///
/// Field projection cannot silently detach the client from its server:
/// ```compile_fail
/// use ic_testkit::pic::PocketIcInstance;
/// fn detach<T>(owned: PocketIcInstance<T>) -> T {
///     owned.instance
/// }
/// ```
/// Nor can a client borrow outlive a temporary owner:
/// ```compile_fail
/// use ic_testkit::pic::PocketIcInstance;
/// let client = PocketIcInstance::from_borrowed_server(String::new()).instance();
/// assert!(client.is_empty());
/// ```
#[must_use]
pub struct PocketIcInstance<T> {
    instance: T,
    server: Option<PocketIcManagedServer>,
}

impl<T> PocketIcInstance<T> {
    /// Retain an existing client whose server lifetime is owned elsewhere.
    ///
    /// This creates no server or instance and claims no server process authority.
    /// The client retains its upstream deletion behavior; its external server
    /// must outlive it. Use a startup configuration for a managed server owner.
    pub const fn from_borrowed_server(instance: T) -> Self {
        Self {
            instance,
            server: None,
        }
    }

    /// Borrow the client while retaining this record's managed-server ownership.
    #[must_use]
    pub const fn instance(&self) -> &T {
        &self.instance
    }

    /// Mutably borrow the client while retaining managed-server ownership.
    /// Supports client setup such as `PocketIc::make_live` without splitting the
    /// owner. Replacing the client transfers responsibility for the old client's
    /// server lifetime to the caller, just as with [`Self::into_parts`].
    ///
    /// The borrow cannot outlive the owner:
    /// ```compile_fail
    /// use ic_testkit::pic::PocketIcInstance;
    /// let client = PocketIcInstance::from_borrowed_server(String::new()).instance_mut();
    /// client.push_str("gateway");
    /// ```
    pub const fn instance_mut(&mut self) -> &mut T {
        &mut self.instance
    }

    /// Borrow the exact managed server, or `None` for an externally owned server.
    #[must_use]
    pub const fn server(&self) -> Option<&PocketIcManagedServer> {
        self.server.as_ref()
    }

    /// Transfer the client and optional server, without deleting or stopping either.
    ///
    /// The caller must retain the server until client deletion is complete. For
    /// a nonblocking client, explicitly await `instance.drop()` before dropping
    /// the returned server. No implicit asynchronous or blocking cleanup is added.
    #[must_use]
    pub fn into_parts(self) -> (T, Option<PocketIcManagedServer>) {
        (self.instance, self.server)
    }
}

/// Bounded lossy UTF-8 output captured from a managed `PocketIC` server.
///
/// Each stream retains at most the first 16 KiB. A textual suffix reports the
/// number of omitted bytes when truncation occurred. Unreadable streams and
/// paths replaced with non-regular files are omitted.
#[derive(Clone, Debug, Default, Eq, PartialEq)]
pub struct PocketIcManagedServerOutput {
    stdout: String,
    stderr: String,
}

/// Original startup failure, bounded server output and typed cleanup failures.
///
/// Match [`Self::failure`] to classify the original cause. Cleanup diagnostics
/// never replace it, and command/server failures retain separate ownership.
/// Output is an inert bounded snapshot; it grants no process authority.
#[derive(Debug)]
pub struct PocketIcStartupError {
    failure: Box<PocketIcStartupFailure>,
    output: PocketIcManagedServerOutput,
    command_cleanup: Option<Box<CleanupError>>,
    server_cleanup: Option<Box<CleanupError>>,
}

impl PocketIcStartupError {
    fn new(failure: PocketIcStartupFailure) -> Self {
        Self {
            failure: Box::new(failure),
            output: PocketIcManagedServerOutput::default(),
            command_cleanup: None,
            server_cleanup: None,
        }
    }

    /// Original cause, unaffected by subsequent cleanup failure.
    #[must_use]
    pub fn failure(&self) -> &PocketIcStartupFailure {
        &self.failure
    }

    /// Bounded output from the server owned by this operation, if any.
    #[must_use]
    pub const fn output(&self) -> &PocketIcManagedServerOutput {
        &self.output
    }

    /// Failure cleaning the command's owned process group/direct child.
    #[must_use]
    pub fn command_cleanup(&self) -> Option<&CleanupError> {
        self.command_cleanup.as_deref()
    }

    /// Failure cleaning the managed server's owned process group/direct child.
    #[must_use]
    pub fn server_cleanup(&self) -> Option<&CleanupError> {
        self.server_cleanup.as_deref()
    }

    fn with_cleanup(
        mut self,
        command: Option<CleanupError>,
        server: Option<(PocketIcManagedServerOutput, Option<CleanupError>)>,
    ) -> Self {
        self.command_cleanup = command.map(Box::new);
        if let Some((output, cleanup)) = server {
            self.output = output;
            self.server_cleanup = cleanup.map(Box::new);
        }
        self
    }
}

/// Original failure, independent of output capture and cleanup outcomes.
#[non_exhaustive]
#[derive(Debug, thiserror::Error)]
pub enum PocketIcStartupFailure {
    /// Neither the shared server URL nor an explicit executable was configured.
    #[error(
        "configure IC_TESTKIT_POCKET_IC_URL or POCKET_IC_BIN; prepare a verified binary with make install-server"
    )]
    NotConfigured,
    /// A selected environment value was empty or was not valid Unicode.
    #[error("invalid selected environment value: {variable}")]
    InvalidEnvironment {
        /// Selected environment variable name.
        variable: &'static str,
    },
    /// The bounded version probe failed, including nonzero exit or timeout.
    #[error("PocketIC version probe failed: {source}")]
    ServerVersionProbe {
        /// Underlying operation failure.
        source: ic_host_process::tool::ToolError,
    },
    /// The selected executable does not report the qualified server identity.
    #[error("PocketIC version mismatch: expected {expected:?}, observed {:?}", String::from_utf8_lossy(.observed))]
    ServerVersionMismatch {
        /// Required server identity.
        expected: String,
        /// Captured version-probe output.
        observed: Vec<u8>,
    },
    /// Command execution failed.
    #[error("command {} failed: {source}", .program.display())]
    CommandRun {
        /// Selected command executable.
        program: PathBuf,
        /// Underlying operation failure.
        source: io::Error,
    },
    /// The caller supplied a zero timeout or unusable hard TTL.
    #[error("{message}")]
    InvalidConfiguration {
        /// Configuration diagnostic.
        message: String,
    },
    /// A caller-provided existing server URL could not be parsed.
    #[error("invalid PocketIC server URL {server_url:?}: {message}")]
    InvalidServerUrl {
        /// Caller-provided server URL.
        server_url: String,
        /// Configuration diagnostic.
        message: String,
    },
    /// Preparing or inspecting bounded startup files failed.
    #[error("failed to {operation} at {}: {source}", .path.display())]
    Io {
        /// Failed filesystem operation.
        operation: &'static str,
        /// Affected filesystem path.
        path: PathBuf,
        /// Underlying operation failure.
        source: io::Error,
    },
    /// The configured `PocketIC` server process could not be spawned.
    #[error("failed to spawn PocketIC server {}: {source}", .server_binary.display())]
    ServerSpawn {
        /// Selected server executable.
        server_binary: PathBuf,
        /// Underlying operation failure.
        source: io::Error,
    },
    /// The managed `PocketIC` server exited during construction or command execution.
    #[error("PocketIC server {} exited with {status} after {elapsed:?}", .server_binary.display())]
    ServerExited {
        /// Selected server executable.
        server_binary: PathBuf,
        /// Observed server exit status.
        status: ExitStatus,
        /// Time spent before the server exited.
        elapsed: Duration,
    },
    /// The managed server did not publish a usable port before the deadline.
    #[error("PocketIC server {} was not ready within {timeout:?}", .server_binary.display())]
    ReadinessTimeout {
        /// Selected server executable.
        server_binary: PathBuf,
        /// Caller-selected deadline budget.
        timeout: Duration,
    },
    /// The managed server published an invalid or oversized port-file value.
    #[error("PocketIC server {} published invalid port {value:?}", .server_binary.display())]
    InvalidServerPort {
        /// Selected server executable.
        server_binary: PathBuf,
        /// Rejected port-file contents.
        value: String,
    },
    /// `PocketIC` instance creation did not finish before the startup deadline.
    #[error("PocketIC instance creation exceeded {timeout:?}")]
    InstanceCreationTimeout {
        /// Caller-selected deadline budget.
        timeout: Duration,
    },
    /// Spawning the bounded builder worker failed.
    #[error("failed to spawn PocketIC builder worker: {source}")]
    BuilderThreadSpawn {
        /// Underlying operation failure.
        source: io::Error,
    },
    /// Upstream `PocketIC` construction panicked before returning an instance.
    #[error("PocketIC startup panicked: {source}")]
    BuilderPanicked {
        /// Underlying operation failure.
        source: super::PocketIcOperationError,
    },
    /// The bounded builder worker ended without returning a result.
    #[error("PocketIC builder worker disconnected without a result")]
    BuilderDisconnected,
    /// The command completed, but the owned server could not be cleaned up.
    #[error("PocketIC server cleanup failed after command exited with {command_status}")]
    ServerCleanup {
        /// Exit status of the completed command.
        command_status: ExitStatus,
    },
}

impl PocketIcStartupConfig {
    /// Select the shared environment contract without discovery or downloads.
    ///
    /// `IC_TESTKIT_POCKET_IC_URL` takes precedence over `POCKET_IC_BIN`. An
    /// explicitly empty or invalid selected value fails rather than falling
    /// back. URL mode never launches a version probe or claims server ownership.
    /// Binary mode resolves the explicit path and checks `--version` against
    /// Testkit's supported protocol range using the shared bounded capture
    /// engine. This is version qualification, not executable-byte admission;
    /// prepare and verify the binary with `ic-testkit-server setup` / `check`.
    /// The probe has its own `timeout`; subsequent startup has the same budget.
    ///
    /// # Errors
    /// Rejects missing or invalid configuration and failed or incompatible version probes.
    pub fn from_env(timeout: Duration) -> Result<Self, PocketIcStartupError> {
        Self::from_environment(timeout, |name| std::env::var_os(name))
    }

    fn from_environment(
        timeout: Duration,
        mut variable: impl FnMut(&str) -> Option<std::ffi::OsString>,
    ) -> Result<Self, PocketIcStartupError> {
        if let Some(value) = variable("IC_TESTKIT_POCKET_IC_URL") {
            let server_url = value
                .into_string()
                .ok()
                .filter(|value| !value.is_empty())
                .ok_or_else(|| {
                    PocketIcStartupError::new(PocketIcStartupFailure::InvalidEnvironment {
                        variable: "IC_TESTKIT_POCKET_IC_URL",
                    })
                })?;
            let config = Self::Connect {
                server_url: (&server_url).into(),
                timeout,
            };
            config.validate()?;
            let _ = configured_builder(PocketIcBuilder::new(), &server_url)?;
            return Ok(config);
        }
        let value = variable("POCKET_IC_BIN")
            .ok_or_else(|| PocketIcStartupError::new(PocketIcStartupFailure::NotConfigured))?;
        if value.is_empty() {
            return Err(PocketIcStartupError::new(
                PocketIcStartupFailure::InvalidEnvironment {
                    variable: "POCKET_IC_BIN",
                },
            ));
        }
        let path = PathBuf::from(value);
        let binary = fs::canonicalize(&path).map_err(|source| {
            PocketIcStartupError::new(PocketIcStartupFailure::Io {
                operation: "resolve configured PocketIC executable",
                path,
                source,
            })
        })?;
        let config = Self::Spawn {
            server_binary: (&binary).into(),
            timeout,
            server_hard_ttl: None,
            server_idle_ttl: None,
            server_output_files: None,
        };
        config.validate()?;
        let evidence = ic_host_process::tool::capture_group_command(
            Command::new(&binary).arg("--version"),
            ic_host_process::tool::OutputLimits {
                stdout: ic_host_process::tool::OutputLimit::Terminate(SERVER_OUTPUT_LIMIT),
                stderr: ic_host_process::tool::OutputLimit::Terminate(SERVER_OUTPUT_LIMIT),
                timeout: Some(timeout),
            },
        )
        .and_then(ic_host_process::tool::CompletedExecution::require_success)
        .map_err(|source| {
            PocketIcStartupError::new(PocketIcStartupFailure::ServerVersionProbe { source })
        })?;
        let expected = super::server::expected_server_version();
        if !super::supports_pocket_ic_server(&evidence.stdout) {
            return Err(PocketIcStartupError::new(
                PocketIcStartupFailure::ServerVersionMismatch {
                    expected,
                    observed: evidence.stdout,
                },
            ));
        }
        Ok(config)
    }

    /// Complete startup deadline for either source.
    #[must_use]
    pub const fn timeout(&self) -> Duration {
        match self {
            Self::Spawn { timeout, .. } | Self::Connect { timeout, .. } => *timeout,
        }
    }

    /// Construct a synchronous instance and retain its exact managed server.
    ///
    /// One deadline covers server readiness and instance creation. Topology,
    /// state paths and request policy remain on the caller's builder. Spawn mode
    /// terminates the server on failure and retains it on success; connect mode
    /// never claims the external server. See [`PocketIcInstance`] for drop limits.
    ///
    /// ```no_run
    /// use ic_platform::pocket_ic::PocketIcBuilder;
    /// use ic_testkit::pic::PocketIcStartupConfig;
    /// use std::{path::Path, time::Duration};
    /// # fn run(binary: &Path) -> Result<(), ic_testkit::pic::PocketIcStartupError> {
    /// let owned = PocketIcStartupConfig::Spawn { server_binary: binary.into(), timeout: Duration::from_secs(60), server_hard_ttl: None, server_idle_ttl: None, server_output_files: None }
    ///     .build_instance(PocketIcBuilder::new().with_application_subnet())?;
    /// let canister = owned.instance().create_canister();
    /// assert!(owned.instance().canister_exists(canister));
    /// // Drop deletes the instance, then terminates and reaps the owned server.
    /// # Ok(())
    /// # }
    /// ```
    ///
    /// # Errors
    /// Returns configuration, startup, deadline or instance-creation failures with cleanup evidence.
    pub fn build_instance(
        self,
        builder: PocketIcBuilder,
    ) -> Result<PocketIcInstance<PocketIc>, PocketIcStartupError> {
        let startup = StartupDeadline::new(&self)?;
        let (server, url) = match self {
            Self::Connect { server_url, .. } => (None, server_url),
            spawn @ Self::Spawn { .. } => {
                let (server, url) = spawn.spawn_server(startup.started)?.wait_ready(&startup)?;
                (Some(server), url)
            }
        };
        build_bounded(builder, &url, &startup, server)
    }

    /// Construct an upstream nonblocking instance on the caller's Tokio runtime.
    ///
    /// Uses the same deadline, readiness checks, typed panic/exit failures and
    /// bounded output as [`Self::build_instance`], without running the synchronous
    /// instance constructor. Requires a Tokio runtime with its time driver enabled.
    /// Readiness and instance waits yield; process spawning and small file reads
    /// remain synchronous, as does exact owned-child cleanup. The deadline cannot
    /// preempt synchronous work inside an upstream future poll (such as copying
    /// a read-only state directory); it is enforced when that poll returns.
    ///
    /// Dropping the future cancels local construction and terminates/reaps its
    /// owned server, including during readiness. Cancellation has no returned error;
    /// select output files to retain diagnostics. A borrowed server is untouched.
    /// If a borrowed server created an instance before timeout or a lost reply, local
    /// cancellation cannot identify or delete that instance; the caller owns this
    /// reconciliation. The deadline covers construction, not subsequent deletion.
    ///
    /// ```no_run
    /// use ic_platform::pocket_ic::PocketIcBuilder;
    /// use ic_testkit::pic::PocketIcStartupConfig;
    /// use std::{path::Path, time::Duration};
    /// # async fn run(binary: &Path) -> Result<(), ic_testkit::pic::PocketIcStartupError> {
    /// let owned = PocketIcStartupConfig::Spawn { server_binary: binary.into(), timeout: Duration::from_secs(60), server_hard_ttl: None, server_idle_ttl: None, server_output_files: None }
    ///     .build_instance_async(PocketIcBuilder::new().with_application_subnet()).await?;
    /// let canister = owned.instance().create_canister().await;
    /// assert!(owned.instance().canister_exists(canister).await);
    /// let (instance, server) = owned.into_parts();
    /// instance.drop().await;
    /// drop(server);
    /// # Ok(())
    /// # }
    /// ```
    ///
    /// # Errors
    /// Returns configuration, startup, deadline or instance-creation failures with cleanup evidence.
    pub async fn build_instance_async(
        self,
        builder: PocketIcBuilder,
    ) -> Result<PocketIcInstance<ic_platform::pocket_ic::nonblocking::PocketIc>, PocketIcStartupError>
    {
        let startup = StartupDeadline::new(&self)?;
        let (server, server_url) = match self {
            Self::Connect { server_url, .. } => (None, server_url),
            spawn @ Self::Spawn { .. } => {
                let mut server = spawn.spawn_server(startup.started)?;
                let server_url = loop {
                    match server.poll_ready(&startup) {
                        Ok(Some(url)) => break url,
                        Ok(None) => {
                            tokio::time::sleep(
                                startup
                                    .deadline
                                    .saturating_duration_since(Instant::now())
                                    .min(STARTUP_POLL_INTERVAL),
                            )
                            .await;
                        }
                        Err(error) => return Err(finalize_failure(error, None, Some(server))),
                    }
                };
                (Some(server), server_url)
            }
        };
        build_async_bounded(builder, &server_url, &startup, server).await
    }

    /// Start a caller-owned managed server without constructing an instance.
    ///
    /// This requires a configuration created by [`Self::Spawn`]. Readiness is
    /// bounded by [`Self::timeout`]. No hard TTL is passed by default; an
    /// explicit `server_hard_ttl` value is passed to the child.
    /// `server_idle_ttl` independently selects the operation-idle
    /// lifetime; otherwise `PocketIC` retains its own idle default.
    /// Readiness requires a nonzero decimal port followed by a newline in a
    /// regular UTF-8 file of at most 64 bytes; oversized files fail with bounded
    /// diagnostics. Non-regular port files fail with [`PocketIcStartupFailure::Io`]
    /// and [`io::ErrorKind::InvalidData`] without waiting for a FIFO writer.
    /// The returned handle terminates the child on drop; use its URL with
    /// [`Self::Connect`] to construct bounded instances.
    ///
    /// # Errors
    /// Rejects invalid configuration and server startup or readiness failures.
    pub fn start_managed_server(self) -> Result<PocketIcManagedServer, PocketIcStartupError> {
        if matches!(self, Self::Connect { .. }) {
            self.validate()?;
            return Err(PocketIcStartupError::new(
                PocketIcStartupFailure::InvalidConfiguration {
                    message: "starting a managed PocketIC server requires a spawn configuration"
                        .to_owned(),
                },
            ));
        }
        let startup = StartupDeadline::new(&self)?;
        let (server, url) = self.spawn_server(startup.started)?.wait_ready(&startup)?;
        Ok(PocketIcManagedServer { server, url })
    }

    // Only validated spawn configurations reach this projection. Keeping it here
    // gives sync, async and server-only construction the same spawn settings.
    fn spawn_server(self, started: Instant) -> Result<ManagedServer, PocketIcStartupError> {
        let Self::Spawn {
            server_binary,
            server_hard_ttl,
            server_idle_ttl,
            server_output_files,
            ..
        } = self
        else {
            unreachable!("connect configurations do not spawn a server")
        };
        ManagedServer::spawn(
            server_binary,
            server_hard_ttl,
            server_idle_ttl,
            server_output_files,
            started,
        )
    }

    /// Run a command with `IC_TESTKIT_POCKET_IC_URL` set to this server.
    ///
    /// Spawn mode retains the managed server until command completion; connect
    /// mode borrows the external server and never terminates it. The command's
    /// IO and other environment selections remain caller-owned. Cancellation
    /// is polled after bounded startup and every 20 ms while the command runs.
    /// It returns an [`io::ErrorKind::Interrupted`] error after cleanup.
    /// If the owned server exits while the command is pending, the command is
    /// terminated and the server's status and bounded diagnostics are returned
    /// as [`PocketIcStartupFailure::ServerExited`]. External servers are not monitored.
    ///
    /// On Unix the command starts in a new owned process group. Completion,
    /// cancellation and observation failures terminate remaining group members
    /// before reaping the leader, using the same lifecycle engine as managed
    /// servers. This does not impose a command deadline or sandbox descendants
    /// that deliberately leave the owned group. Other hosts own the direct child.
    ///
    /// # Errors
    /// Returns startup, command execution, cancellation or server cleanup failures.
    pub fn run_command(
        self,
        command: &mut Command,
        mut cancelled: impl FnMut() -> bool,
    ) -> Result<ExitStatus, PocketIcStartupError> {
        self.validate()?;
        let program = PathBuf::from(command.get_program());
        let command_error = |source| {
            PocketIcStartupError::new(PocketIcStartupFailure::CommandRun {
                program: program.clone(),
                source,
            })
        };
        if cancelled() {
            return Err(command_error(io::Error::from(io::ErrorKind::Interrupted)));
        }
        let (server, url) = match self {
            Self::Connect { server_url, .. } => (None, server_url),
            spawn @ Self::Spawn { .. } => {
                let server = spawn.start_managed_server()?;
                let url = server.url().to_owned();
                (Some(server.server), url)
            }
        };
        let mut server = server;
        if cancelled() {
            return Err(finalize_failure(
                command_error(io::Error::from(io::ErrorKind::Interrupted)),
                None,
                server,
            ));
        }
        command.env("IC_TESTKIT_POCKET_IC_URL", url);
        let mut owned_child = match OwnedChild::spawn(command, CleanupPolicy::KillAndWait) {
            Ok(child) => child,
            Err(source) => {
                return Err(finalize_failure(
                    PocketIcStartupError::new(PocketIcStartupFailure::Io {
                        operation: "spawn command with PocketIC server",
                        path: PathBuf::from(command.get_program()),
                        source,
                    }),
                    None,
                    server,
                ));
            }
        };
        let result = loop {
            if cancelled() {
                break Err(command_error(io::Error::from(io::ErrorKind::Interrupted)));
            }
            match owned_child.try_wait() {
                Ok(Some(status)) => break Ok(status),
                Ok(None) => {
                    if let Some(managed) = server.as_mut() {
                        match managed.try_wait() {
                            Ok(Some(status)) => break Err(managed.exit_failure(status)),
                            Ok(None) => {}
                            Err(error) => break Err(error),
                        }
                    }
                    thread::sleep(STARTUP_POLL_INTERVAL);
                }
                Err(source) => break Err(command_error(source)),
            }
        };
        match result {
            Err(error) => Err(finalize_failure(error, Some(&mut owned_child), server)),
            Ok(status) => {
                if let Some(server) = server {
                    let (output, cleanup) = server.terminate_and_capture();
                    if let Some(cleanup) = cleanup {
                        return Err(PocketIcStartupError::new(
                            PocketIcStartupFailure::ServerCleanup {
                                command_status: status,
                            },
                        )
                        .with_cleanup(None, Some((output, Some(cleanup)))));
                    }
                }
                Ok(status)
            }
        }
    }

    fn validate(&self) -> Result<(), PocketIcStartupError> {
        if self.timeout().is_zero() {
            return Err(PocketIcStartupError::new(
                PocketIcStartupFailure::InvalidConfiguration {
                    message: "PocketIC startup timeout must be greater than zero".to_owned(),
                },
            ));
        }
        if let Self::Spawn {
            server_hard_ttl,
            server_idle_ttl,
            ..
        } = self
        {
            for (ttl, message) in [
                (
                    server_hard_ttl,
                    "PocketIC server hard TTL must be at least one second",
                ),
                (
                    server_idle_ttl,
                    "PocketIC server idle TTL must be at least one second",
                ),
            ] {
                if ttl.is_some_and(|ttl| ttl.as_secs() == 0) {
                    return Err(PocketIcStartupError::new(
                        PocketIcStartupFailure::InvalidConfiguration {
                            message: message.to_owned(),
                        },
                    ));
                }
            }
        }
        Ok(())
    }
}

impl PocketIcManagedServer {
    /// OS process ID of the owned server child, for caller-managed monitoring.
    ///
    /// This identifies the server process, not its descendants, and does not
    /// establish that it is still running. The OS may reuse the ID after the
    /// child exits and is reaped. Dropping this handle terminates and waits for
    /// the child; retaining the ID does not retain server ownership.
    #[must_use]
    pub fn process_id(&self) -> u32 {
        self.server.child.id()
    }

    /// Loopback URL published by the managed server.
    #[must_use]
    pub fn url(&self) -> &str {
        &self.url
    }

    /// Current bounded stdout and stderr captured from the managed server.
    ///
    /// This reads at most the first 16 KiB from each retained output file,
    /// renders it as lossy UTF-8, and adds an omitted-byte suffix when truncated.
    /// The diagnostic read is bounded; files can grow while the server runs.
    #[must_use]
    pub fn output(&self) -> PocketIcManagedServerOutput {
        self.server.files.capture()
    }
}

impl PocketIcManagedServerOutput {
    /// Bounded lossy UTF-8 standard output.
    #[must_use]
    pub fn stdout(&self) -> &str {
        &self.stdout
    }

    /// Bounded lossy UTF-8 standard error.
    #[must_use]
    pub fn stderr(&self) -> &str {
        &self.stderr
    }
}

struct StartupDeadline {
    started: Instant,
    timeout: Duration,
    deadline: Instant,
}

impl StartupDeadline {
    fn new(config: &PocketIcStartupConfig) -> Result<Self, PocketIcStartupError> {
        config.validate()?;
        let started = Instant::now();
        let timeout = config.timeout();
        Ok(Self {
            started,
            timeout,
            deadline: startup_deadline(started, timeout)?,
        })
    }

    fn check_instance_deadline(
        &self,
        server: &mut Option<ManagedServer>,
    ) -> Result<Instant, PocketIcStartupError> {
        let now = Instant::now();
        if now >= self.deadline {
            return Err(finalize_failure(
                PocketIcStartupError::new(PocketIcStartupFailure::InstanceCreationTimeout {
                    timeout: self.timeout,
                }),
                None,
                server.take(),
            ));
        }
        Ok(now)
    }

    fn admit_instance<T>(
        &self,
        instance: T,
        server_url: &str,
        server: &mut Option<ManagedServer>,
    ) -> Result<PocketIcInstance<T>, PocketIcStartupError> {
        self.check_instance_deadline(server)?;
        check_server(server)?;
        Ok(PocketIcInstance {
            instance,
            server: server.take().map(|server| PocketIcManagedServer {
                server,
                url: server_url.to_owned(),
            }),
        })
    }
}

fn startup_deadline(started: Instant, timeout: Duration) -> Result<Instant, PocketIcStartupError> {
    started.checked_add(timeout).ok_or_else(|| {
        PocketIcStartupError::new(PocketIcStartupFailure::InvalidConfiguration {
            message: "PocketIC startup timeout exceeds the platform clock range".to_owned(),
        })
    })
}

fn build_bounded(
    builder: PocketIcBuilder,
    server_url: &str,
    startup: &StartupDeadline,
    mut server: Option<ManagedServer>,
) -> Result<PocketIcInstance<PocketIc>, PocketIcStartupError> {
    let builder = configured_builder(builder, server_url)
        .map_err(|error| finalize_failure(error, None, server.take()))?;
    let (sender, receiver) = mpsc::sync_channel(1);
    if let Err(source) = thread::Builder::new()
        .name("ic-testkit-pocket-ic-startup".to_owned())
        .spawn(move || {
            let result = transport::capture_operation(|| builder.build());
            let _ = sender.send(result);
        })
    {
        return Err(finalize_failure(
            PocketIcStartupError::new(PocketIcStartupFailure::BuilderThreadSpawn { source }),
            None,
            server,
        ));
    }
    await_builder(receiver, server_url, startup, &mut server)
}

#[expect(
    clippy::needless_pass_by_value,
    reason = "consume the receiver so failed startup disconnects late builders"
)]
fn await_builder(
    receiver: mpsc::Receiver<Result<PocketIc, super::PocketIcOperationError>>,
    server_url: &str,
    startup: &StartupDeadline,
    server: &mut Option<ManagedServer>,
) -> Result<PocketIcInstance<PocketIc>, PocketIcStartupError> {
    loop {
        let now = startup.check_instance_deadline(server)?;
        let remaining = startup.deadline.saturating_duration_since(now);
        let wait = if server.is_some() {
            remaining.min(STARTUP_POLL_INTERVAL)
        } else {
            remaining
        };
        match receiver.recv_timeout(wait) {
            Ok(Ok(pocket_ic)) => {
                return startup.admit_instance(pocket_ic, server_url, server);
            }
            Ok(Err(source)) => {
                return Err(finalize_failure(
                    PocketIcStartupError::new(PocketIcStartupFailure::BuilderPanicked { source }),
                    None,
                    server.take(),
                ));
            }
            Err(RecvTimeoutError::Disconnected) => {
                return Err(finalize_failure(
                    PocketIcStartupError::new(PocketIcStartupFailure::BuilderDisconnected),
                    None,
                    server.take(),
                ));
            }
            Err(RecvTimeoutError::Timeout) => {
                check_server(server)?;
            }
        }
    }
}

fn configured_builder(
    builder: PocketIcBuilder,
    server_url: &str,
) -> Result<PocketIcBuilder, PocketIcStartupError> {
    server_url
        .parse()
        .map(|url| builder.with_server_url(url))
        .map_err(|error| {
            PocketIcStartupError::new(PocketIcStartupFailure::InvalidServerUrl {
                server_url: server_url.to_owned(),
                message: error.to_string(),
            })
        })
}

fn check_server(server: &mut Option<ManagedServer>) -> Result<(), PocketIcStartupError> {
    if let Some(managed) = server.as_mut() {
        let error = match managed.try_wait() {
            Ok(Some(status)) => Some(managed.exit_failure(status)),
            Ok(None) => None,
            Err(error) => Some(error),
        };
        if let Some(error) = error {
            return Err(finalize_failure(error, None, server.take()));
        }
    }
    Ok(())
}

async fn build_async_bounded(
    builder: PocketIcBuilder,
    server_url: &str,
    startup: &StartupDeadline,
    mut server: Option<ManagedServer>,
) -> Result<PocketIcInstance<ic_platform::pocket_ic::nonblocking::PocketIc>, PocketIcStartupError> {
    let builder = configured_builder(builder, server_url)
        .map_err(|error| finalize_failure(error, None, server.take()))?;
    let mut construction = std::pin::pin!(builder.build_async());
    let mut construction = std::pin::pin!(std::future::poll_fn(|context| {
        match transport::capture_operation(|| construction.as_mut().poll(context)) {
            Ok(poll) => poll.map(Ok),
            Err(source) => std::task::Poll::Ready(Err(PocketIcStartupError::new(
                PocketIcStartupFailure::BuilderPanicked { source },
            ))),
        }
    }));
    // Drop this local before the pending upstream future on cancellation.
    let mut server = server;
    loop {
        let now = startup.check_instance_deadline(&mut server)?;
        check_server(&mut server)?;
        let next_check = if server.is_some() {
            now + startup
                .deadline
                .saturating_duration_since(now)
                .min(STARTUP_POLL_INTERVAL)
        } else {
            startup.deadline
        };
        match tokio::time::timeout_at(next_check.into(), construction.as_mut()).await {
            Ok(Ok(instance)) => {
                return startup.admit_instance(instance, server_url, &mut server);
            }
            Ok(Err(error)) => return Err(finalize_failure(error, None, server)),
            Err(_) => {}
        }
    }
}

/// Finalize an owned operation once, preserving the primary failure.
fn finalize_failure(
    error: PocketIcStartupError,
    command: Option<&mut OwnedChild>,
    server: Option<ManagedServer>,
) -> PocketIcStartupError {
    error.with_cleanup(
        command.and_then(|child| child.terminate().err()),
        server.map(ManagedServer::terminate_and_capture),
    )
}

struct ManagedServer {
    child: OwnedChild,
    binary: PathBuf,
    files: StartupFiles,
    started: Instant,
}

enum PortFileState {
    Pending,
    Ready(u16),
    Invalid(String),
}

impl ManagedServer {
    fn wait_ready(
        mut self,
        startup: &StartupDeadline,
    ) -> Result<(Self, String), PocketIcStartupError> {
        loop {
            match self.poll_ready(startup) {
                Ok(Some(url)) => return Ok((self, url)),
                Ok(None) => thread::sleep(
                    startup
                        .deadline
                        .saturating_duration_since(Instant::now())
                        .min(STARTUP_POLL_INTERVAL),
                ),
                Err(error) => return Err(finalize_failure(error, None, Some(self))),
            }
        }
    }

    fn spawn(
        binary: PathBuf,
        hard_ttl: Option<Duration>,
        idle_ttl: Option<Duration>,
        output_files: Option<(PathBuf, PathBuf)>,
        started: Instant,
    ) -> Result<Self, PocketIcStartupError> {
        let (files, stdout, stderr) = StartupFiles::create(output_files)?;
        let mut command = Command::new(&binary);
        if let Some(hard_ttl) = hard_ttl {
            command
                .arg("--hard-ttl")
                .arg(hard_ttl.as_secs().to_string());
        }
        if let Some(idle_ttl) = idle_ttl {
            command.arg("--ttl").arg(idle_ttl.as_secs().to_string());
        }
        command
            .arg("--port-file")
            .arg(&files.port)
            .stdout(Stdio::from(stdout))
            .stderr(Stdio::from(stderr));
        let child =
            OwnedChild::spawn(&mut command, CleanupPolicy::KillAndWait).map_err(|source| {
                PocketIcStartupError::new(PocketIcStartupFailure::ServerSpawn {
                    server_binary: binary.clone(),
                    source,
                })
            })?;
        Ok(Self {
            child,
            binary,
            files,
            started,
        })
    }

    fn poll_ready(
        &mut self,
        startup: &StartupDeadline,
    ) -> Result<Option<String>, PocketIcStartupError> {
        if let Some(status) = self.try_wait()? {
            return Err(self.exit_failure(status));
        }
        if Instant::now() >= startup.deadline {
            return Err(PocketIcStartupError::new(
                PocketIcStartupFailure::ReadinessTimeout {
                    server_binary: self.binary.clone(),
                    timeout: startup.timeout,
                },
            ));
        }
        match self.files.read_port()? {
            PortFileState::Pending => Ok(None),
            PortFileState::Ready(port) => Ok(Some(format!("http://127.0.0.1:{port}/"))),
            PortFileState::Invalid(value) => Err(PocketIcStartupError::new(
                PocketIcStartupFailure::InvalidServerPort {
                    server_binary: self.binary.clone(),
                    value,
                },
            )),
        }
    }

    fn try_wait(&mut self) -> Result<Option<ExitStatus>, PocketIcStartupError> {
        self.child.try_wait().map_err(|source| {
            PocketIcStartupError::new(PocketIcStartupFailure::Io {
                operation: "inspect PocketIC server child",
                path: self.binary.clone(),
                source,
            })
        })
    }

    fn exit_failure(&self, status: ExitStatus) -> PocketIcStartupError {
        PocketIcStartupError::new(PocketIcStartupFailure::ServerExited {
            server_binary: self.binary.clone(),
            status,
            elapsed: self.started.elapsed(),
        })
    }

    fn terminate_and_capture(self) -> (PocketIcManagedServerOutput, Option<CleanupError>) {
        let Self {
            mut child, files, ..
        } = self;
        let cleanup = child.terminate().err();
        // Finish the child's Drop cleanup before reading the final diagnostics.
        drop(child);
        (files.capture(), cleanup)
    }
}

struct StartupFiles {
    #[cfg_attr(
        not(test),
        expect(
            dead_code,
            reason = "retains directory ownership until cleanup or drop"
        )
    )]
    directory: OwnedDirectory,
    port: PathBuf,
    stdout: PathBuf,
    stderr: PathBuf,
}

impl StartupFiles {
    fn read_port(&self) -> Result<PortFileState, PocketIcStartupError> {
        let port_path = &self.port;
        let mut contents = String::new();
        match ic_host_fs::read::open_regular_file(port_path)
            .map_err(|error| match error {
                ic_host_artifacts::artifact::ArtifactError::Io(source) => source,
                error => io::Error::new(io::ErrorKind::InvalidData, error),
            })
            .and_then(|file| {
                file.take((SERVER_PORT_FILE_LIMIT + 1) as u64)
                    .read_to_string(&mut contents)
            }) {
            Ok(_) => {}
            Err(error) if error.kind() == io::ErrorKind::NotFound => {
                return Ok(PortFileState::Pending);
            }
            Err(source) => {
                return Err(PocketIcStartupError::new(PocketIcStartupFailure::Io {
                    operation: "read PocketIC server port file",
                    path: port_path.clone(),
                    source,
                }));
            }
        }
        if contents.len() > SERVER_PORT_FILE_LIMIT {
            return Ok(PortFileState::Invalid(format!(
                "{} (port file exceeds {SERVER_PORT_FILE_LIMIT} bytes)",
                contents.trim()
            )));
        }
        if !contents.contains('\n') {
            return Ok(PortFileState::Pending);
        }
        let value = contents.trim().to_owned();
        match value.parse::<u16>() {
            Ok(port) if port != 0 => Ok(PortFileState::Ready(port)),
            _ => Ok(PortFileState::Invalid(value)),
        }
    }

    fn capture(&self) -> PocketIcManagedServerOutput {
        PocketIcManagedServerOutput {
            stdout: read_bounded_lossy(&self.stdout),
            stderr: read_bounded_lossy(&self.stderr),
        }
    }

    fn create(
        output_files: Option<(PathBuf, PathBuf)>,
    ) -> Result<(Self, File, File), PocketIcStartupError> {
        let output_files = output_files
            .map(|(stdout, stderr)| {
                Ok::<_, PocketIcStartupError>((
                    std::path::absolute(&stdout)
                        .map_err(|source| startup_file_error("resolve", &stdout, source))?,
                    std::path::absolute(&stderr)
                        .map_err(|source| startup_file_error("resolve", &stderr, source))?,
                ))
            })
            .transpose()?;
        let directory = OwnedDirectory::new(
            create_unique_directory(
                &std::env::temp_dir(),
                "ic-testkit-pocket-ic-startup",
                0o700,
                None,
            )
            .map_err(|source| startup_file_error("create", &std::env::temp_dir(), source))?,
        );
        let base = directory.path();
        let (stdout_path, stderr_path) =
            output_files.unwrap_or_else(|| (base.join("stdout"), base.join("stderr")));
        let files = Self {
            port: base.join("port"),
            stdout: stdout_path,
            stderr: stderr_path,
            directory,
        };
        let stdout = create_new_file(&files.stdout)
            .map_err(|source| startup_file_error("create", &files.stdout, source))?;
        let stderr = create_new_file(&files.stderr)
            .map_err(|source| startup_file_error("create", &files.stderr, source))?;
        Ok((files, stdout, stderr))
    }
}

fn create_new_file(path: &Path) -> io::Result<File> {
    let mut options = OpenOptions::new();
    options.write(true).create_new(true);
    #[cfg(unix)]
    options.mode(0o600);
    options.open(path)
}

fn startup_file_error(
    operation: &'static str,
    path: &Path,
    source: io::Error,
) -> PocketIcStartupError {
    PocketIcStartupError::new(PocketIcStartupFailure::Io {
        operation,
        path: path.to_owned(),
        source,
    })
}

fn read_bounded_lossy(path: &Path) -> String {
    let Ok((bytes, length)) = ic_host_fs::read::read_file_prefix(path, SERVER_OUTPUT_LIMIT) else {
        return String::new();
    };
    let mut output = String::from_utf8_lossy(&bytes).into_owned();
    let omitted = length.saturating_sub(bytes.len() as u64);
    if omitted > 0 {
        let _ = write!(output, "\n<truncated {omitted} bytes>");
    }
    output
}

impl std::fmt::Display for PocketIcStartupError {
    fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        std::fmt::Display::fmt(self.failure.as_ref(), formatter)?;
        if !self.output.stdout.is_empty() {
            write!(formatter, "; server stdout: {}", self.output.stdout)?;
        }
        if !self.output.stderr.is_empty() {
            write!(formatter, "; server stderr: {}", self.output.stderr)?;
        }
        for (owner, cleanup) in [
            ("command", self.command_cleanup()),
            ("server", self.server_cleanup()),
        ] {
            if let Some(error) = cleanup {
                write!(formatter, "; {owner} cleanup also failed: {error}")?;
            }
        }
        Ok(())
    }
}

impl std::error::Error for PocketIcStartupError {
    fn source(&self) -> Option<&(dyn std::error::Error + 'static)> {
        std::error::Error::source(self.failure.as_ref())
    }
}

#[cfg(test)]
mod tests {
    mod instances;

    use std::{
        fs,
        path::PathBuf,
        time::{Duration, Instant},
    };

    use super::{
        OwnedDirectory, PocketIcStartupConfig, PocketIcStartupError, PocketIcStartupFailure,
        StartupFiles, create_unique_directory,
    };
    use ic_platform::pocket_ic::PocketIcBuilder;

    #[cfg(unix)]
    use ic_host_process::testing::write_executable_script;
    #[cfg(unix)]
    use std::{
        io::Write as _,
        os::unix::fs::{OpenOptionsExt as _, PermissionsExt as _},
        process::Command,
        sync::mpsc,
    };

    #[test]
    fn completed_instances_must_still_meet_the_startup_deadline() {
        let started = Instant::now();
        let startup = super::StartupDeadline {
            started,
            timeout: Duration::from_secs(60),
            deadline: started + Duration::from_secs(60),
        };
        let mut server = None;
        let instance = startup
            .admit_instance(7_u8, "http://localhost/", &mut server)
            .unwrap();
        assert_eq!(*instance.instance(), 7);
        assert!(server.is_none());
        let expired = super::StartupDeadline {
            deadline: started,
            ..startup
        };
        let error = expired
            .admit_instance(7_u8, "http://localhost/", &mut server)
            .err()
            .expect("completed instance after the deadline must refuse admission");
        assert!(
            matches!(error.failure(), PocketIcStartupFailure::InstanceCreationTimeout { timeout } if *timeout == Duration::from_secs(60))
        );
        assert!(server.is_none());
    }

    #[test]
    fn original_failure_output_and_both_typed_cleanup_reports_survive_projection() {
        use std::error::Error as _;
        for failure in [
            PocketIcStartupFailure::ReadinessTimeout {
                server_binary: "server".into(),
                timeout: Duration::from_secs(1),
            },
            PocketIcStartupFailure::InstanceCreationTimeout {
                timeout: Duration::from_secs(1),
            },
            PocketIcStartupFailure::InvalidServerPort {
                server_binary: "server".into(),
                value: "invalid".into(),
            },
            PocketIcStartupFailure::BuilderPanicked {
                source: super::super::PocketIcOperationError::new("original-panic"),
            },
            PocketIcStartupFailure::BuilderDisconnected,
            PocketIcStartupFailure::BuilderThreadSpawn {
                source: std::io::Error::from_raw_os_error(9),
            },
            PocketIcStartupFailure::Io {
                operation: "original-io",
                path: "port".into(),
                source: std::io::Error::from_raw_os_error(9),
            },
        ] {
            let primary_text = failure.to_string();
            let output = super::PocketIcManagedServerOutput {
                stdout: "original-stdout".into(),
                stderr: "original-stderr".into(),
            };
            let command = super::CleanupError {
                status: None,
                group_error: None,
                term_error: Some(std::io::Error::from_raw_os_error(4)),
                kill_error: Some(std::io::Error::from_raw_os_error(5)),
                wait_error: None,
            };
            let server = super::CleanupError {
                status: None,
                group_error: Some(std::io::Error::from_raw_os_error(1)),
                term_error: None,
                kill_error: None,
                wait_error: Some(std::io::Error::from_raw_os_error(10)),
            };
            let error = PocketIcStartupError::new(failure)
                .with_cleanup(Some(command), Some((output, Some(server))));
            assert_eq!(error.failure().to_string(), primary_text);
            assert_eq!(error.output().stdout(), "original-stdout");
            assert_eq!(error.output().stderr(), "original-stderr");
            let command = error.command_cleanup().unwrap();
            let server = error.server_cleanup().unwrap();
            assert_eq!(command.kill_error.as_ref().unwrap().raw_os_error(), Some(5));
            assert_eq!(command.term_error.as_ref().unwrap().raw_os_error(), Some(4));
            assert_eq!(server.group_error.as_ref().unwrap().raw_os_error(), Some(1));
            assert_eq!(server.wait_error.as_ref().unwrap().raw_os_error(), Some(10));
            if matches!(
                error.failure(),
                PocketIcStartupFailure::Io { .. }
                    | PocketIcStartupFailure::BuilderThreadSpawn { .. }
            ) {
                assert_eq!(
                    error
                        .source()
                        .unwrap()
                        .downcast_ref::<std::io::Error>()
                        .unwrap()
                        .raw_os_error(),
                    Some(9)
                );
            }
            if let PocketIcStartupFailure::BuilderPanicked { source } = error.failure() {
                assert_eq!(
                    error
                        .source()
                        .unwrap()
                        .downcast_ref::<super::super::PocketIcOperationError>(),
                    Some(source)
                );
            }
            let displayed = error.to_string();
            assert!(displayed.starts_with(&primary_text));
            for marker in [
                "original-stdout",
                "original-stderr",
                "command cleanup also failed",
                "server cleanup also failed",
            ] {
                assert!(displayed.contains(marker));
            }
        }
        let error = PocketIcStartupError::new(PocketIcStartupFailure::NotConfigured);
        assert_eq!(error.to_string(), error.failure().to_string());
        assert!(error.command_cleanup().is_none());
        assert!(error.server_cleanup().is_none());
    }

    #[cfg(unix)]
    #[test]
    fn builder_failures_finalize_the_owned_server_and_keep_output() {
        for disconnected in [false, true] {
            let script = TestServerScript::new(
                "builder-failure",
                "#!/bin/sh\nprintf '%s\n%s\n' \"$$\" \"$2\"\nprintf 'builder-diagnostic' >&2\nprintf '34567\n' > \"$2\"\nexec sleep 30\n",
            );
            let managed = PocketIcStartupConfig::Spawn {
                server_binary: script.path(),
                timeout: Duration::from_secs(2),
                server_hard_ttl: None,
                server_idle_ttl: None,
                server_output_files: None,
            }
            .start_managed_server()
            .unwrap();
            let pid = managed.process_id();
            let directory = managed.server.files.directory.path().to_path_buf();
            let (sender, receiver) = mpsc::channel();
            if !disconnected {
                sender
                    .send(Err(super::super::PocketIcOperationError::new(
                        "original-builder-panic",
                    )))
                    .unwrap();
            }
            drop(sender);
            let error = super::await_builder(
                receiver,
                "http://127.0.0.1:34567/",
                &super::StartupDeadline {
                    started: Instant::now(),
                    deadline: Instant::now() + Duration::from_secs(2),
                    timeout: Duration::from_secs(2),
                },
                &mut Some(managed.server),
            )
            .err()
            .unwrap();
            assert!(matches!(
                (disconnected, error.failure()),
                (true, PocketIcStartupFailure::BuilderDisconnected)
                    | (false, PocketIcStartupFailure::BuilderPanicked { .. })
            ));
            assert_eq!(error.output().stderr(), "builder-diagnostic");
            assert!(error.server_cleanup().is_none());
            assert!(!directory.exists());
            assert_eq!(process_state(pid), None);
        }
    }

    #[cfg(unix)]
    #[test]
    fn port_io_and_command_spawn_failures_finalize_the_owned_server() {
        for port_io in [true, false] {
            let port = if port_io {
                "mkfifo \"$2\""
            } else {
                "printf '34567\n' > \"$2\""
            };
            let script = TestServerScript::new(
                "owned-io-failure",
                &format!(
                    "#!/bin/sh\nprintf '%s\n%s\n' \"$$\" \"$2\"\nprintf 'original-server-diagnostic' >&2\n{port}\nexec sleep 30\n"
                ),
            );
            let config = PocketIcStartupConfig::Spawn {
                server_binary: script.path(),
                timeout: Duration::from_secs(2),
                server_hard_ttl: None,
                server_idle_ttl: None,
                server_output_files: None,
            };
            let error = if port_io {
                config.start_managed_server().err().unwrap()
            } else {
                config
                    .run_command(&mut Command::new("/missing/ic-testkit-command"), || false)
                    .unwrap_err()
            };
            let PocketIcStartupFailure::Io { source, .. } = error.failure() else {
                panic!("expected original IO failure: {error:?}");
            };
            assert_eq!(
                source.kind(),
                if port_io {
                    std::io::ErrorKind::InvalidData
                } else {
                    std::io::ErrorKind::NotFound
                }
            );
            assert_eq!(error.output().stderr(), "original-server-diagnostic");
            let mut lines = error.output().stdout().lines();
            let pid = lines.next().unwrap().parse().unwrap();
            let port = PathBuf::from(lines.next().unwrap());
            assert_eq!(process_state(pid), None);
            assert!(!port.parent().unwrap().exists());
            assert!(error.server_cleanup().is_none());
        }
    }

    #[cfg(unix)]
    #[test]
    fn command_cancellation_before_startup_has_no_spawn_effects() {
        let error = PocketIcStartupConfig::Spawn {
            server_binary: ("/missing/server").into(),
            timeout: Duration::from_secs(1),
            server_hard_ttl: None,
            server_idle_ttl: None,
            server_output_files: None,
        }
        .run_command(&mut Command::new("/missing/command"), || true)
        .unwrap_err();
        assert!(
            matches!(error.failure(), PocketIcStartupFailure::CommandRun { source, .. } if source.kind() == std::io::ErrorKind::Interrupted)
        );
        assert_eq!(error.output().stdout(), "");
        assert_eq!(error.output().stderr(), "");
        assert!(error.command_cleanup().is_none());
        assert!(error.server_cleanup().is_none());
    }

    #[cfg(unix)]
    #[test]
    fn cancellation_callback_panic_still_reaps_the_owned_command() {
        let script = TestServerScript::new(
            "cancel-panic",
            "#!/bin/sh\nprintf '%s' \"$$\" > \"$1\"\nexec sleep 30\n",
        );
        let pid_file = script.path().with_extension("pid");
        let result = std::panic::catch_unwind(|| {
            PocketIcStartupConfig::Connect {
                server_url: ("http://127.0.0.1:12345/").into(),
                timeout: Duration::from_secs(1),
            }
            .run_command(Command::new(script.path()).arg(&pid_file), || {
                assert!(
                    !fs::read_to_string(&pid_file).is_ok_and(|value| !value.is_empty()),
                    "caller cancellation failed"
                );
                false
            })
        });
        assert!(result.is_err());
        let pid = fs::read_to_string(&pid_file).unwrap().parse().unwrap();
        assert!(process_state(pid).is_none_or(|state| state == 'Z'));
        fs::remove_file(pid_file).unwrap();
    }

    #[cfg(unix)]
    #[test]
    fn environment_selection_prefers_urls_and_fails_closed() {
        use std::os::unix::ffi::OsStringExt as _;

        let timeout = Duration::from_secs(1);
        let config = PocketIcStartupConfig::from_environment(timeout, |name| {
            Some(
                if name == "IC_TESTKIT_POCKET_IC_URL" {
                    "http://127.0.0.1:12345/"
                } else {
                    "/missing/server"
                }
                .into(),
            )
        })
        .unwrap();
        assert!(
            matches!(config, PocketIcStartupConfig::Connect { ref server_url, .. } if server_url == "http://127.0.0.1:12345/")
        );
        assert!(matches!(
            PocketIcStartupConfig::from_environment(timeout, |_| None)
                .as_ref()
                .map_err(PocketIcStartupError::failure),
            Err(PocketIcStartupFailure::NotConfigured)
        ));
        assert!(matches!(
            PocketIcStartupConfig::from_environment(timeout, |_| Some("".into()))
                .as_ref()
                .map_err(PocketIcStartupError::failure),
            Err(PocketIcStartupFailure::InvalidEnvironment {
                variable: "IC_TESTKIT_POCKET_IC_URL"
            })
        ));
        assert!(matches!(
            PocketIcStartupConfig::from_environment(timeout, |_| Some("bad URL".into()))
                .as_ref()
                .map_err(PocketIcStartupError::failure),
            Err(PocketIcStartupFailure::InvalidServerUrl { .. })
        ));
        assert!(matches!(
            PocketIcStartupConfig::from_environment(timeout, |_| Some(
                std::ffi::OsString::from_vec(vec![0xff])
            ))
            .as_ref()
            .map_err(PocketIcStartupError::failure),
            Err(PocketIcStartupFailure::InvalidEnvironment { .. })
        ));
    }

    #[cfg(unix)]
    #[test]
    fn version_capture_cleans_wrapper_descendants_on_exit_and_timeout() {
        for timeout in [false, true] {
            let qualified = format!(
                "printf 'pocket-ic-server {}\\n'",
                crate::pic::POCKET_IC_SERVER_VERSION
            );
            let ending = if timeout { "wait" } else { &qualified };
            let script = TestServerScript::new(
                "version-group",
                &format!(
                    "#!/bin/sh\nsleep 30 >/dev/null 2>&1 &\nprintf '%s' \"$!\" > \"$0.pid\"\n{ending}\n"
                ),
            );
            let result = PocketIcStartupConfig::from_environment(Duration::from_secs(1), |name| {
                (name == "POCKET_IC_BIN").then(|| script.path().into_os_string())
            });
            if timeout {
                assert!(matches!(
                    result.as_ref().map_err(PocketIcStartupError::failure),
                    Err(PocketIcStartupFailure::ServerVersionProbe { .. })
                ));
            } else {
                assert!(result.is_ok());
            }
            let pid_file = script.path().with_extension("pid");
            let pid = fs::read_to_string(&pid_file).unwrap().parse().unwrap();
            let deadline = Instant::now() + Duration::from_secs(3);
            while process_state(pid).is_some_and(|state| state != 'Z') {
                assert!(
                    Instant::now() < deadline,
                    "version wrapper descendant remained alive"
                );
                std::thread::sleep(Duration::from_millis(20));
            }
            fs::remove_file(pid_file).unwrap();
        }
    }

    #[cfg(unix)]
    #[test]
    fn owned_builder_panic_retains_server_output_and_reaps_child() {
        let script = TestServerScript::new(
            "builder-panic",
            "#!/bin/sh\nprintf '%s' \"$$\" > \"$0.pid\"\nprintf 'builder stdout'\nprintf 'builder stderr' >&2\nprintf '34567\\n' > \"$2\"\nexec sleep 30\n",
        );
        let error = PocketIcStartupConfig::Spawn {
            server_binary: script.path(),
            timeout: Duration::from_secs(2),
            server_hard_ttl: None,
            server_idle_ttl: None,
            server_output_files: None,
        }
        .build_instance(PocketIcBuilder::new())
        .err()
        .unwrap();
        let PocketIcStartupFailure::BuilderPanicked { source } = error.failure() else {
            panic!("expected builder failure, got {error:?}");
        };
        assert_ne!(source.message(), "");
        assert_eq!(
            std::error::Error::source(&error)
                .unwrap()
                .downcast_ref::<super::super::PocketIcOperationError>(),
            Some(source)
        );
        assert_eq!(error.output().stdout(), "builder stdout");
        assert_eq!(error.output().stderr(), "builder stderr");
        assert!(error.server_cleanup().is_none());
        let pid_file = script.path().with_extension("pid");
        let pid = fs::read_to_string(&pid_file).unwrap().parse().unwrap();
        assert_eq!(process_state(pid), None);
        fs::remove_file(pid_file).unwrap();
    }

    #[cfg(unix)]
    #[test]
    fn environment_binary_selection_uses_bounded_shared_version_capture() {
        let qualified = format!(
            "printf 'pocket-ic-server {}\\n'",
            crate::pic::POCKET_IC_SERVER_VERSION
        );
        let failed = format!("{qualified}; exit 23");
        for (label, body, expected) in [
            ("qualified", qualified.as_str(), 0),
            ("compatible-16.0", "printf 'pocket-ic-server 16.0.0\\n'", 0),
            ("compatible-patch", "printf 'pocket-ic-server 16.0.1\\n'", 0),
            ("wrong-version", "printf 'pocket-ic-server 15.0.0\\n'", 1),
            ("future-major", "printf 'pocket-ic-server 17.0.0\\n'", 1),
            ("malformed-version", "printf 'pocket-ic-server 16.1\\n'", 1),
            ("failed-version", failed.as_str(), 2),
            ("invalid-utf8", "printf '\\377'", 1),
            ("version-timeout", "exec sleep 30", 2),
        ] {
            let script = TestServerScript::new(
                label,
                &format!("#!/bin/sh\n[ \"$1\" = --version ] || exit 99\n{body}\n"),
            );
            // A successful process launch needs scheduling headroom on loaded CI
            // hosts; only the deliberately stalled probe needs a short deadline.
            let timeout = if label == "version-timeout" {
                Duration::from_millis(200)
            } else {
                Duration::from_secs(5)
            };
            let result = PocketIcStartupConfig::from_environment(timeout, |name| {
                (name == "POCKET_IC_BIN").then(|| script.path().into_os_string())
            });
            match (
                expected,
                result.as_ref().map_err(PocketIcStartupError::failure),
            ) {
                (0, Ok(config)) => {
                    let binary = script.path().canonicalize().unwrap();
                    assert!(
                        matches!(config, PocketIcStartupConfig::Spawn { server_binary, .. } if server_binary == &binary)
                    );
                }
                (1, Err(PocketIcStartupFailure::ServerVersionMismatch { .. }))
                | (2, Err(PocketIcStartupFailure::ServerVersionProbe { .. })) => {}
                (_, result) => panic!("unexpected {label} result: {result:?}"),
            }
        }
    }

    #[cfg(unix)]
    fn process_state(pid: u32) -> Option<char> {
        ic_host_process::testing::process_state(pid).expect("inspect managed test process state")
    }

    #[cfg(unix)]
    #[test]
    fn reading_large_sparse_server_output_is_bounded() {
        let (files, _, _) = StartupFiles::create(None).expect("allocate startup files");
        let mut file = fs::File::create(&files.stdout).expect("create sparse log");
        file.write_all(b"server started\n").expect("write prefix");
        let size = 8_u64 * 1024 * 1024 * 1024;
        file.set_len(size).expect("extend sparse log");
        let output = super::read_bounded_lossy(&files.stdout);
        assert!(output.starts_with("server started\n"));
        assert!(output.ends_with(&format!(
            "<truncated {} bytes>",
            size - super::SERVER_OUTPUT_LIMIT as u64
        )));
        assert!(output.len() < super::SERVER_OUTPUT_LIMIT + 100);
    }

    #[cfg(unix)]
    #[test]
    fn startup_readers_reject_fifos_without_waiting_for_a_writer() {
        let (files, stdout, stderr) = StartupFiles::create(None).expect("allocate startup files");
        drop((stdout, stderr));
        fs::remove_file(&files.stdout).unwrap();
        fs::remove_file(&files.stderr).unwrap();
        assert!(
            Command::new("mkfifo")
                .args([&files.port, &files.stdout, &files.stderr])
                .status()
                .expect("create FIFO startup files")
                .success()
        );
        for path in [&files.port, &files.stdout, &files.stderr] {
            // A delayed writer bounds a blocked read and records whether the
            // reader needed it. Completion cancels the writer; with no reader,
            // a nonblocking open fails and is retried if the reader starts late.
            let fifo = path.clone();
            let (stop_writer, stopped) = mpsc::channel();
            let writer = std::thread::spawn(move || {
                loop {
                    match stopped.recv_timeout(Duration::from_millis(200)) {
                        Ok(()) | Err(mpsc::RecvTimeoutError::Disconnected) => return false,
                        Err(mpsc::RecvTimeoutError::Timeout) => {}
                    }
                    if fs::OpenOptions::new()
                        .write(true)
                        .custom_flags(libc::O_NONBLOCK)
                        .open(&fifo)
                        .is_ok()
                    {
                        return true;
                    }
                }
            });
            let result = if path == &files.port {
                Some(files.read_port())
            } else {
                assert_eq!(super::read_bounded_lossy(path), "");
                None
            };
            let _ = stop_writer.send(());
            assert!(
                !writer.join().expect("join delayed FIFO writer"),
                "startup reader waited for a writer: {}",
                path.display(),
            );
            if let Some(result) = result {
                assert!(matches!(
                    result.as_ref().map_err(PocketIcStartupError::failure),
                    Err(PocketIcStartupFailure::Io { source, .. })
                        if source.kind() == std::io::ErrorKind::InvalidData
                ));
            }
        }
    }

    #[test]
    fn port_file_readiness_preserves_partial_writes_and_rejects_oversized_contents() {
        let (files, _, _) = StartupFiles::create(None).expect("allocate startup files");
        assert!(matches!(
            files.read_port().unwrap(),
            super::PortFileState::Pending
        ));
        for contents in ["", "34567"] {
            fs::write(&files.port, contents).unwrap();
            assert!(matches!(
                files.read_port().unwrap(),
                super::PortFileState::Pending
            ));
        }
        for (contents, expected) in [("1\n", 1), ("65535\n", 65535), (" 34567\r\n", 34567)] {
            fs::write(&files.port, contents).unwrap();
            assert!(matches!(
                files.read_port().unwrap(),
                super::PortFileState::Ready(port) if port == expected
            ));
        }
        for contents in ["0\n", "65536\n", "invalid\n", "1\n2\n"] {
            fs::write(&files.port, contents).unwrap();
            assert!(matches!(
                files.read_port().unwrap(),
                super::PortFileState::Invalid(_)
            ));
        }
        fs::write(&files.port, [0xff, b'\n']).unwrap();
        assert!(matches!(
            files.read_port().as_ref().map_err(PocketIcStartupError::failure),
            Err(PocketIcStartupFailure::Io { source, .. })
                if source.kind() == std::io::ErrorKind::InvalidData
        ));
        for contents in ["1\n".to_owned() + &" ".repeat(128), "0".repeat(128)] {
            fs::write(&files.port, contents).unwrap();
            assert!(
                matches!(files.read_port().unwrap(), super::PortFileState::Invalid(_)),
                "oversized port contents must fail even without a newline",
            );
        }
        // A large backing file must not enlarge the returned diagnostic.
        fs::File::options()
            .write(true)
            .open(&files.port)
            .unwrap()
            .set_len(1024 * 1024)
            .unwrap();
        assert!(matches!(
            files.read_port().unwrap(),
            super::PortFileState::Invalid(value) if value.len() < 256
        ));
    }

    #[test]
    fn startup_config_requires_positive_bounds() {
        let config = PocketIcStartupConfig::Spawn {
            server_binary: "pocket-ic".into(),
            timeout: Duration::from_secs(1),
            server_hard_ttl: None,
            server_idle_ttl: Some(Duration::from_millis(1)),
            server_output_files: None,
        };
        assert!(matches!(
            config.validate().unwrap_err().failure(),
            PocketIcStartupFailure::InvalidConfiguration { .. }
        ));
        let error = PocketIcStartupConfig::Connect {
            server_url: ("http://127.0.0.1:1/").into(),
            timeout: Duration::ZERO,
        }
        .validate()
        .expect_err("zero startup timeout must fail");
        assert!(matches!(
            error.failure(),
            PocketIcStartupFailure::InvalidConfiguration { .. }
        ));

        let error = PocketIcStartupConfig::Spawn {
            server_binary: ("pocket-ic").into(),
            timeout: Duration::from_secs(1),
            server_hard_ttl: Some(Duration::from_millis(1)),
            server_idle_ttl: None,
            server_output_files: None,
        }
        .validate()
        .expect_err("subsecond server hard TTL must fail");
        assert!(matches!(
            error.failure(),
            PocketIcStartupFailure::InvalidConfiguration { .. }
        ));
    }

    #[test]
    fn startup_files_leave_the_server_owned_port_path_absent() {
        let (files, stdout, stderr) = StartupFiles::create(None).expect("allocate startup files");
        let directory = files.directory.path().to_path_buf();

        assert!(directory.is_dir());
        assert!(!files.port.exists());
        assert!(files.stdout.is_file());
        assert!(files.stderr.is_file());
        #[cfg(unix)]
        {
            let mode = fs::metadata(&directory)
                .expect("inspect private startup directory")
                .permissions()
                .mode();
            assert_eq!(mode & 0o077, 0);
        }

        drop(stdout);
        drop(stderr);
        drop(files);
        assert!(!directory.exists());
    }

    #[cfg(unix)]
    #[test]
    fn caller_output_files_survive_startup_command_and_cancellation_cleanup() {
        for outcome in [
            "timeout",
            "startup-exit",
            "server-exit",
            "command-exit",
            "cancel",
            "success",
        ] {
            let (owner, _, _) = StartupFiles::create(None).unwrap();
            let stdout = owner.directory.path().join("retained-stdout");
            let stderr = owner.directory.path().join("retained-stderr");
            let ending = match outcome {
                "timeout" => "exec sleep 30",
                "startup-exit" => "exit 41",
                "server-exit" => {
                    "printf '34567\\n' > \"$2\"; while [ ! -s \"$0.command\" ]; do sleep 0.02; done; exit 42"
                }
                _ => "printf '34567\\n' > \"$2\"; exec sleep 30",
            };
            let script = TestServerScript::new(
                outcome,
                &format!(
                    "#!/bin/sh\nprintf '%s\\n%s\\n' \"$$\" \"$2\" > \"$0.pid\"\ndd if=/dev/zero bs=1024 count=20 2>/dev/null\nprintf raw-stdout-end\ndd if=/dev/zero bs=1024 count=20 >&2 2>/dev/null\nprintf raw-stderr-end >&2\n{ending}\n"
                ),
            );
            let pid_file = script.path().with_extension("pid");
            let command_file = script.path().with_extension("command");
            let config = PocketIcStartupConfig::Spawn {
                server_binary: script.path(),
                timeout: Duration::from_millis(if outcome == "timeout" { 1000 } else { 2000 }),
                server_hard_ttl: None,
                server_idle_ttl: None,
                server_output_files: Some(((&stdout).into(), (&stderr).into())),
            };
            if outcome == "timeout" || outcome == "startup-exit" {
                let error = config.start_managed_server().err().unwrap();
                match (outcome, error.failure()) {
                    ("timeout", PocketIcStartupFailure::ReadinessTimeout { .. }) => {
                        assert!(error.output().stdout().contains("truncated"));
                        assert!(!error.output().stderr().contains("raw-stderr-end"));
                    }
                    ("startup-exit", PocketIcStartupFailure::ServerExited { status, .. }) => {
                        assert_eq!(status.code(), Some(41));
                    }
                    (_, error) => panic!("unexpected startup result: {error:?}"),
                }
            } else {
                let command_end = match outcome {
                    "command-exit" => "exit 37",
                    "success" => "exit 0",
                    _ => "exec sleep 30",
                };
                let result = config.run_command(
                    Command::new("/bin/sh")
                        .args([
                            "-c",
                            &format!("printf '%s' \"$$\" > \"$1\"; {command_end}"),
                            "fixture",
                        ])
                        .arg(&command_file),
                    || {
                        outcome == "cancel"
                            && fs::metadata(&command_file).is_ok_and(|m| m.len() > 0)
                    },
                );
                let result = result.as_ref().map_err(PocketIcStartupError::failure);
                match (outcome, result) {
                    ("command-exit", Ok(status)) => assert_eq!(status.code(), Some(37)),
                    ("success", Ok(status)) => assert!(status.success()),
                    ("server-exit", Err(PocketIcStartupFailure::ServerExited { status, .. })) => {
                        assert_eq!(status.code(), Some(42));
                    }
                    ("cancel", Err(PocketIcStartupFailure::CommandRun { source, .. })) => {
                        assert_eq!(source.kind(), std::io::ErrorKind::Interrupted);
                    }
                    (_, result) => panic!("unexpected command result: {result:?}"),
                }
                let pid = fs::read_to_string(&command_file).unwrap().parse().unwrap();
                assert!(process_state(pid).is_none_or(|state| state == 'Z'));
            }
            for (path, suffix) in [(&stdout, b"raw-stdout-end"), (&stderr, b"raw-stderr-end")] {
                let bytes = fs::read(path).unwrap();
                assert_eq!(bytes.len(), 20 * 1024 + suffix.len());
                assert!(bytes.ends_with(suffix));
                assert_eq!(
                    fs::metadata(path).unwrap().permissions().mode() & 0o777,
                    0o600
                );
            }
            let report = fs::read_to_string(&pid_file).unwrap();
            let mut lines = report.lines();
            assert_eq!(process_state(lines.next().unwrap().parse().unwrap()), None);
            assert!(
                !PathBuf::from(lines.next().unwrap())
                    .parent()
                    .unwrap()
                    .exists()
            );
            fs::remove_file(pid_file).unwrap();
        }
    }

    #[cfg(unix)]
    #[test]
    fn caller_output_files_refuse_existing_entries_and_keep_partial_preparation() {
        use std::os::unix::fs::symlink;

        let (owner, _, _) = StartupFiles::create(None).unwrap();
        let stdout = owner.directory.path().join("retained-stdout");
        let stderr = owner.directory.path().join("retained-stderr");
        let unrelated = owner.directory.path().join("unrelated");
        fs::write(&unrelated, b"original").unwrap();
        symlink(&unrelated, &stderr).unwrap();
        let error = StartupFiles::create(Some((stdout.clone(), stderr.clone())))
            .err()
            .unwrap();
        assert!(
            matches!(error.failure(), PocketIcStartupFailure::Io { source, .. } if source.kind() == std::io::ErrorKind::AlreadyExists)
        );
        assert!(stdout.is_file());
        assert_eq!(fs::read(&unrelated).unwrap(), b"original");
        fs::write(&stdout, b"retained attempt").unwrap();
        assert!(StartupFiles::create(Some((stdout.clone(), stderr.clone()))).is_err());
        assert_eq!(fs::read(&stdout).unwrap(), b"retained attempt");
        fs::remove_file(stderr.clone()).unwrap();
        assert!(
            Command::new("mkfifo")
                .arg(&stderr)
                .status()
                .unwrap()
                .success()
        );
        fs::remove_file(stdout.clone()).unwrap();
        assert!(StartupFiles::create(Some((stdout.clone(), stderr))).is_err());
        assert!(stdout.is_file());
    }

    #[cfg(unix)]
    #[test]
    fn managed_startup_reports_an_exited_server_with_bounded_output() {
        let script = TestServerScript::new(
            "exit",
            "#!/bin/sh\nif [ \"$1\" != \"--port-file\" ] || [ -e \"$2\" ]; then exit 97; fi\nprintf 'synthetic server stdout'\nprintf 'synthetic bind failure' >&2\nexit 23\n",
        );

        let result = PocketIcStartupConfig::Spawn {
            server_binary: script.path(),
            timeout: Duration::from_secs(2),
            server_hard_ttl: None,
            server_idle_ttl: None,
            server_output_files: None,
        }
        .build_instance(PocketIcBuilder::new().with_application_subnet());

        let error = result.err().unwrap();
        let PocketIcStartupFailure::ServerExited {
            server_binary,
            status,
            ..
        } = error.failure()
        else {
            panic!("expected a structured server exit, got {error:?}");
        };
        assert_eq!(server_binary, &script.path());
        assert_eq!(status.code(), Some(23));
        assert_eq!(error.output().stdout(), "synthetic server stdout");
        assert_eq!(error.output().stderr(), "synthetic bind failure");
    }

    #[cfg(unix)]
    #[test]
    fn managed_startup_rejects_oversized_port_files_and_cleans_up() {
        let script = TestServerScript::new(
            "oversized-port",
            "#!/bin/sh\nprintf '%s\\n%s\\n' \"$$\" \"$2\"\nprintf '34567\\n%064s' '' > \"$2.pending\"\nmv \"$2.pending\" \"$2\"\nexec sleep 30\n",
        );
        let result = PocketIcStartupConfig::Spawn {
            server_binary: script.path(),
            timeout: Duration::from_secs(2),
            server_hard_ttl: None,
            server_idle_ttl: None,
            server_output_files: None,
        }
        .start_managed_server();
        let error = result.err().unwrap();
        let PocketIcStartupFailure::InvalidServerPort { value, .. } = error.failure() else {
            panic!("oversized port publication must fail readiness");
        };
        assert!(value.contains("port file exceeds"));
        assert!(value.len() < 256);
        let mut lines = error.output().stdout().lines();
        let pid = lines.next().unwrap().parse::<u32>().unwrap();
        let port_path = PathBuf::from(lines.next().unwrap());
        assert!(!port_path.parent().unwrap().exists());
        assert_eq!(process_state(pid), None, "failed server must be reaped");
    }

    #[cfg(unix)]
    #[test]
    fn managed_startup_terminates_a_server_that_never_becomes_ready() {
        let script = TestServerScript::new(
            "timeout",
            "#!/bin/sh\nif [ \"$1\" != \"--port-file\" ] || [ -e \"$2\" ]; then exit 97; fi\nexec sleep 30\n",
        );
        let timeout = Duration::from_millis(100);
        let started = Instant::now();

        let result = PocketIcStartupConfig::Spawn {
            server_binary: script.path(),
            timeout,
            server_hard_ttl: None,
            server_idle_ttl: None,
            server_output_files: None,
        }
        .build_instance(PocketIcBuilder::new().with_application_subnet());

        assert!(
            started.elapsed() < Duration::from_secs(2),
            "bounded startup should not wait for the sleeping child"
        );
        let error = result.err().unwrap();
        assert!(
            matches!(error.failure(), PocketIcStartupFailure::ReadinessTimeout {
            server_binary, timeout: actual_timeout,
        } if server_binary == &script.path() && *actual_timeout == timeout)
        );
        assert!(error.server_cleanup().is_none());
    }

    #[cfg(unix)]
    #[test]
    fn managed_server_handle_exposes_process_id_url_output_and_raii_ownership() {
        let script = TestServerScript::new(
            "handle",
            "#!/bin/sh\nif [ \"$1\" != \"--port-file\" ] || [ -e \"$2\" ]; then echo 'unexpected managed server arguments' >&2; exit 97; fi\nprintf 'managed server ready: %s' \"$$\"\nprintf '34567\\n' > \"$2\"\nexec sleep 30\n",
        );

        let server = PocketIcStartupConfig::Spawn {
            server_binary: script.path(),
            timeout: Duration::from_secs(2),
            server_hard_ttl: None,
            server_idle_ttl: None,
            server_output_files: None,
        }
        .start_managed_server()
        .expect("start caller-owned managed server");

        assert_eq!(server.url(), "http://127.0.0.1:34567/");
        assert_eq!(
            server.output().stdout(),
            format!("managed server ready: {}", server.process_id())
        );
        assert_eq!(server.output().stderr(), "");
        let pid = server.process_id();
        assert!(process_state(pid).is_some_and(|state| state != 'Z'));
        drop(server);
        assert_eq!(process_state(pid), None, "owned server must be reaped");
    }

    #[cfg(unix)]
    #[test]
    fn managed_server_cleans_descendants_on_drop_timeout_and_exit() {
        for mode in ["drop", "timeout", "exit"] {
            let publish = if mode == "drop" {
                "printf '34567\\n' > \"$2\"\n"
            } else {
                ""
            };
            let finish = if mode == "exit" {
                "sleep 0.03\nexit 23\n"
            } else {
                "exec sleep 30\n"
            };
            let script = TestServerScript::new(
                mode,
                &format!("#!/bin/sh\nsleep 30 &\nprintf '%s' \"$!\"\n{publish}{finish}"),
            );
            let result = PocketIcStartupConfig::Spawn {
                server_binary: script.path(),
                timeout: Duration::from_millis(300),
                server_hard_ttl: None,
                server_idle_ttl: None,
                server_output_files: None,
            }
            .start_managed_server();
            let output = match result {
                Ok(server) => {
                    let output = server.output().stdout().to_owned();
                    drop(server);
                    output
                }
                Err(error) => {
                    match error.failure() {
                        PocketIcStartupFailure::ReadinessTimeout { .. } => {
                            assert_eq!(mode, "timeout");
                            assert!(error.server_cleanup().is_none());
                        }
                        PocketIcStartupFailure::ServerExited { status, .. } => {
                            assert_eq!(mode, "exit");
                            assert_eq!(status.code(), Some(23));
                        }
                        _ => panic!("unexpected {mode} startup result: {error}"),
                    }
                    error.output().stdout().to_owned()
                }
            };
            let pid = output
                .parse::<u32>()
                .expect("server published its descendant PID");
            let deadline = Instant::now() + Duration::from_secs(2);
            while process_state(pid).is_some_and(|state| state != 'Z') {
                assert!(
                    Instant::now() < deadline,
                    "{mode} left its descendant running"
                );
                std::thread::sleep(Duration::from_millis(10));
            }
        }
    }

    #[cfg(unix)]
    #[test]
    fn managed_server_passes_an_explicit_hard_ttl() {
        let script = TestServerScript::new(
            "hard-ttl",
            "#!/bin/sh\nif [ \"$1\" != \"--hard-ttl\" ] || [ \"$2\" != \"17\" ] || [ \"$3\" != \"--port-file\" ] || [ -e \"$4\" ]; then exit 97; fi\nprintf '34567\\n' > \"$4\"\nexec sleep 30\n",
        );

        let server = PocketIcStartupConfig::Spawn {
            server_binary: script.path(),
            timeout: Duration::from_secs(2),
            server_hard_ttl: Some(Duration::from_secs(17)),
            server_idle_ttl: None,
            server_output_files: None,
        }
        .start_managed_server()
        .expect("start managed server with an explicit hard TTL");

        assert_eq!(server.url(), "http://127.0.0.1:34567/");
    }

    #[cfg(unix)]
    #[test]
    #[ignore = "requires POCKET_IC_BIN=<caller-provided PocketIC server binary>"]
    fn caller_provided_server_publishes_port_constructs_instance_and_cleans_up() {
        let binary = std::env::var_os("POCKET_IC_BIN")
            .map(PathBuf::from)
            .expect("set POCKET_IC_BIN to the exact server binary");
        // Observe this exact child's port path without inspecting other startup
        // directories created concurrently.
        let wrapper = TestServerScript::new(
            "real-owned-instance",
            "#!/bin/sh\nprintf '%s\n%s\n' \"$$\" \"$4\" > \"$0.record\"\nexec \"$POCKET_IC_BIN\" \"$@\"\n",
        );
        let owned = PocketIcStartupConfig::Spawn {
            server_binary: wrapper.path(),
            timeout: Duration::from_secs(30),
            server_hard_ttl: Some(Duration::from_secs(60)),
            server_idle_ttl: None,
            server_output_files: None,
        }
        .build_instance(PocketIcBuilder::new().with_application_subnet())
        .expect("owned managed spawn must construct an instance");
        let record_file = wrapper.path().with_extension("record");
        let record = fs::read_to_string(&record_file).unwrap();
        let mut fields = record.lines();
        let pid = fields.next().unwrap().parse().unwrap();
        let port_path = PathBuf::from(fields.next().unwrap());
        let startup_directory = port_path.parent().unwrap();
        drop(owned);
        assert!(!startup_directory.exists());
        assert_eq!(process_state(pid), None);
        fs::remove_file(record_file).unwrap();

        let server = PocketIcStartupConfig::Spawn {
            server_binary: (&binary).into(),
            timeout: Duration::from_secs(30),
            server_hard_ttl: Some(Duration::from_secs(60)),
            server_idle_ttl: None,
            server_output_files: None,
        }
        .start_managed_server()
        .expect("caller-provided PocketIC server must publish its port");
        let files = &server.server.files;
        let startup_directory = files.directory.path().to_path_buf();

        assert!(files.port.is_file());
        let pocket_ic = PocketIcStartupConfig::Connect {
            server_url: (server.url()).into(),
            timeout: Duration::from_secs(30),
        }
        .build_instance(PocketIcBuilder::new().with_application_subnet())
        .expect("construct instance through caller-provided server");

        drop(pocket_ic);
        drop(server);
        assert!(!startup_directory.exists());
    }

    #[cfg(unix)]
    struct TestServerScript {
        path: PathBuf,
        #[expect(
            dead_code,
            reason = "retains directory ownership until cleanup or drop"
        )]
        directory: OwnedDirectory,
    }

    #[cfg(unix)]
    impl TestServerScript {
        fn new(label: &str, contents: &str) -> Self {
            let directory = OwnedDirectory::new(
                create_unique_directory(
                    &std::env::temp_dir(),
                    &format!("ic-testkit-pocket-ic-{label}"),
                    0o700,
                    None,
                )
                .expect("create test server directory"),
            );
            let path = directory.path().join("server");
            write_executable_script(&path, contents).expect("write executable test script");
            Self { path, directory }
        }

        fn path(&self) -> PathBuf {
            self.path.clone()
        }
    }
}