Struct PipeInner 

pub struct PipeInner { /* private fields */ }

Shared script-pipe backend: buffer plus writer/keeper accounting. Clone the Arc, never the state: every handle aliases one channel.

Implementations

impl PipeInner

pub fn reader_handle(self: &Arc<Self>) -> SharedInput

One shared reader half over this backend. Each call mints a fresh PipeReader over the same buffer, so readers share bytes (not positions) exactly like the registry-era shared input.

pub fn writer_handle(self: &Arc<Self>) -> SharedOutput

One attached writer half over this backend. Dropping the handle detaches, preserving EOF-from-detach semantics.

pub fn writer_count(&self) -> usize

Live data-writer attachments (excludes keeper pins). Used for INSPECT() diagnostics; never blocks.

pub fn buffered_bytes(&self) -> u64

Bytes currently buffered for readers. Used for diagnostics.

pub fn peek_bytes(&self) -> Result<Vec<u8>>

Non-destructive snapshot of buffered bytes for pipe-content assertions. Never waits: returns what is buffered right now.

pub fn force_close(&self)

Explicitly close the pipe: readers drain buffered bytes, then observe EOF regardless of live writers or keeper pins. General primitive (sockets have shutdown, files have close); pipes previously had detach-only EOF. A later writer attachment resurrects the pipe per standard attach semantics, so callers must not reuse closed pipes for new sessions.

pub fn pin_keeper(&self)

Pin a keeper slot so transient writer churn can never observe zero writers. Called synchronously on the spawning thread before an ASYNC worker starts; the returned guard unpins on drop when the worker exits, restoring normal EOF semantics afterwards. Never touches closed: pinning a pipe that already reached EOF must not resurrect it into a blocking pipe.

pub fn unpin_keeper(&self)

Release one keeper slot. When the last transient writer and the last keeper are both gone the pipe closes and blocked readers see EOF.

pub fn read_into_timeout( &self, buf: &mut [u8], backstop: Duration, ) -> Result<Option<usize>>

Timeout-bounded variant of the blocking buffer read for bridge worker loops: returns Ok(None) when the backstop elapses with no data and no close, so cancellation resolves on a tick instead of hanging on a condvar. Bridge-only caller; every DSL reader keeps the blocking read with unchanged semantics.