init f creates a new pipe, applies f to its writer end, and returns its reader end. init closes the writer end when the result of f becomes determined. If f raises, the writer end is closed and the exception is raised to the caller of init.

init_reader is symmetric. It creates a new pipe, applies f to its reader end, and returns its writer end. init calls close_read when the result of f becomes determined or if f raises, and any exception is raised to the caller of init_reader.

create () creates a new pipe. It is preferable to use init or init_reader instead of create, since they provide exception handling and automatic closing of the pipe.

of_list l returns a closed pipe reader filled with the contents of l.

unfold ~init ~f returns a pipe that it fills with 'as by repeatedly applying f to values of the state type 's. When f returns None, the resulting pipe is closed. unfold respects pushback on the resulting pipe.

For example, to create a pipe of natural numbers:

        Pipe.unfold ~init:0 ~f:(fun n -> return (Some (n, n+1)))
      

of_sequence sequence returns a pipe reader that gets filled with the elements of sequence. of_sequence respects pushback on the resulting pipe.

close t closes the write end of the pipe:

• Future write attempts will fail, raising an exception.

• If, at the time of the close, there are reads blocked waiting for data, these reads will unblock, producing `Eof.

• Future read attempts will drain the data that was in the pipe at the time of the close, until the pipe's buffer has been exhausted; subsequent reads will immediately get `Eof.

Thus, after a pipe has been closed, reads never block.

Close is idempotent.

close_read t closes both the read and write ends of the pipe. It does everything close does, and in addition:

• all pending flushes become determined with `Reader_closed.
• the pipe buffer is cleared.
• all subsequent reads will get `Eof

is_closed t returns true iff close t or close_read t has been called.

closed t returns a deferred that becomes determined when close t or close_read t is called.

Deferreds returned by upstream_flushed and downstream_flushed become determined when all values written prior to the call have been consumed, or if the reader end of the pipe is closed. The difference between "upstream" and "downstream" comes if one has a chain of pipes that are linked (e.g. by Pipe.map):

      P1 --> P2 --> P3
    

Calling downstream_flushed P2 ensures that everything in P2 has made it out of P3. Calling upstream_flushed P2 ensures that everything in P1 has made it out of P3. More generally, downstream_flushed starts at the current pipe and follows the chain to the final downstream consumer(s). upstream_flushed follows the chain to the initial upstream pipe(s), and then calls downstream_flushed.

For a pipe in isolation, "consumed" means "read from the pipe". However, for pipes linked together with transfer or any function built from transfer, "consumed" means "propagated all the way downstream through the chain and read from the final pipe in the chain". Furthermore, for a pipe ultimately connected to an Async.Writer, "consumed" means the OS write() system call has completed on the bytes read from the final pipe in the chain.

The following Pipe functions automatically link their input and output pipes together so that *_flushed on upstream pipes will propagate to downstream pipes: transfer*, map*, filter_map*, filter, interleave, concat. There is *not* automatic linking with iter*; however, user code can customize the behavior of flush functions using Consumer, see below.

add_consumer reader ~downstream_flushed creates a new consumer of reader, and causes future calls to flushed_downstream reader to take this consumer into account. Thereafter, Pipe.flushed_downstream reader will first ensure that values previously written to reader have been read, then that they have been sent downstream by the consumer that read them, and finally that they have been flushed downstream.

One should only supply the resulting consumer to read operations on reader. Using a consumer created from one reader with another reader will raise an exception.

length t returns the number of elements currently queued in t

is_empty t is true iff there are no values in the pipe.

pushback writer becomes determined when either writer has been closed or the pipe can accept a new write.

write writer a enqueues a in writer, returning a pushback deferred, as described above.

transfer_in writer ~from:q transfers the elements from q into writer, leaving q empty, and returning a pushback deferred.

write_without_pushback and transfer_in_without_pushback are alternatives to transfer_in and write that can be used when you don't care about the pushback deferred. They add data to the pipe and return immediately.

The following equivalences hold:

• write t a = write_without_pushback t a; pushback t
• transfer_in t ~from = transfer_in_without_pushback t ~from; pushback t

If is_closed writer, then all of these functions raise.

write_when_ready writer ~f waits until there is space available in the pipe, and then calls f write, where write can be used by f to write a single value into the pipe at a time. write_when_ready guarantees that the pipe is open when it calls f, and hence that the writes will succeed, unless f itself closes the pipe.

write_if_open w e is equivalent to:

      let x = e in
      if not (is_closed w) then write w x else Deferred.unit
    

Note the difference in allocation and potential side effects when w is closed and e is a complex expression.

write_without_pushback_if_open is the same as write_if_open, except it calls write_without_pushback instead of write.

read' pipe reads values available in the pipe, as soon as any value becomes available. The resulting queue will satisfy 0 < Queue.length q <= max_queue_length. read' raises if max_queue_length <= 0. The consumer is used to extend the meaning of values being flushed (see the Consumer module above).

read pipe reads a single value from the pipe. The consumer is used to extend the meaning of values being flushed (see the Consumer module above).

read_at_most t ~num_values is read' t ~max_queue_length:num_values.

read_exactly r ~num_values reads exactly num_values items, unless EOF is encountered. read_exactly performs a sequence of read_at_most operations, so there is no guarantee that the queue of values it returns comprise a contiguous segment of the written stream of values -- other readers might pick off elements in-between read_exactly's atomic reads. read_exactly raises if num_values <= 0. The consumer is used to extend the meaning of values being flushed (see the Consumer module above).

read_now' reader reads values from reader that are immediately available. The resulting queue will satisfy 0 <= Q.length q <= max_queue_length. If reader is closed, read_now' returns `Eof. If reader is empty, read_now' returns `Nothing_available. The consumer is used to extend the meaning of values being flushed (see the Consumer module above).

read_now is like read_now', except that it reads a single value rather than everything that is available.

read_now_at_most t ~num_values is read_now' t ~max_queue_length:num_values

clear reader consumes all of the values currently in reader, and all blocked flushes become determined with `Ok.

read_all reader reads all the values from the pipe until it is closed. An alternative name might be Reader.to_queue.

values_available reader returns a deferred that becomes determined when there are values in the pipe. If there are multiple readers (a rare situation), there is no guarantee that some other reader hasn't become active because of ordinary Async scheduling and removed some or all of the values between the time the result of values_available becomes determined and the time something waiting upon that result runs.

values_available is useful when one wants to choose on values being available in a pipe, so that one can be sure and not remove values and drop them on the floor.

values_available is roughly equivalent to read' ~max_queue_length:0.

read_choice reader is:

      choice
        (values_available reader)
        (fun (_ : [ `Ok | `Eof ]) -> read_now reader)
    

read_choice consumes a value from reader iff the choice is taken. read_choice exists to discourage the broken idiom:

      choice (read reader) (fun ...)
    

which is broken because it reads from reader even if the choice isn't taken. `Nothing_available can only be returned if there is a race condition with one or more other consumers.

read_choice_single_consumer_exn reader [%here] is like read_choice reader, but it raises in the case of `Nothing_available. It is intended to be used when reader has no other consumers.

fold' reader ~init ~f reads a batch of elements from reader, supplies them to f, waits for f to finish, and then repeats. fold' finishes when the call to f on the final batch of elements from reader finishes.

fold reader ~init ~f folds over the elements of reader, consuming them as they come in. fold finishes when the final call to f returns.

iter' reader ~f repeatedly applies f to batches of elements of reader, waiting for each call to f to finish before continuing. The deferred returned by iter' becomes determined when the call to f on the final batch of elements finishes. ~continue_on_error:true causes the iteration to continue even if f raises. ~consumer is used to extend the meaning of values being flushed (see the Consumer module above).

iter t f is a specialization of iter' that applies the f to each element in the batch, waiting for one call to f to finish before making the next call to f.

iter_without_pushback t ~f applies f to each element in t, without giving f a chance to pushback on the iteration continuing. If f raises on some element of t, iter_without_pushback will not consume any further elements. iter_without_pushback will not make more than max_iterations_per_job calls to f in a single Async_job; this can be used to increase Async-scheduling fairness.

transfer' input output ~f repeatedly reads a batch of elements from input, applies f to the batch, writes the result as a batch to output, and then waits on pushback in output before continuing. transfer' finishes if input is closed or output is closed. If output is closed, then transfer' closes input.

transfer is like transfer', except that it processes one element at time.

transfer_id is a specialization of transfer' with f = Fn.id.

map' input ~f returns a reader, output, and repeatedly applies f to batches of elements from input, with the results appearing in output. If values are not being consumed from output, map' will pushback and stop consuming values from input. If output is closed, then map' will close input.

map is like map', except that it processes one element at time.

filter_map' input ~f returns a reader, output, and repeatedly applies f to elements from input, with the results that aren't None appearing in output. If values are not being consumed from output, filter_map' will pushback and stop consuming values from input. If output is closed, then filter_map' will close input.

filter_map is a specialized version of filter_map'.

filter input ~f returns a reader, output, and copies to output each element from input that satisfies the predicate f. If output is closed, then filter closes input.

interleave inputs returns a reader output, and, for each input, transfers batches of values from that input to output, using transfer_id. Each input is transferred to output independently. So, batches of values from different inputs can be in flight to output simultaneously, but at most one batch at a time from any particular input. The operation is complete when either all the inputs produce EOF, or when output is closed by the downstream consumer (in which case interleave closes all the inputs).

merge inputs ~cmp returns a reader, output, that merges all the inputs. Assuming that for each input, values are sorted according to the comparison function cmp, values for each input will be transfered to output and the values returned by output will be sorted according to cmp.

concat inputs return a reader, output, with the values from each pipe in inputs in sequence. concat closes output once it reaches EOF on the final input. If output is closed, then concat closes all its inputs.

to_stream_deprecated reader returns a stream that reads everything from the pipe. This function is deprecated because one should change the code that is consuming a stream to instead consume from a pipe reader.

of_stream_deprecated reader return a pipe that has one element for every element on the stream. This function is deprecated because one should change the code that is producing a stream to instead produce a pipe reader.

drain reader repeatedly reads values from reader and throws them away.

drain_and_count is like drain, except it also counts the number of values it has read.

to_list input reads everything from input; on EOF, it produces the accumulated list of these values.

hash a hash function suitable for pipes

equal on pipes is physical equality.

Every pipe has a "size budget", which governs the pushback that is used to discourage writers from enqueueing arbitrarily large amounts of data. As long as the length of the pipe exceeds the size budget, writers will not be notified to do further writing. Whenever the length is less than or equal to the size budget, writers will be notified to continue.

Every pipe's initial size budget is zero.

set_size_budget t i changes the size budget of t to i. Any nonnegative value is allowed.

show_debug_messages, if true will cause a message to be printed at the start of each operation, showing the pipe and other arguments.

check_invariant, if true, will cause pipes' invariants to be checked at the start of each operation.