Code Monkey home page Code Monkey logo

async's Introduction

Overview

async is a C library for Linux and macOS that provides:

  • a single-threaded main loop that dispatches callback events
  • edge-triggered file-descriptor monitoring (using Linux epoll(2) or BSD kqueue(2) system calls)
  • timers
  • TCP client and server connections
  • other useful byte streams

Building

async uses SCons and pkg-config for building and depends on the following libraries:

Before building async for the first time, run

git submodule update --init

To build async, run

scons [ prefix=<prefix> ]

from the top-level async directory. The prefix argument is a directory, /usr/local by default, where the build system searches for async dependencies and installs async.

To install async, run

sudo scons [ prefix=<prefix> ] install

The Structure of an Async Application

Here is the skeleton of a simple async application:

#include <async/async.h>

typedef struct {
    async_t *async;
    int fd;
    /*...*/
} my_app_t;

static void process_event(my_app_t *app)
{
    char buf[100];
    ssize_t count = read(app->fd, buf, sizeof buf);
    /*...*/
}

static void initialize(my_app_t *app)
{
    app->fd = my_channel_open(/*...*/);
    action_1 callback = { app, (act_1) process_event };
    int status = async_register(app->async, app->fd, callback);
    if (status < 0) {
        /* see errno */
    }
    async_execute(app->async, callback);    /* the first time's on us */
}

int main(void)
{
    my_app_t app;
    app.async = make_async();
    initialize(&app);
    async_loop(app.async);
    destroy_async(app.async);   /* if you want */
    return 0;
}

Notes:

  • A single async object is created for the lifetime of the process.
  • The async_loop() function blocks "for ever" (use async_quit_loop() to exit the loop).
  • One or more file descriptors are handed to async with async_register(). Only socket-like ("selectable") file descriptors are suitable.
  • Registration makes the file descriptor nonblocking as a side effect.
  • Event handlers must never sleep or block but must return ASAP. Use state machines to manage control flow.
  • A file descriptor is only guaranteed a notification when its I/O state changes after registration. Initially, it is considered readable and writable. If read(), recv*(), accept(), write(), send*() or connect() returns with EAGAIN or EINPROGRESS, the I/O state changes and the application can rely on a callback as soon as the file operation should be tried again. That is why the sample application above uses async_execute() to schedule the callback during initialization.
  • The application is responsible for exhausting its inputs before a callback is guaranteed. However, care must be taken so the application does not spin in a loop servicing one file descriptor for ever. At times, control should be relinquished to other activities. Resume reading the input by calling async_execute() before returning from the callback.
  • Callbacks may be spurious.
  • async_loop() returns a negative integer (with errno set) in case of an I/O error. In particular, a signal will make async_loop() return with EINTR.

Timers and Tasks

You start a timer with

async_timer_t *my_timer =
    async_timer_start(app->async, async_now(app->async) + 5 * ASYNC_S,
    (action_1) { app, (act_1) callback });

which causes callback(app) to be called by async 5 seconds from now. Note that the expiry is expressed as a point in time instead of a delay; async_now() returns an unsigned, non-wrapping, ever-increasing nanosecond counter. If it refers to the past, the timer expires immediately—however, the callback is always invoked from async_loop() and not from async_timer_start().

A running timer can be canceled with async_timer_cancel(async, timer). Canceling a non-running timer is a sure way to crash the process.

If you want a task executed right away, call async_execute(async, callback). A direct function call is more immediate, of course, but "backgrounding" tasks often has benefits: you can complete state transitions before the scheduled action is taken and multiple activities are interleaved fairly by the main loop.

Like async_timer_start(), async_execute() returns a timer object. It is usually ignored but could be used by the application to cancel the pending action before it is executed.

Notifications

The <async/notification.h> module provides a notification object that wraps an action and can safely schedule it from a signal handler or separate thread. You create a notification with

notification_t *my_notification =
    make_notification(app->async, (action_1) { app, (act_1) callback });

The notification can be issued at any time with issue_notification(my_notification) and can be destroyed with destroy_notification(my_notification).

Byte Streams and Yields

async includes a collection of byte stream and yield types. The types are implemented in C++'esque C which allows for interfaces and virtual functions.

A byte stream type implements the <async/bytestream_1.h> interface. That is, any byte stream object can be "typecast" into a bytestream_1 object with a function. Thus, a chunk encoder object is converted into a bytestream_1 object with chunkencoder_as_bytestream_1(encoder) and a pacer stream object is converted with pacerstream_as_bytestream_1(pacer).

(Interface types sport a numeric tag. An interface never changes. If a change is needed, the numeric tag is incremented, and both versions of the interface remain valid.)

The bytestream_1 interface has four methods:

  • ssize_t read(void *stream, void *buf, size_t count)

  • void close(void *stream)

  • void register_callback(void *stream, action_1 action)

  • void unregister_callback(void *stream)

Each byte stream type additionally contains a constructor to create an object of the type, and the bytestream_1 methods as functions prefixed with the type name.

The semantics of the methods is mostly obvious. The close() function acts as a destructor for the object. (Closed stream objects become inactive immediately and are deallocated by the main loop.)

Byte streams are often chained much like Unix pipelines for similarly diverse effects.

A yield is a sequence of arbitrary data objects, typically driven by I/O events. A yield type implements the <async/yield_1.h> interface. That is, any yield object can be "typecast" into a yield_1 object with a function.

Similarly to the byte stream interface, the yield_1 interface has four methods:

  • void *receive(void *yield)

  • void close(void *yield)

  • void register_callback(void *yield, action_1 action)

  • void unregister_callback(void *yield)

The receive method returns an object whose type depends on the specific yield. Each yield type additionally contains a constructor to create an object of the type, and the yield_1 methods as functions prefixed with the type name.

A yield is usually implemented on top of a byte stream that decodes a single object out of a source stream.

Base64 encoder

<async/base64encoder.h>

A byte stream that encodes another stream in Base64 encoding.

Base64 decoder

<async/base64decoder.h>

A byte stream that decodes another stream in Base64 encoding.

Blob stream

<async/blobstream.h>

A byte array delivered as a byte stream.

Blocking stream

<async/blockingstream.h>

Any blocking file descriptor (e.g., a regular file) dressed as a byte stream; use with care.

Chunk encoder

<async/chunkencoder.h>

A byte stream that encodes another stream in the chunked transfer encoding format.

Chunk decoder

<async/chunkdecoder.h>

A byte stream that decodes another stream in the chunked transfer encoding format.

Clobber stream

<async/clobberstream.h>

A byte stream that corrupts another stream.

Concat stream

<async/concatstream.h>

A byte stream that concatenates a fixed number of streams into one (like cat).

Dry stream

<async/drystream.h>

A byte stream that always returns EAGAIN.

Empty stream

<async/emptystream.h>

A byte stream that gives out an immediate EOF (like /dev/null).

Error stream

<async/errorstream.h>

A byte stream that always returns a given error.

Farewell stream

<async/farewellstream.h>

A byte stream that wraps another stream and invokes a callback as soon as close() is called on it.

Iconv stream

<async/iconvstream.h>

A byte stream that converts a character stream from one character encoding to another.

JSON encoder

<async/jsonencoder.h>

A byte stream that encodes a JSON object.

Multipart decoder

<async/multipartdecoder.h>

A byte stream that decodes one part of a RFC 2046 multipart body stream.

Naive encoder

async/naiveencoder.h>

A byte stream that encodes another stream by appending a terminator byte and optionally escaping special bytes.

Naive decoder

async/naivedecoder.h>

A byte stream that decodes another stream in the "naive" encoding.

Nice stream

<async/nicestream.h>

A byte stream that wraps another stream and occasionally returns an EAGAIN if the stream seems to be readable for ever.

Pacer stream

<async/pacerstream.h>

A byte stream that wraps another stream and gives out bytes at a steady rate. Allows for an implementation of periodic timers (every read byte corresponds to a timer tick).

Pause stream

<async/pausestream.h>

A blocking stream that returns EAGAIN when a given number of bytes has been read. Reading can be resumed by raising the limit.

Pipe stream

<async/pipestream.h>

Any nonblocking file descriptor dressed as a byte stream.

Probe stream

<async/probestream.h>

A byte stream that can be used to trace reading and closing activities of another stream.

Queue stream

<async/queuestream.h>

A byte stream that concatenates streams dynamically on the fly.

String stream

<async/stringstream.h>

A C string delivered as a byte stream.

Substream

<async/substream.h>

A byte stream that delivers a portion of another stream (like head and tail).

Switch stream

<async/switchstream.h>

A byte stream that wraps another stream and can switch to a different one on the fly.

Trickle stream

<async/tricklestream.h>

A byte stream that slows down the transmission of another stream to a single byte per a given interval.

Zero stream

<async/zerostream.h>

A byte stream that delivers a never-ending supply of zero bytes (like /dev/zero).

JSON yield

<async/jsonyield.h>

A yield of JSON objects delimited with the NUL byte using ESC as the escape character. The yield input is a bytestream_1 object and the outputs are json_thing_t objects.

Multipart deserializer

<async/multipartdeserializer.h>

A yield of the parts of a RFC 2046 multipart body. The yield input is a bytestream_1 object and the outputs are bytestream_1 objects.

Naive framer

<async/naiveframer.h>

A yield of frames delimited with "naive" encoding. The yield input is a bytestream_1 object and the outputs are bytestream_1 objects.

JSON decoder

The <async/jsondecoder.h> module provides an object that decodes a single JSON object from a byte stream. As opposed to other decoders, the JSON decoder is not a byte stream.

TCP

The <async/tcp_connection.h> module integrates TCP (and Unix stream socket) connections with async and the byte stream mechanism. A TCP connection is presented as a pair of byte streams. The TCP connection supplies a byte stream for receiving bytes from the socket. The application must supply a byte stream for sending bytes to the socket. Thus, bytes are only read, never written. A typical usage pattern is to supply the TCP connection with an outbound queue stream and send messages with the queuestream_enqueue() function.

To create a TCP (or Unix stream socket) client, call tcp_connect(), tcp_register_callback() and tcp_set_output_stream().

To create a TCP (or Unix stream socket) server, call tcp_listen(), tcp_register_server_callback() and tcp_accept().

To create a socket pair, call socketpair() followed by tcp_adopt_connection().

Note that tcp_close() closes both associated byte streams. You can close the byte streams individually using the bytestream_1 facilities, or, use the tcp_shut_down() method. Even after both byte streams have been closed/shut down, tcp_close() needs to be called to free up the resources.

async's People

Contributors

rauhma avatar nikobockerman avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.