Socket#

Added in version 2.22.

class Socket(**properties: Any)#

Superclasses: Object

Implemented Interfaces: DatagramBased, Initable

A GSocket is a low-level networking primitive. It is a more or less direct mapping of the BSD socket API in a portable GObject based API. It supports both the UNIX socket implementations and winsock2 on Windows.

GSocket is the platform independent base upon which the higher level network primitives are based. Applications are not typically meant to use it directly, but rather through classes like SocketClient, SocketService and SocketConnection. However there may be cases where direct use of GSocket is useful.

GSocket implements the Initable interface, so if it is manually constructed by e.g. new you must call init and check the results before using the object. This is done automatically in new and new_from_fd, so these functions can return NULL.

Sockets operate in two general modes, blocking or non-blocking. When in blocking mode all operations (which don’t take an explicit blocking parameter) block until the requested operation is finished or there is an error. In non-blocking mode all calls that would block return immediately with a G_IO_ERROR_WOULD_BLOCK error. To know when a call would successfully run you can call condition_check, or condition_wait. You can also use create_source and attach it to a MainContext to get callbacks when I/O is possible. Note that all sockets are always set to non blocking mode in the system, and blocking mode is emulated in GSocket.

When working in non-blocking mode applications should always be able to handle getting a G_IO_ERROR_WOULD_BLOCK error even when some other function said that I/O was possible. This can easily happen in case of a race condition in the application, but it can also happen for other reasons. For instance, on Windows a socket is always seen as writable until a write returns G_IO_ERROR_WOULD_BLOCK.

GSocket’s can be either connection oriented or datagram based. For connection oriented types you must first establish a connection by either connecting to an address or accepting a connection from another address. For connectionless socket types the target/source address is specified or received in each I/O operation.

All socket file descriptors are set to be close-on-exec.

Note that creating a GSocket causes the signal SIGPIPE to be ignored for the remainder of the program. If you are writing a command-line utility that uses GSocket, you may need to take into account the fact that your program will not automatically be killed if it tries to write to stdout after it has been closed.

Like most other APIs in GLib, GSocket is not inherently thread safe. To use a GSocket concurrently from multiple threads, you must implement your own locking.

Nagle’s algorithm#

Since GLib 2.80, GSocket will automatically set the TCP_NODELAY option on all G_SOCKET_TYPE_STREAM sockets. This disables Nagle’s algorithm <https://en.wikipedia.org/wiki/Nagle``%27s_algorithm`>`_ as it typically does more harm than good on modern networks.

If your application needs Nagle’s algorithm enabled, call set_option after constructing a GSocket to enable it:

socket = g_socket_new (, G_SOCKET_TYPE_STREAM, );
if (socket != NULL)
  {
    g_socket_set_option (socket, IPPROTO_TCP, TCP_NODELAY, FALSE, &local_error);
    // handle error if needed
  }

Constructors#

class Socket
classmethod new(family: SocketFamily, type: SocketType, protocol: SocketProtocol) Socket#

Creates a new Socket with the defined family, type and protocol. If protocol is 0 (DEFAULT) the default protocol type for the family and type is used.

The protocol is a family and type specific int that specifies what kind of protocol to use. SocketProtocol lists several common ones. Many families only support one protocol, and use 0 for this, others support several and using 0 means to use the default protocol for the family and type.

The protocol id is passed directly to the operating system, so you can use protocols not listed in SocketProtocol if you know the protocol number used for it.

Added in version 2.22.

Parameters:
  • family – the socket family to use, e.g. IPV4.

  • type – the socket type to use.

  • protocol – the id of the protocol to use, or 0 for default.

classmethod new_from_fd(fd: int) Socket#

Creates a new Socket from a native file descriptor or winsock SOCKET handle.

This reads all the settings from the file descriptor so that all properties should work. Note that the file descriptor will be set to non-blocking mode, independent on the blocking mode of the Socket.

On success, the returned Socket takes ownership of fd. On failure, the caller must close fd themselves.

Since GLib 2.46, it is no longer a fatal error to call this on a non-socket descriptor. Instead, a GError will be set with code FAILED

Added in version 2.22.

Parameters:

fd – a native socket file descriptor.

Methods#

class Socket
accept(cancellable: Cancellable | None = None) Socket#

Accept incoming connections on a connection-based socket. This removes the first outstanding connection request from the listening socket and creates a Socket object for it.

The socket must be bound to a local address with bind() and must be listening for incoming connections (listen()).

If there are no outstanding connections then the operation will block or return WOULD_BLOCK if non-blocking I/O is enabled. To be notified of an incoming connection, wait for the %G_IO_IN condition.

Added in version 2.22.

Parameters:

cancellable – a %GCancellable or None

bind(address: SocketAddress, allow_reuse: bool) bool#

When a socket is created it is attached to an address family, but it doesn’t have an address in this family. bind() assigns the address (sometimes called name) of the socket.

It is generally required to bind to a local address before you can receive connections. (See listen() and accept() ). In certain situations, you may also want to bind a socket that will be used to initiate connections, though this is not normally required.

If socket is a TCP socket, then allow_reuse controls the setting of the SO_REUSEADDR socket option; normally it should be True for server sockets (sockets that you will eventually call accept() on), and False for client sockets. (Failing to set this flag on a server socket may cause bind() to return ADDRESS_IN_USE if the server program is stopped and then immediately restarted.)

If socket is a UDP socket, then allow_reuse determines whether or not other UDP sockets can be bound to the same address at the same time. In particular, you can have several UDP sockets bound to the same address, and they will all receive all of the multicast and broadcast packets sent to that address. (The behavior of unicast UDP packets to an address with multiple listeners is not defined.)

Added in version 2.22.

Parameters:
  • address – a SocketAddress specifying the local address.

  • allow_reuse – whether to allow reusing this address

check_connect_result() bool#

Checks and resets the pending connect error for the socket. This is used to check for errors when connect() is used in non-blocking mode.

Added in version 2.22.

close() bool#

Closes the socket, shutting down any active connection.

Closing a socket does not wait for all outstanding I/O operations to finish, so the caller should not rely on them to be guaranteed to complete even if the close returns with no error.

Once the socket is closed, all other operations will return CLOSED. Closing a socket multiple times will not return an error.

Sockets will be automatically closed when the last reference is dropped, but you might want to call this function to make sure resources are released as early as possible.

Beware that due to the way that TCP works, it is possible for recently-sent data to be lost if either you close a socket while the %G_IO_IN condition is set, or else if the remote connection tries to send something to you after you close the socket but before it has finished reading all of the data you sent. There is no easy generic way to avoid this problem; the easiest fix is to design the network protocol such that the client will never send data “out of turn”. Another solution is for the server to half-close the connection by calling shutdown() with only the shutdown_write flag set, and then wait for the client to notice this and close its side of the connection, after which the server can safely call close(). (This is what TcpConnection does if you call set_graceful_disconnect(). But of course, this only works if the client will close its connection after the server does.)

Added in version 2.22.

condition_check(condition: IOCondition) IOCondition#

Checks on the readiness of socket to perform operations. The operations specified in condition are checked for and masked against the currently-satisfied conditions on socket. The result is returned.

Note that on Windows, it is possible for an operation to return WOULD_BLOCK even immediately after condition_check() has claimed that the socket is ready for writing. Rather than calling condition_check() and then writing to the socket if it succeeds, it is generally better to simply try writing to the socket right away, and try again later if the initial attempt returns WOULD_BLOCK.

It is meaningless to specify %G_IO_ERR or %G_IO_HUP in condition; these conditions will always be set in the output if they are true.

This call never blocks.

Added in version 2.22.

Parameters:

condition – a IOCondition mask to check

condition_timed_wait(condition: IOCondition, timeout_us: int, cancellable: Cancellable | None = None) bool#

Waits for up to timeout_us microseconds for condition to become true on socket. If the condition is met, True is returned.

If cancellable is cancelled before the condition is met, or if timeout_us (or the socket’s Socket:timeout) is reached before the condition is met, then False is returned and error, if non-None, is set to the appropriate value (CANCELLED or TIMED_OUT).

If you don’t want a timeout, use condition_wait(). (Alternatively, you can pass -1 for timeout_us.)

Note that although timeout_us is in microseconds for consistency with other GLib APIs, this function actually only has millisecond resolution, and the behavior is undefined if timeout_us is not an exact number of milliseconds.

Added in version 2.32.

Parameters:
  • condition – a IOCondition mask to wait for

  • timeout_us – the maximum time (in microseconds) to wait, or -1

  • cancellable – a Cancellable, or None

condition_wait(condition: IOCondition, cancellable: Cancellable | None = None) bool#

Waits for condition to become true on socket. When the condition is met, True is returned.

If cancellable is cancelled before the condition is met, or if the socket has a timeout set and it is reached before the condition is met, then False is returned and error, if non-None, is set to the appropriate value (CANCELLED or TIMED_OUT).

See also condition_timed_wait().

Added in version 2.22.

Parameters:
connect(address: SocketAddress, cancellable: Cancellable | None = None) bool#

Connect the socket to the specified remote address.

For connection oriented socket this generally means we attempt to make a connection to the address. For a connection-less socket it sets the default address for send() and discards all incoming datagrams from other sources.

Generally connection oriented sockets can only connect once, but connection-less sockets can connect multiple times to change the default address.

If the connect call needs to do network I/O it will block, unless non-blocking I/O is enabled. Then PENDING is returned and the user can be notified of the connection finishing by waiting for the G_IO_OUT condition. The result of the connection must then be checked with check_connect_result().

Added in version 2.22.

Parameters:
  • address – a SocketAddress specifying the remote address.

  • cancellable – a %GCancellable or None

connection_factory_create_connection() SocketConnection#

Creates a SocketConnection subclass of the right type for socket.

Added in version 2.22.

get_available_bytes() int#

Get the amount of data pending in the OS input buffer, without blocking.

If socket is a UDP or SCTP socket, this will return the size of just the next packet, even if additional packets are buffered after that one.

Note that on Windows, this function is rather inefficient in the UDP case, and so if you know any plausible upper bound on the size of the incoming packet, it is better to just do a receive() with a buffer of that size, rather than calling get_available_bytes() first and then doing a receive of exactly the right size.

Added in version 2.32.

get_blocking() bool#

Gets the blocking mode of the socket. For details on blocking I/O, see set_blocking().

Added in version 2.22.

get_broadcast() bool#

Gets the broadcast setting on socket; if True, it is possible to send packets to broadcast addresses.

Added in version 2.32.

get_credentials() Credentials#

Returns the credentials of the foreign process connected to this socket, if any (e.g. it is only supported for UNIX sockets).

If this operation isn’t supported on the OS, the method fails with the NOT_SUPPORTED error. On Linux this is implemented by reading the %SO_PEERCRED option on the underlying socket.

This method can be expected to be available on the following platforms:

  • Linux since GLib 2.26

  • OpenBSD since GLib 2.30

  • Solaris, Illumos and OpenSolaris since GLib 2.40

  • NetBSD since GLib 2.42

  • macOS, tvOS, iOS since GLib 2.66

Other ways to obtain credentials from a foreign peer includes the UnixCredentialsMessage type and send_credentials() / receive_credentials() functions.

Added in version 2.26.

get_family() SocketFamily#

Gets the socket family of the socket.

Added in version 2.22.

get_fd() int#

Returns the underlying OS socket object. On unix this is a socket file descriptor, and on Windows this is a Winsock2 SOCKET handle. This may be useful for doing platform specific or otherwise unusual operations on the socket.

Added in version 2.22.

get_keepalive() bool#

Gets the keepalive mode of the socket. For details on this, see set_keepalive().

Added in version 2.22.

get_listen_backlog() int#

Gets the listen backlog setting of the socket. For details on this, see set_listen_backlog().

Added in version 2.22.

get_local_address() SocketAddress#

Try to get the local address of a bound socket. This is only useful if the socket has been bound to a local address, either explicitly or implicitly when connecting.

Added in version 2.22.

get_multicast_loopback() bool#

Gets the multicast loopback setting on socket; if True (the default), outgoing multicast packets will be looped back to multicast listeners on the same host.

Added in version 2.32.

get_multicast_ttl() int#

Gets the multicast time-to-live setting on socket; see set_multicast_ttl() for more details.

Added in version 2.32.

get_option(level: int, optname: int) tuple[bool, int]#

Gets the value of an integer-valued option on socket, as with getsockopt(). (If you need to fetch a non-integer-valued option, you will need to call getsockopt() directly.)

The `<gio/gnetworking.h> <networking.html>`_ header pulls in system headers that will define most of the standard/portable socket options. For unusual socket protocols or platform-dependent options, you may need to include additional headers.

Note that even for socket options that are a single byte in size, value is still a pointer to a int variable, not a guint8; get_option() will handle the conversion internally.

Added in version 2.36.

Parameters:
  • level – the “API level” of the option (eg, SOL_SOCKET)

  • optname – the “name” of the option (eg, SO_BROADCAST)

get_protocol() SocketProtocol#

Gets the socket protocol id the socket was created with. In case the protocol is unknown, -1 is returned.

Added in version 2.22.

get_remote_address() SocketAddress#

Try to get the remote address of a connected socket. This is only useful for connection oriented sockets that have been connected.

Added in version 2.22.

get_socket_type() SocketType#

Gets the socket type of the socket.

Added in version 2.22.

get_timeout() int#

Gets the timeout setting of the socket. For details on this, see set_timeout().

Added in version 2.26.

get_ttl() int#

Gets the unicast time-to-live setting on socket; see set_ttl() for more details.

Added in version 2.32.

is_closed() bool#

Checks whether a socket is closed.

Added in version 2.22.

is_connected() bool#

Check whether the socket is connected. This is only useful for connection-oriented sockets.

If using shutdown(), this function will return True until the socket has been shut down for reading and writing. If you do a non-blocking connect, this function will not return True until after you call check_connect_result().

Added in version 2.22.

join_multicast_group(group: InetAddress, source_specific: bool, iface: str | None = None) bool#

Registers socket to receive multicast messages sent to group. socket must be a DATAGRAM socket, and must have been bound to an appropriate interface and port with bind().

If iface is None, the system will automatically pick an interface to bind to based on group.

If source_specific is True, source-specific multicast as defined in RFC 4604 is used. Note that on older platforms this may fail with a NOT_SUPPORTED error.

To bind to a given source-specific multicast address, use join_multicast_group_ssm() instead.

Added in version 2.32.

Parameters:
  • group – a InetAddress specifying the group address to join.

  • source_specificTrue if source-specific multicast should be used

  • iface – Name of the interface to use, or None

join_multicast_group_ssm(group: InetAddress, source_specific: InetAddress | None = None, iface: str | None = None) bool#

Registers socket to receive multicast messages sent to group. socket must be a DATAGRAM socket, and must have been bound to an appropriate interface and port with bind().

If iface is None, the system will automatically pick an interface to bind to based on group.

If source_specific is not None, use source-specific multicast as defined in RFC 4604. Note that on older platforms this may fail with a NOT_SUPPORTED error.

Note that this function can be called multiple times for the same group with different source_specific in order to receive multicast packets from more than one source.

Added in version 2.56.

Parameters:
  • group – a InetAddress specifying the group address to join.

  • source_specific – a InetAddress specifying the source-specific multicast address or None to ignore.

  • iface – Name of the interface to use, or None

leave_multicast_group(group: InetAddress, source_specific: bool, iface: str | None = None) bool#

Removes socket from the multicast group defined by group, iface, and source_specific (which must all have the same values they had when you joined the group).

socket remains bound to its address and port, and can still receive unicast messages after calling this.

To unbind to a given source-specific multicast address, use leave_multicast_group_ssm() instead.

Added in version 2.32.

Parameters:
  • group – a InetAddress specifying the group address to leave.

  • source_specificTrue if source-specific multicast was used

  • iface – Interface used

leave_multicast_group_ssm(group: InetAddress, source_specific: InetAddress | None = None, iface: str | None = None) bool#

Removes socket from the multicast group defined by group, iface, and source_specific (which must all have the same values they had when you joined the group).

socket remains bound to its address and port, and can still receive unicast messages after calling this.

Added in version 2.56.

Parameters:
  • group – a InetAddress specifying the group address to leave.

  • source_specific – a InetAddress specifying the source-specific multicast address or None to ignore.

  • iface – Name of the interface to use, or None

listen() bool#

Marks the socket as a server socket, i.e. a socket that is used to accept incoming requests using accept().

Before calling this the socket must be bound to a local address using bind().

To set the maximum amount of outstanding clients, use set_listen_backlog().

Added in version 2.22.

receive(cancellable: Cancellable | None = None) tuple[int, bytes]#

Receive data (up to size bytes) from a socket. This is mainly used by connection-oriented sockets; it is identical to receive_from() with address set to None.

For DATAGRAM and SEQPACKET sockets, receive() will always read either 0 or 1 complete messages from the socket. If the received message is too large to fit in buffer, then the data beyond size bytes will be discarded, without any explicit indication that this has occurred.

For STREAM sockets, receive() can return any number of bytes, up to size. If more than size bytes have been received, the additional data will be returned in future calls to receive().

If the socket is in blocking mode the call will block until there is some data to receive, the connection is closed, or there is an error. If there is no data available and the socket is in non-blocking mode, a WOULD_BLOCK error will be returned. To be notified when data is available, wait for the %G_IO_IN condition.

On error -1 is returned and error is set accordingly.

Added in version 2.22.

Parameters:

cancellable – a %GCancellable or None

receive_bytes(size: int, timeout_us: int, cancellable: Cancellable | None = None) Bytes#

Receives data (up to size bytes) from a socket.

This function is a variant of receive which returns a Bytes rather than a plain buffer.

Pass -1 to timeout_us to block indefinitely until data is received (or the connection is closed, or there is an error). Pass 0 to use the default timeout from timeout, or pass a positive number to wait for that many microseconds for data before returning G_IO_ERROR_TIMED_OUT.

Added in version 2.80.

Parameters:
  • size – the number of bytes you want to read from the socket

  • timeout_us – the timeout to wait for, in microseconds, or -1 to block indefinitely

  • cancellable – a %GCancellable, or NULL

receive_bytes_from(size: int, timeout_us: int, cancellable: Cancellable | None = None) tuple[Bytes, SocketAddress]#

Receive data (up to size bytes) from a socket.

This function is a variant of receive_from which returns a Bytes rather than a plain buffer.

If address is non-None then address will be set equal to the source address of the received packet.

The address is owned by the caller.

Pass -1 to timeout_us to block indefinitely until data is received (or the connection is closed, or there is an error). Pass 0 to use the default timeout from timeout, or pass a positive number to wait for that many microseconds for data before returning G_IO_ERROR_TIMED_OUT.

Added in version 2.80.

Parameters:
  • size – the number of bytes you want to read from the socket

  • timeout_us – the timeout to wait for, in microseconds, or -1 to block indefinitely

  • cancellable – a Cancellable, or NULL

receive_from(cancellable: Cancellable | None = None) tuple[int, SocketAddress, bytes]#

Receive data (up to size bytes) from a socket.

If address is non-None then address will be set equal to the source address of the received packet. address is owned by the caller.

See receive() for additional information.

Added in version 2.22.

Parameters:

cancellable – a %GCancellable or None

receive_message(vectors: Sequence[InputVector], cancellable: Cancellable | None = None) tuple[int, SocketAddress, list[SocketControlMessage], int]#

Receive data from a socket. For receiving multiple messages, see receive_messages(); for easier use, see receive() and receive_from().

If address is non-None then address will be set equal to the source address of the received packet. address is owned by the caller.

vector must point to an array of InputVector structs and num_vectors must be the length of this array. These structs describe the buffers that received data will be scattered into. If num_vectors is -1, then vectors is assumed to be terminated by a InputVector with a None buffer pointer.

As a special case, if num_vectors is 0 (in which case, vectors may of course be None), then a single byte is received and discarded. This is to facilitate the common practice of sending a single ‘0’ byte for the purposes of transferring ancillary data.

messages, if non-None, will be set to point to a newly-allocated array of SocketControlMessage instances or None if no such messages was received. These correspond to the control messages received from the kernel, one SocketControlMessage per message from the kernel. This array is None-terminated and must be freed by the caller using free() after calling unref() on each element. If messages is None, any control messages received will be discarded.

num_messages, if non-None, will be set to the number of control messages received.

If both messages and num_messages are non-None, then num_messages gives the number of SocketControlMessage instances in messages (ie: not including the None terminator).

flags is an in/out parameter. The commonly available arguments for this are available in the GSocketMsgFlags enum, but the values there are the same as the system values, and the flags are passed in as-is, so you can pass in system-specific flags too (and receive_message() may pass system-specific flags out). Flags passed in to the parameter affect the receive operation; flags returned out of it are relevant to the specific returned message.

As with receive(), data may be discarded if socket is DATAGRAM or SEQPACKET and you do not provide enough buffer space to read a complete message. You can pass PEEK in flags to peek at the current message without removing it from the receive queue, but there is no portable way to find out the length of the message other than by reading it into a sufficiently-large buffer.

If the socket is in blocking mode the call will block until there is some data to receive, the connection is closed, or there is an error. If there is no data available and the socket is in non-blocking mode, a WOULD_BLOCK error will be returned. To be notified when data is available, wait for the %G_IO_IN condition.

On error -1 is returned and error is set accordingly.

Added in version 2.22.

Parameters:
  • vectors – an array of InputVector structs

  • cancellable – a %GCancellable or None

receive_messages(messages: Sequence[InputMessage], flags: int, cancellable: Cancellable | None = None) int#

Receive multiple data messages from socket in one go. This is the most complicated and fully-featured version of this call. For easier use, see receive(), receive_from(), and receive_message().

messages must point to an array of InputMessage structs and num_messages must be the length of this array. Each InputMessage contains a pointer to an array of InputVector structs describing the buffers that the data received in each message will be written to. Using multiple InputVector is more memory-efficient than manually copying data out of a single buffer to multiple sources, and more system-call-efficient than making multiple calls to receive(), such as in scenarios where a lot of data packets need to be received (e.g. high-bandwidth video streaming over RTP/UDP).

flags modify how all messages are received. The commonly available arguments for this are available in the GSocketMsgFlags enum, but the values there are the same as the system values, and the flags are passed in as-is, so you can pass in system-specific flags too. These flags affect the overall receive operation. Flags affecting individual messages are returned in InputMessage.flags.

The other members of InputMessage are treated as described in its documentation.

If Socket:blocking is True the call will block until num_messages have been received, or the end of the stream is reached.

If Socket:blocking is False the call will return up to num_messages without blocking, or WOULD_BLOCK if no messages are queued in the operating system to be received.

In blocking mode, if Socket:timeout is positive and is reached before any messages are received, TIMED_OUT is returned, otherwise up to num_messages are returned. (Note: This is effectively the behaviour of MSG_WAITFORONE with recvmmsg().)

To be notified when messages are available, wait for the %G_IO_IN condition. Note though that you may still receive WOULD_BLOCK from receive_messages() even if you were previously notified of a %G_IO_IN condition.

If the remote peer closes the connection, any messages queued in the operating system will be returned, and subsequent calls to receive_messages() will return 0 (with no error set).

On error -1 is returned and error is set accordingly. An error will only be returned if zero messages could be received; otherwise the number of messages successfully received before the error will be returned.

Added in version 2.48.

Parameters:
  • messages – an array of InputMessage structs

  • flags – an int containing GSocketMsgFlags flags for the overall operation, which may additionally contain other platform specific flags

  • cancellable – a %GCancellable or None

receive_with_blocking(blocking: bool, cancellable: Cancellable | None = None) tuple[int, bytes]#

This behaves exactly the same as receive(), except that the choice of blocking or non-blocking behavior is determined by the blocking argument rather than by socket’s properties.

Added in version 2.26.

Parameters:
  • blocking – whether to do blocking or non-blocking I/O

  • cancellable – a %GCancellable or None

send(buffer: Sequence[int], cancellable: Cancellable | None = None) int#

Tries to send size bytes from buffer on the socket. This is mainly used by connection-oriented sockets; it is identical to send_to() with address set to None.

If the socket is in blocking mode the call will block until there is space for the data in the socket queue. If there is no space available and the socket is in non-blocking mode a WOULD_BLOCK error will be returned. To be notified when space is available, wait for the %G_IO_OUT condition. Note though that you may still receive WOULD_BLOCK from send() even if you were previously notified of a %G_IO_OUT condition. (On Windows in particular, this is very common due to the way the underlying APIs work.)

On error -1 is returned and error is set accordingly.

Added in version 2.22.

Parameters:
  • buffer – the buffer containing the data to send.

  • cancellable – a %GCancellable or None

send_message(address: SocketAddress | None, vectors: Sequence[OutputVector], messages: Sequence[SocketControlMessage] | None, flags: int, cancellable: Cancellable | None = None) int#

Send data to address on socket. For sending multiple messages see send_messages(); for easier use, see send() and send_to().

If address is None then the message is sent to the default receiver (set by connect()).

vectors must point to an array of OutputVector structs and num_vectors must be the length of this array. (If num_vectors is -1, then vectors is assumed to be terminated by a OutputVector with a None buffer pointer.) The OutputVector structs describe the buffers that the sent data will be gathered from. Using multiple OutputVector is more memory-efficient than manually copying data from multiple sources into a single buffer, and more network-efficient than making multiple calls to send().

messages, if non-None, is taken to point to an array of num_messages SocketControlMessage instances. These correspond to the control messages to be sent on the socket. If num_messages is -1 then messages is treated as a None-terminated array.

flags modify how the message is sent. The commonly available arguments for this are available in the GSocketMsgFlags enum, but the values there are the same as the system values, and the flags are passed in as-is, so you can pass in system-specific flags too.

If the socket is in blocking mode the call will block until there is space for the data in the socket queue. If there is no space available and the socket is in non-blocking mode a WOULD_BLOCK error will be returned. To be notified when space is available, wait for the %G_IO_OUT condition. Note though that you may still receive WOULD_BLOCK from send() even if you were previously notified of a %G_IO_OUT condition. (On Windows in particular, this is very common due to the way the underlying APIs work.)

The sum of the sizes of each OutputVector in vectors must not be greater than %G_MAXSSIZE. If the message can be larger than this, then it is mandatory to use the send_message_with_timeout() function.

On error -1 is returned and error is set accordingly.

Added in version 2.22.

Parameters:
send_message_with_timeout(address: SocketAddress | None, vectors: Sequence[OutputVector], messages: Sequence[SocketControlMessage] | None, flags: int, timeout_us: int, cancellable: Cancellable | None = None) tuple[PollableReturn, int]#

This behaves exactly the same as send_message(), except that the choice of timeout behavior is determined by the timeout_us argument rather than by socket’s properties.

On error FAILED is returned and error is set accordingly, or if the socket is currently not writable WOULD_BLOCK is returned. bytes_written will contain 0 in both cases.

Added in version 2.60.

Parameters:
send_messages(messages: Sequence[OutputMessage], flags: int, cancellable: Cancellable | None = None) int#

Send multiple data messages from socket in one go. This is the most complicated and fully-featured version of this call. For easier use, see send(), send_to(), and send_message().

messages must point to an array of OutputMessage structs and num_messages must be the length of this array. Each OutputMessage contains an address to send the data to, and a pointer to an array of OutputVector structs to describe the buffers that the data to be sent for each message will be gathered from. Using multiple OutputVector is more memory-efficient than manually copying data from multiple sources into a single buffer, and more network-efficient than making multiple calls to send(). Sending multiple messages in one go avoids the overhead of making a lot of syscalls in scenarios where a lot of data packets need to be sent (e.g. high-bandwidth video streaming over RTP/UDP), or where the same data needs to be sent to multiple recipients.

flags modify how the message is sent. The commonly available arguments for this are available in the GSocketMsgFlags enum, but the values there are the same as the system values, and the flags are passed in as-is, so you can pass in system-specific flags too.

If the socket is in blocking mode the call will block until there is space for all the data in the socket queue. If there is no space available and the socket is in non-blocking mode a WOULD_BLOCK error will be returned if no data was written at all, otherwise the number of messages sent will be returned. To be notified when space is available, wait for the %G_IO_OUT condition. Note though that you may still receive WOULD_BLOCK from send() even if you were previously notified of a %G_IO_OUT condition. (On Windows in particular, this is very common due to the way the underlying APIs work.)

On error -1 is returned and error is set accordingly. An error will only be returned if zero messages could be sent; otherwise the number of messages successfully sent before the error will be returned.

Added in version 2.44.

Parameters:
send_to(address: SocketAddress | None, buffer: Sequence[int], cancellable: Cancellable | None = None) int#

Tries to send size bytes from buffer to address. If address is None then the message is sent to the default receiver (set by connect()).

See send() for additional information.

Added in version 2.22.

Parameters:
  • address – a SocketAddress, or None

  • buffer – the buffer containing the data to send.

  • cancellable – a %GCancellable or None

send_with_blocking(buffer: Sequence[int], blocking: bool, cancellable: Cancellable | None = None) int#

This behaves exactly the same as send(), except that the choice of blocking or non-blocking behavior is determined by the blocking argument rather than by socket’s properties.

Added in version 2.26.

Parameters:
  • buffer – the buffer containing the data to send.

  • blocking – whether to do blocking or non-blocking I/O

  • cancellable – a %GCancellable or None

set_blocking(blocking: bool) None#

Sets the blocking mode of the socket. In blocking mode all operations (which don’t take an explicit blocking parameter) block until they succeed or there is an error. In non-blocking mode all functions return results immediately or with a WOULD_BLOCK error.

All sockets are created in blocking mode. However, note that the platform level socket is always non-blocking, and blocking mode is a GSocket level feature.

Added in version 2.22.

Parameters:

blocking – Whether to use blocking I/O or not.

set_broadcast(broadcast: bool) None#

Sets whether socket should allow sending to broadcast addresses. This is False by default.

Added in version 2.32.

Parameters:

broadcast – whether socket should allow sending to broadcast addresses

set_keepalive(keepalive: bool) None#

Sets or unsets the %SO_KEEPALIVE flag on the underlying socket. When this flag is set on a socket, the system will attempt to verify that the remote socket endpoint is still present if a sufficiently long period of time passes with no data being exchanged. If the system is unable to verify the presence of the remote endpoint, it will automatically close the connection.

This option is only functional on certain kinds of sockets. (Notably, TCP sockets.)

The exact time between pings is system- and protocol-dependent, but will normally be at least two hours. Most commonly, you would set this flag on a server socket if you want to allow clients to remain idle for long periods of time, but also want to ensure that connections are eventually garbage-collected if clients crash or become unreachable.

Added in version 2.22.

Parameters:

keepalive – Value for the keepalive flag

set_listen_backlog(backlog: int) None#

Sets the maximum number of outstanding connections allowed when listening on this socket. If more clients than this are connecting to the socket and the application is not handling them on time then the new connections will be refused.

Note that this must be called before listen() and has no effect if called after that.

Added in version 2.22.

Parameters:

backlog – the maximum number of pending connections.

set_multicast_loopback(loopback: bool) None#

Sets whether outgoing multicast packets will be received by sockets listening on that multicast address on the same host. This is True by default.

Added in version 2.32.

Parameters:

loopback – whether socket should receive messages sent to its multicast groups from the local host

set_multicast_ttl(ttl: int) None#

Sets the time-to-live for outgoing multicast datagrams on socket. By default, this is 1, meaning that multicast packets will not leave the local network.

Added in version 2.32.

Parameters:

ttl – the time-to-live value for all multicast datagrams on socket

set_option(level: int, optname: int, value: int) bool#

Sets the value of an integer-valued option on socket, as with setsockopt(). (If you need to set a non-integer-valued option, you will need to call setsockopt() directly.)

The `<gio/gnetworking.h> <networking.html>`_ header pulls in system headers that will define most of the standard/portable socket options. For unusual socket protocols or platform-dependent options, you may need to include additional headers.

Added in version 2.36.

Parameters:
  • level – the “API level” of the option (eg, SOL_SOCKET)

  • optname – the “name” of the option (eg, SO_BROADCAST)

  • value – the value to set the option to

set_timeout(timeout: int) None#

Sets the time in seconds after which I/O operations on socket will time out if they have not yet completed.

On a blocking socket, this means that any blocking Socket operation will time out after timeout seconds of inactivity, returning TIMED_OUT.

On a non-blocking socket, calls to condition_wait() will also fail with TIMED_OUT after the given time. Sources created with create_source() will trigger after timeout seconds of inactivity, with the requested condition set, at which point calling receive(), send(), check_connect_result(), etc, will fail with TIMED_OUT.

If timeout is 0 (the default), operations will never time out on their own.

Note that if an I/O operation is interrupted by a signal, this may cause the timeout to be reset.

Added in version 2.26.

Parameters:

timeout – the timeout for socket, in seconds, or 0 for none

set_ttl(ttl: int) None#

Sets the time-to-live for outgoing unicast packets on socket. By default the platform-specific default value is used.

Added in version 2.32.

Parameters:

ttl – the time-to-live value for all unicast packets on socket

shutdown(shutdown_read: bool, shutdown_write: bool) bool#

Shut down part or all of a full-duplex connection.

If shutdown_read is True then the receiving side of the connection is shut down, and further reading is disallowed.

If shutdown_write is True then the sending side of the connection is shut down, and further writing is disallowed.

It is allowed for both shutdown_read and shutdown_write to be True.

One example where it is useful to shut down only one side of a connection is graceful disconnect for TCP connections where you close the sending side, then wait for the other side to close the connection, thus ensuring that the other side saw all sent data.

Added in version 2.22.

Parameters:
  • shutdown_read – whether to shut down the read side

  • shutdown_write – whether to shut down the write side

speaks_ipv4() bool#

Checks if a socket is capable of speaking IPv4.

IPv4 sockets are capable of speaking IPv4. On some operating systems and under some combinations of circumstances IPv6 sockets are also capable of speaking IPv4. See RFC 3493 section 3.7 for more information.

No other types of sockets are currently considered as being capable of speaking IPv4.

Added in version 2.22.

Properties#

class Socket
props.blocking: bool#

The type of the None singleton.

Added in version 2.22.

props.broadcast: bool#

The type of the None singleton.

Added in version 2.32.

props.family: SocketFamily#

The type of the None singleton.

Added in version 2.22.

props.fd: int#

The type of the None singleton.

Added in version 2.22.

props.keepalive: bool#

The type of the None singleton.

Added in version 2.22.

props.listen_backlog: int#

The type of the None singleton.

Added in version 2.22.

props.local_address: SocketAddress#

The type of the None singleton.

Added in version 2.22.

props.multicast_loopback: bool#

The type of the None singleton.

Added in version 2.32.

props.multicast_ttl: int#

The type of the None singleton.

Added in version 2.32.

props.protocol: SocketProtocol#

The type of the None singleton.

Added in version 2.22.

props.remote_address: SocketAddress#

The type of the None singleton.

Added in version 2.22.

props.timeout: int#

The type of the None singleton.

Added in version 2.26.

props.ttl: int#

The type of the None singleton.

Added in version 2.32.

props.type: SocketType#

The type of the None singleton.

Added in version 2.22.

Fields#

class Socket
parent_instance#
priv#