Functions for interacting with the external world through ports.

Ports provide a mechanism to start operating system processes external to the Erlang VM and communicate with them via message passing.

Example

iex> port = Port.open({:spawn, "cat"}, [:binary])
iex> send(port, {self(), {:command, "hello"}})
iex> send(port, {self(), {:command, "world"}})
iex> flush()
{#Port<0.1444>, {:data, "hello"}}
{#Port<0.1444>, {:data, "world"}}
iex> send(port, {self(), :close})
:ok
iex> flush()
{#Port<0.1464>, :closed}
:ok

In the example above, we have created a new port that executes the program cat. cat is a program available on Unix-like operating systems that receives data from multiple inputs and concatenates them in the output.

After the port was created, we sent it two commands in the form of messages using send/2. The first command has the binary payload of "hello" and the second has "world".

After sending those two messages, we invoked the IEx helper flush(), which printed all messages received from the port, in this case we got "hello" and "world" back. Note that the messages are in binary because we passed the :binary option when opening the port in Port.open/2. Without such option, it would have yielded a list of bytes.

Once everything was done, we closed the port.

Elixir provides many conveniences for working with ports and some drawbacks. We will explore those below.

Message and function APIs

There are two APIs for working with ports. It can be either asynchronous via message passing, as in the example above, or by calling the functions on this module.

The messages supported by ports and their counterpart function APIs are listed below:

• {pid, {:command, binary}} - sends the given data to the port. See command/3.

• {pid, :close} - closes the port. Unless the port is already closed, the port will reply with {port, :closed} message once it has flushed its buffers and effectively closed. See close/1.

• {pid, {:connect, new_pid}} - sets the new_pid as the new owner of the port. Once a port is opened, the port is linked and connected to the caller process and communication to the port only happens through the connected process. This message makes new_pid the new connected processes. Unless the port is dead, the port will reply to the old owner with {port, :connected}. See connect/2.

On its turn, the port will send the connected process the following messages:

• {port, {:data, data}} - data sent by the port
• {port, :closed} - reply to the {pid, :close} message
• {port, :connected} - reply to the {pid, {:connect, new_pid}} message
• {:EXIT, port, reason} - exit signals in case the port crashes. If reason is not :normal, this message will only be received if the owner process is trapping exits

Open mechanisms

The port can be opened through four main mechanisms.

As a short summary, prefer to using the :spawn and :spawn_executable options mentioned below. The other two options, :spawn_driver and :fd are for advanced usage within the VM. Also consider using System.cmd/3 if all you want is to execute a program and retrieve its return value.

Windows argument splitting and untrusted arguments

On Unix systems, arguments are passed to a new operating system process as an array of strings but on Windows it is up to the child process to parse them and some Windows programs may apply their own rules, which are inconsistent with the standard C runtime argv parsing

This is particularly troublesome when invoking .bat or .com files as these run implicitly through cmd.exe, whose argument parsing is vulnerable to malicious input and can be used to run arbitrary shell commands.

Therefore, if you are running on Windows and you execute batch files or .com applications, you must not pass untrusted input as arguments to the program. You may avoid accidentally executing them by explicitly passing the extension of the program you want to run, such as .exe, and double check the program is indeed not a batch file or .com application.

This affects both spawn and spawn_executable.

spawn

The :spawn tuple receives a binary that is going to be executed as a full invocation. For example, we can use it to invoke "echo hello" directly:

iex> port = Port.open({:spawn, "echo hello"}, [:binary])
iex> flush()
{#Port<0.1444>, {:data, "hello\n"}}

:spawn will retrieve the program name from the argument and traverse your operating system $PATH environment variable looking for a matching program.

Although the above is handy, it means it is impossible to invoke an executable that has whitespaces on its name or in any of its arguments. For those reasons, most times it is preferable to execute :spawn_executable.

spawn_executable

Spawn executable is a more restricted and explicit version of spawn. It expects full file paths to the executable you want to execute. If they are in your $PATH, they can be retrieved by calling System.find_executable/1:

iex> path = System.find_executable("echo")
iex> port = Port.open({:spawn_executable, path}, [:binary, args: ["hello world"]])
iex> flush()
{#Port<0.1380>, {:data, "hello world\n"}}

When using :spawn_executable, the list of arguments can be passed via the :args option as done above. For the full list of options, see the documentation for the Erlang function :erlang.open_port/2.

fd

The :fd name option allows developers to access in and out file descriptors used by the Erlang VM. You would use those only if you are reimplementing core part of the Runtime System, such as the :user and :shell processes.

Zombie operating system processes

A port can be closed via the close/1 function or by sending a {pid, :close} message. However, if the VM crashes, a long-running program started by the port will have its stdin and stdout channels closed but it won't be automatically terminated.

While most Unix command line tools will exit once its communication channels are closed, not all command line applications will do so. You can easily check this by starting the port and then shutting down the VM and inspecting your operating system to see if the port process is still running.

While we encourage graceful termination by detecting if stdin/stdout has been closed, we do not always have control over how third-party software terminates. In those cases, you can wrap the application in a script that checks for stdin. Here is such script that has been verified to work on bash shells:

#!/usr/bin/env bash

# Start the program in the background
exec "$@" &
pid1=$!

# Silence warnings from here on
exec >/dev/null 2>&1

# Read from stdin in the background and
# kill running program when stdin closes
exec 0<&0 $(
  while read; do :; done
  kill -KILL $pid1
) &
pid2=$!

# Clean up
wait $pid1
ret=$?
kill -KILL $pid2
exit $ret

Note the program above hijacks stdin, so you won't be able to communicate with the underlying software via stdin (on the positive side, software that reads from stdin typically terminates when stdin closes).

Now instead of:

Port.open(
  {:spawn_executable, "/path/to/program"},
  args: ["a", "b", "c"]
)

You may invoke:

Port.open(
  {:spawn_executable, "/path/to/wrapper"},
  args: ["/path/to/program", "a", "b", "c"]
)