Managing external processes from R is not trivial, and this class aims to help with this deficiency. It is essentially a small wrapper around the system base R function, to return the process id of the started process, and set its standard output and error streams. The process id is then used to manage the process.

Usage

p <- process$new(command = NULL, args,
                 stdin = NULL, stdout = NULL, stderr = NULL,
                 connections = list(), poll_connection = NULL,
                 env = NULL, cleanup = TRUE, cleanup_tree = FALSE,
                 wd = NULL, echo_cmd = FALSE, supervise = FALSE,
                 windows_verbatim_args = FALSE,
                 windows_hide_window = FALSE,
                 encoding = "", post_process = NULL)
    p$is_alive()
p$signal(signal)
p$interrupt()
p$kill(grace = 0.1)
p$kill_tree(grace = 0.1)
p$wait(timeout = -1)
p$get_pid()
p$get_exit_status()
p$get_start_time()
    p$read_output(n = -1)
p$read_error(n = -1)
p$read_output_lines(n = -1)
p$read_error_lines(n = -1)
p$has_input_connection()
p$has_output_connection()
p$has_error_connection()
p$has_poll_connection()
p$get_input_connection()
p$get_output_connection()
p$get_error_connection()
p$get_poll_connection()
p$is_incomplete_output()
p$is_incomplete_error()
p$read_all_output()
p$read_all_error()
p$read_all_output_lines()
p$read_all_error_lines()
p$write_input(str, sep = "\n")
p$get_input_file()
p$get_output_file()
p$get_error_file()
    p$poll_io(timeout)
    p$get_result()
    p$as_ps_handle()
p$get_name()
p$get_exe()
p$get_cmdline()
p$get_status()
p$get_username()
p$get_wd()
p$get_cpu_times()
p$get_memory_info()
p$suspend()
p$resume()
    format(p)
print(p)

Arguments

  • p: process object.

  • command: Character scalar, the command to run. Note that this argument is not passed to a shell, so no tilde-expansion or variable substitution is performed on it. It should not be quoted with base::shQuote(). See base::normalizePath() for tilde-expansion.

  • args: Character vector, arguments to the command. They will be used as is, without a shell. They don't need to be escaped.

  • stdin: What to do with the standard input. Possible values: NULL: set to the null device, i.e. no standard input is provided; a string, supply the specified string as standard input; "|": create a (writeable) connection for stdin.

  • stdout: What to do with the standard output. Possible values: NULL: discard it; a string, redirect it to this file; "|": create a connection for it.

  • stderr: What to do with the standard error. Possible values: NULL: discard it; a string, redirect it to this file; "|": create a connection for it.

  • connections: A list of connections to pass to the child process. This is an experimental feature currently.

  • poll_connection: Whether to create an extra connection to the process that allows polling, even if the standard input and standard output are not pipes. If this is NULL (the default), then this connection will be only created if standard output and standard error are not pipes, and connections is an empty list. If the poll connection is created, you can query it via p$get_poll_connection() and it is also included in the response to p$poll_io() and poll(). The numeric file descriptor of the poll connection comes right after stderr (2), and the connections listed in connections.

  • env: Environment variables of the child process. If NULL, the parent's environment is inherited. On Windows, many programs cannot function correctly if some environment variables are not set, so we always set HOMEDRIVE, HOMEPATH, LOGONSERVER, PATH, SYSTEMDRIVE, SYSTEMROOT, TEMP, USERDOMAIN, USERNAME, USERPROFILE and WINDIR.

  • cleanup: Whether to kill the process when the process object is garbage collected.

  • cleanup_tree: Whether to kill the process and its child process tree when the process object is garbage collected.

  • wd: working directory of the process. It must exist. If NULL, then the current working directory is used.

  • echo_cmd: Whether to print the command to the screen before running it.

  • supervise: Whether to register the process with a supervisor. If TRUE, the supervisor will ensure that the process is killed when the R process exits.

  • windows_verbatim_args: Whether to omit quoting the arguments on Windows. It is ignored on other platforms.

  • windows_hide_window: Whether to hide the application's window on Windows. It is ignored on other platforms.

  • signal: An integer scalar, the id of the signal to send to the process. See tools::pskill() for the list of signals.

  • grace: Currently not used.

  • timeout: Timeout in milliseconds, for the wait or the I/O polling.

  • n: Number of characters or lines to read.

  • str: Character or raw vector to write to the standard input of the process. If a character vector with a marked encoding, it will be converted to encoding.

  • sep: Separator to add between str elements if it is a character vector. It is ignored if str is a raw vector.

  • encoding: The encoding to assume for stdin, stdout and stderr. By default the encoding of the current locale is used. Note that processx always reencodes the output of the stdoutand stderr streams in UTF-8 currently. If you want to read them without any conversion, on all platforms, specify "UTF-8" as encoding.

  • post_process: An optional function to run when the process has finished. Currently it only runs if $get_result() is called. It is only run once.

Details

$new() starts a new process in the background, and then returns immediately.

$is_alive() checks if the process is alive. Returns a logical scalar.

$signal() sends a signal to the process. On Windows only the SIGINT, SIGTERM and SIGKILL signals are interpreted, and the special 0 signal, The first three all kill the process. The 0 signal return TRUE if the process is alive, and FALSE otherwise. On Unix all signals are supported that the OS supports, and the 0 signal as well.

$interrupt() sends an interrupt to the process. On Unix this is a SIGINT signal, and it is usually equivalent to pressing CTRL+C at the terminal prompt. On Windows, it is a CTRL+BREAK keypress. Applications may catch these events. By default they will quit.

$kill() kills the process. It also kills all of its child processes, except if they have created a new process group (on Unix), or job object (on Windows). It returns TRUE if the process was killed, and FALSE if it was no killed (because it was already finished/dead when processx tried to kill it).

$kill_tree() performs process tree cleanup, it kills the process (if still alive), together with any child (or grandchild, etc.) processes. It uses the ps package, so that needs to be installed, and ps needs to support the current platform as well. Process tree cleanup works by marking the process with an environment variable, which is inherited in all child processes. This allows finding descendents, even if they are orphaned, i.e. they are not connected to the root of the tree cleanup in the process tree any more. $kill_tree() returns a named integer vector of the process ids that were killed, the names are the names of the processes (e.g. "sleep", "notepad.exe", "Rterm.exe", etc.).

$wait() waits until the process finishes, or a timeout happens. Note that if the process never finishes, and the timeout is infinite (the default), then R will never regain control. It returns the process itself, invisibly. In some rare cases, $wait() might take a bit longer than specified to time out. This happens on Unix, when another package overwrites the processx SIGCHLD signal handler, after the processx process has started. One such package is parallel, if used with fork clusters, e.g. through parallel::mcparallel().

$get_pid() returns the process id of the process.

$get_exit_status returns the exit code of the process if it has finished and NULL otherwise. On Unix, in some rare cases, the exit status might be NA. This happens if another package (or R itself) overwrites the processx SIGCHLD handler, after the processx process has started. In these cases processx cannot determine the real exit status of the process. One such package is parallel, if used with fork clusters, e.g. through the parallel::mcparallel() function.

$get_start_time() returns the time when the process was started.

$is_supervised() returns whether the process is being tracked by supervisor process.

$supervise() if passed TRUE, tells the supervisor to start tracking the process. If FALSE, tells the supervisor to stop tracking the process. Note that even if the supervisor is disabled for a process, if it was started with cleanup = TRUE, the process will still be killed when the object is garbage collected.

$read_output() reads from the standard output connection of the process. If the standard output connection was not requested, then then it returns an error. It uses a non-blocking text connection. This will work only if stdout="|" was used. Otherwise, it will throw an error.

$read_error() is similar to $read_output, but it reads from the standard error stream.

$read_output_lines() reads lines from standard output connection of the process. If the standard output connection was not requested, then then it returns an error. It uses a non-blocking text connection. This will work only if stdout="|" was used. Otherwise, it will throw an error.

$read_error_lines() is similar to $read_output_lines, but it reads from the standard error stream.

$has_input_connection() return TRUE if there is a connection object for standard input; in other words, if stdout="|". It returns FALSE otherwise.

$has_output_connection() returns TRUE if there is a connection object for standard output; in other words, if stdout="|". It returns FALSE otherwise.

$has_error_connection() returns TRUE if there is a connection object for standard error; in other words, if stderr="|". It returns FALSE otherwise.

$has_poll_connection() return TRUE if there is a poll connection, FALSE otherwise.

$get_input_connection() returns a connection object, to the standard input stream of the process.

$get_output_connection() returns a connection object, to the standard output stream of the process.

$get_error_conneciton() returns a connection object, to the standard error stream of the process.

$get_poll_connetion() returns the poll connection, if the process has one.

$is_incomplete_output() return FALSE if the other end of the standard output connection was closed (most probably because the process exited). It return TRUE otherwise.

$is_incomplete_error() return FALSE if the other end of the standard error connection was closed (most probably because the process exited). It return TRUE otherwise.

$read_all_output() waits for all standard output from the process. It does not return until the process has finished. Note that this process involves waiting for the process to finish, polling for I/O and potentically several readLines() calls. It returns a character scalar. This will return content only if stdout="|" was used. Otherwise, it will throw an error.

$read_all_error() waits for all standard error from the process. It does not return until the process has finished. Note that this process involves waiting for the process to finish, polling for I/O and potentically several readLines() calls. It returns a character scalar. This will return content only if stderr="|" was used. Otherwise, it will throw an error.

$read_all_output_lines() waits for all standard output lines from a process. It does not return until the process has finished. Note that this process involves waiting for the process to finish, polling for I/O and potentically several readLines() calls. It returns a character vector. This will return content only if stdout="|" was used. Otherwise, it will throw an error.

$read_all_error_lines() waits for all standard error lines from a process. It does not return until the process has finished. Note that this process involves waiting for the process to finish, polling for I/O and potentically several readLines() calls. It returns a character vector. This will return content only if stderr="|" was used. Otherwise, it will throw an error.

$write_input() writes the character vector (separated by sep) to the standard input of the process. It will be converted to the specified encoding. This operation is non-blocking, and it will return, even if the write fails (because the write buffer is full), or if it suceeds partially (i.e. not the full string is written). It returns with a raw vector, that contains the bytes that were not written. You can supply this raw vector to $write_input() again, until it is fully written, and then the return value will be raw(0) (invisibly).

$get_input_file() if the stdin argument was a filename, this returns the absolute path to the file. If stdin was "|" or NULL, this simply returns that value.

$get_output_file() if the stdout argument was a filename, this returns the absolute path to the file. If stdout was "|" or NULL, this simply returns that value.

$get_error_file() if the stderr argument was a filename, this returns the absolute path to the file. If stderr was "|" or NULL, this simply returns that value.

$poll_io() polls the process's connections for I/O. See more in the Polling section, and see also the poll() function to poll on multiple processes.

$get_result() returns the result of the post processesing function. It can only be called once the process has finished. If the process has no post-processing function, then NULL is returned.

$as_ps_handle() returns a ps::ps_handle object, corresponding to the process. The following methods use the ps package on a ps_handle object, created automatically:

format(p) or p$format() creates a string representation of the process, usually for printing.

print(p) or p$print() shows some information about the process on the screen, whether it is running and it's process id, etc.

Polling

The poll_io() function polls the standard output and standard error connections of a process, with a timeout. If there is output in either of them, or they are closed (e.g. because the process exits) poll_io() returns immediately.

In addition to polling a single process, the poll() function can poll the output of several processes, and returns as soon as any of them has generated output (or exited).

Examples

# CRAN does not like long-running examples
# NOT RUN { p <- process$new("sleep", "2") p$is_alive() p p$kill() p$is_alive() p <- process$new("sleep", "2") p$is_alive() Sys.sleep(3) p$is_alive() # }