process.pl -- Create processes and redirect I/O

The module library(process) implements interaction with child processes and unifies older interfaces such as shell/[1,2], open(pipe(command), ...) etc. This library is modelled after SICStus 4.

The main interface is formed by process_create/3. If the process id is requested the process must be waited for using process_wait/2. Otherwise the process resources are reclaimed automatically.

In addition to the predicates, this module defines a file search path (see file_search_path/2 and absolute_file_name/3) named path that locates files on the system's search path for executables. E.g. the following finds the executable for ls:

?- absolute_file_name(path(ls), Path, [access(execute)]).

Incompatibilities and current limitations

Where SICStus distinguishes between an internal process id and the OS process id, this implementation does not make this distinction. This implies that is_process/1 is incomplete and unreliable.
It is unclear what the detached(true) option is supposed to do. Disable signals in the child? Use setsid() to detach from the session? The current implementation uses setsid() on Unix systems.
An extra option env([Name=Value, ...]) is added to process_create/3. As of version 4.1 SICStus added environment(List) which modifies the environment. A compatible option was added to SWI-Prolog 7.7.23.

Compatibility: - SICStus 4
To be done: - Implement detached option in process_create/3

process_create(+Exe, +Args:list, +Options) is det

Create a new process running the file Exe and using arguments from the given list. Exe is a file specification as handed to absolute_file_name/3. Typically one use the path file alias to specify an executable file on the current PATH. Args is a list of arguments that are handed to the new process. On Unix systems, each element in the list becomes a separate argument in the new process. In Windows, the arguments are simply concatenated to form the commandline. Each argument itself is either a primitive or a list of primitives. A primitive is either atomic or a term file(Spec). Using file(Spec), the system inserts a filename using the OS filename conventions which is properly quoted if needed.

Options:

stdin(Spec)

stdout(Spec)

stderr(Spec)

Bind the standard streams of the new process. Spec is one of the terms below. If pipe(Pipe) is used, the Prolog stream is a stream in text-mode using the encoding of the default locale. The encoding can be changed using set_stream/2, or by using the two-argument form of pipe, which accepts an encoding(Encoding) option. The options stdout and stderr may use the same stream, in which case both output streams are connected to the same Prolog stream.

std: Just share with the Prolog I/O streams. On Unix, if the user_input, etc. are bound to a file handle but not to 0,1,2 the process I/O is bound to the file handles of these streams.
null: Bind to a null stream. Reading from such a stream returns end-of-file, writing produces no output
pipe(-Stream)
pipe(-Stream, +StreamOptions): Attach input and/or output to a Prolog stream. The optional StreamOptions argument is a list of options that affect the stream. Currently only the options type(+Type) and encoding(+Encoding) are supported, which have the same meaning as the stream properties of the same name (see stream_property/2). StreamOptions is provided mainly for SICStus compatibility - the SWI-Prolog predicate set_stream/2 can be used for the same purpose.
stream(+Stream): Attach input or output to an existing Prolog stream. This stream must be associated with an OS file handle (see stream_property/2, property file_no). This option is not provided by the SICStus implementation.

cwd(+Directory)

Run the new process in Directory. Directory can be a compound specification, which is converted using absolute_file_name/3. See also process_set_method/1.

env(+List)

As environment(List), but only the specified variables are passed, i.e., no variables are inherited.

environment(+List)

Specify additional environment variables for the new process. List is a list of Name=Value terms, where Value is expanded the same way as the Args argument. If neither env nor environment is passed the environment is inherited from the Prolog process. At most one env(List) or environment(List) term may appear in the options. If multiple appear a permission_error is raised for the second option.

process(-PID)

Unify PID with the process id of the created process.

detached(+Bool)

In Unix: If true, detach the process from the terminal Currently mapped to setsid(); Also creates a new process group for the child In Windows: If true, detach the process from the current job via the CREATE_BREAKAWAY_FROM_JOB flag. In Vista and beyond, processes launched from the shell directly have the 'compatibility assistant' attached to them automatically unless they have a UAC manifest embedded in them. This means that you will get a permission denied error if you try and assign the newly-created PID to a job you create yourself.

window(+Bool)

If true, create a window for the process (Windows only)

priority(+Priority)

In Unix: specifies the process priority for the newly created process. Priority must be an integer between -20 and 19. Positive values are nicer to others, and negative values are less so. The default is zero. Users are free to lower their own priority. Only the super-user may raise it to less-than zero.

If the user specifies the process(-PID) option, he must call process_wait/2 to reclaim the process. Without this option, the system will wait for completion of the process after the last pipe stream is closed.

If the process is not waited for, it must succeed with status 0. If not, an process_error is raised.

Windows notes

On Windows this call is an interface to the CreateProcess() API. The commandline consists of the basename of Exe and the arguments formed from Args. Arguments are separated by a single space. If all characters satisfy iswalnum() it is unquoted. If the argument contains a double-quote it is quoted using single quotes. If both single and double quotes appear a domain_error is raised, otherwise double-quote are used.

The CreateProcess() API has many options. Currently only the CREATE_NO_WINDOW options is supported through the window(+Bool) option. If omitted, the default is to use this option if the application has no console. Future versions are likely to support more window specific options and replace win_exec/2.

Examples

First, a very simple example that behaves the same as shell('ls -l'), except for error handling:

?- process_create(path(ls), ['-l'], []).

The following example uses grep to find all matching lines in a file.

grep(File, Pattern, Lines) :-
        setup_call_cleanup(
            process_create(path(grep), [ Pattern, file(File) ],
                           [ stdout(pipe(Out))
                           ]),
            read_lines(Out, Lines),
            close(Out)).

read_lines(Out, Lines) :-
        read_line_to_codes(Out, Line1),
        read_lines(Line1, Out, Lines).

read_lines(end_of_file, _, []) :- !.
read_lines(Codes, Out, [Line|Lines]) :-
        atom_codes(Line, Codes),
        read_line_to_codes(Out, Line2),
        read_lines(Line2, Out, Lines).

Errors: - process_error(Exe, Status) where Status is one of exit(Code) or killed(Signal). Raised if the process is waited for (i.e., Options does not include process(-PID)), and does not exit with status 0.
bug: - On Windows, environment(List) is handled as env(List), i.e., the environment is not inherited.

process_which(+Exe, -Path) is semidet

True when Path is an absolute file name for the specification Exe. This deals with the search path as well as extensions used by the OS.

process_id(-PID) is det

True if PID is the process id of the running Prolog process.

deprecated: - Use current_prolog_flag(pid, PID)

process_id(+Process, -PID) is det

PID is the process id of Process. Given that they are united in SWI-Prolog, this is a simple unify.

is_process(+PID) is semidet

True if PID might be a process. Succeeds for any positive integer.

process_release(+PID)

Release process handle. In this implementation this is the same as process_wait(PID, _).

process_wait(+PID, -Status) is det

process_wait(+PID, -Status, +Options) is det

True if PID completed with Status. This call normally blocks until the process is finished. Options:

timeout(+Timeout): Default: infinite. If this option is a number, the waits for a maximum of Timeout seconds and unifies Status with timeout if the process does not terminate within Timeout. In this case PID is not invalidated. On Unix systems only timeout 0 and infinite are supported. A 0-value can be used to poll the status of the process.
release(+Bool): Do/do not release the process. We do not support this flag and a domain_error is raised if release(false) is provided.

Arguments:

Status - is one of exit(Code) or killed(Signal), where Code and Signal are integers. If the timeout option is used Status is unified with timeout after the wait timed out.

process_kill(+PID) is det

process_kill(+PID, +Signal) is det

Send signal to process PID. Default is term. Signal is an integer, Unix signal name (e.g. SIGSTOP) or the more Prolog friendly variation one gets after removing SIG and downcase the result: stop. On Windows systems, Signal is ignored and the process is terminated using the TerminateProcess() API. On Windows systems PID must be obtained from process_create/3, while any PID is allowed on Unix systems.

Compatibility: - SICStus does not accept the prolog friendly version. We choose to do so for compatibility with on_signal/3.

process_group_kill(+PID) is det

process_group_kill(+PID, +Signal) is det

Send signal to the group containing process PID. Default is term. See process_wait/1 for a description of signal handling. In Windows, the same restriction on PID applies: it must have been created from process_create/3, and the the group is terminated via the TerminateJobObject API.

process_set_method(+Method) is det

Determine how the process is created on Unix systems. Method is one of spawn (default), fork or vfork. If the method is spawn but this cannot be used because it is either not supported by the OS or the cwd(Dir) option is given fork is used.

The problem is to be understood as follows. The official portable and safe method to create a process is using the fork() system call. This call however copies the process page tables and get seriously slow as the (Prolog) process is multiple giga bytes large. Alternatively, we may use vfork() which avoids copying the process space. But, the safe usage as guaranteed by the POSIX standard of vfork() is insufficient for our purposes. On practical systems your mileage may vary. Modern posix systems also provide posix_spawn(), which provides a safe and portable alternative for the fork() and exec() sequence that may be implemented using fork() or may use a fast but safe alternative. Unfortunately posix_spawn() doesn't support the option to specify the working directory for the child and we cannot use working_directory/2 as the working directory is shared between threads.

Summarizing, the default is safe and tries to be as fast as possible. On some scenarios and on some OSes it is possible to do better. It is generally a good idea to avoid using the cwd(Dir) option of process_create/3 as without we can use posix_spawn().

Re-exported predicates

The following predicates are exported from this file while their implementation is defined in imported modules or non-module files loaded by this module.