Up

module Netsys_win32

: sig

Primitives for Win32

Event objects

#
type w32_event
#
val create_event : unit -> w32_event

Create an event object

#
val set_event : w32_event -> unit

Set the object to "signaled" state

#
val reset_event : w32_event -> unit

Set the object to "non-signaled" state

#
val test_event : w32_event -> bool

Test whether the object is in signaled state

#
val event_wait : w32_event -> float -> bool

Wait until the event is set to signaled state. The float argument is the timeout in seconds. The function returns whether the object is in signaled state.

#
val event_descr : w32_event -> Unix.file_descr

Returns the proxy descriptor for the event. See lookup below for more on proxy descriptors. This function always returns the same descriptor. The user has to close this descriptor if this function is called.

Primitives for sockets

#
val wsa_event_select : w32_event -> Unix.file_descr -> Netsys_posix.poll_req_events -> unit

associate event objects with socket conditions

#
val wsa_maximum_wait_events : unit -> int

max size of the array passed to wsa_wait_for_multiple_events

#
val wsa_wait_for_multiple_events : w32_event array -> int -> int option

Waits until one of the events in the array is in signaled state, or until a timeout happens. The int is the timeout in milliseconds. A negative timeout means infinity.

The function returns the first index in the array that is signaled.

On timeout, None is returned.

The return value WSA_WAIT_IO_COMPLETION is mapped to the Unix error EINTR.

#
val wsa_enum_network_events : Unix.file_descr -> w32_event -> Netsys_posix.poll_act_events

Checks whether an event has been recorded

#
val real_select : Unix.file_descr list -> Unix.file_descr list -> Unix.file_descr list -> float -> Unix.file_descr list * Unix.file_descr list * Unix.file_descr list

Up to Ocaml 3.10, this function is identical to Unix.select. In 3.11, the latter was changed to a smart implementation that promises to handle other types of handles in addition to sockets. As we do the same in Netsys, this would be a duplication of work. Also, the older implementation is more mature.

Support for named pipes

Win32 named pipes work very much like Unix Domain sockets, only that the Win32 interface is different. This wrapper, however, mimicks socket behaviour as far as possible (and we also use an socket-like API with listen and accept). There is a w32_pipe_server representing pipe servers. An individual pipe is wrapped into a w32_pipe.

Win32 named pipes do not allow to check whether an operation would block before starting the operation. There is so-called overlapped I/O, but it works differently than Unix-style multiplexing.

The following functions add a layer to the Win32 primitives that helps using pipes in a way similar to multiplexing. We allocate buffers for input and output, and the functions pipe_read and pipe_write access these buffers in the first place. When reading, but the read buffer is empty, we start an overlapped read operation from the pipe handle. The arriving data refills the read buffer, and a w32_event is signaled to wake up any pending event loop. During the pending read from the pipe handle, the read buffer is locked, and pipe_read will return EWOULDBLOCK.

Writing is slightly more difficult. The first pipe_write puts the data into the write buffer, and immediately starts an overlapped I/O operation to write the data to the pipe handle. During this operation the write buffer is locked, and cannot be further used to accumulate data, even if there is space. So pipe_write will return EWOULDBLOCK while the operation takes place. A w32_event is signaled when the write operation is over.

The only downside of this approach is that the caller has to use pipe_read and pipe_write to access pipes, instead of Unix.read and Unix.write. If generic r/w functions are required that work for numerous kinds of descriptors, there are Netsys.gread and Netsys.gwrite which support named pipes.

#
type w32_pipe_server

A pipe server. Note that there is no such thing in the Win32 API. Actually, a w32_pipe_server contains the server endpoints of a number of pipes, and a few helper objects.

#
type w32_pipe

A pipe endpoint

#
type pipe_mode =
# | Pipe_in
# | Pipe_out
# | Pipe_duplex
#
val rev_mode : pipe_mode -> pipe_mode

Reverses the direction

#
val create_local_pipe_server : string -> pipe_mode -> int -> w32_pipe_server

create_local_named_pipe name mode n: Create a pipe server. The name must have the format "\\.\pipe\<name>". In n the maximum number of instances is passed. The server is set up with a security descriptor so only clients on the same system can connect.

In the following, a terminology has been chosen that is similar to those of the socket API. The terms are different from those Microsoft prefers, however.

#
val pipe_listen : w32_pipe_server -> int -> unit

Creates the backlog queue with n prepared server endpoints.

One can check for new client connections by looking at the pipe_connect_event.

#
val pipe_accept : w32_pipe_server -> w32_pipe

Waits until the connect event is signaled (usually meaning that a new client connection is available), and returns the new pipe.

#
val pipe_connect : string -> pipe_mode -> w32_pipe

pipe_connect name mode: Creates a client pipe handle, and tries to connect to the pipe server name. The function fails with the Unix error EAGAIN if there are currently no listening instances of the pipe at the server.

The name must be of the form "\\.\pipe\<name>" (excluding connects to pipes on remote systems). This function allows only connects to local pipe servers, and enforces anonymous impersonation.

Note that you also can connect to named pipes using open_in and Unix.openfile, and that these functions do not protect against malicious servers that impersonate as the caller.

#
val pipe_pair : pipe_mode -> w32_pipe * w32_pipe

Returns a pair of connected pipes (using automatically generated names). The left pipe is in the passed pipe_mode, and the right pipe is in the matching complementaty mode.

#
val pipe_read : w32_pipe -> string -> int -> int -> int

pipe_read p s pos len: Tries to read data from the pipe. If data is available, it is put into the len bytes at position pos of the string s, and the actual number of read bytes is returned.

If no data is available, the function fails with a Unix error of EAGAIN.

If the end of the pipe is reached, the function returns 0.

#
val pipe_write : w32_pipe -> string -> int -> int -> int

pipe_write p s pos len: Tries to write data to the pipe. If space is available, the data is taken from the len bytes at position pos of the string s, and the actual number of written bytes is returned.

If no space is available, the function fails with a Unix error of EAGAIN.

#
val pipe_shutdown : w32_pipe -> unit

Cancels all pending I/O operations and closes the pipe handle.

Note that there is no way to close only one direction of bidirectional pipes.

See the comments on closing pipes below.

#
val pipe_shutdown_server : w32_pipe_server -> unit

Closes the pipe server: All endpoints in the backlog queue are shutdown. Note that this can result in crashed connections - if the kernel establishes a connection but it is not yet pipe_accepted, it is simply destroyed by this function.

#
val pipe_connect_event : w32_pipe_server -> w32_event

The event object signals when a new client connection is available

#
val pipe_rd_event : w32_pipe -> w32_event
#
val pipe_wr_event : w32_pipe -> w32_event

The event objects signaling that read and write operations are possible. The read event is in signaled state when the read buffer is non-empty (even for write-only pipes). The write event is in signaled state when the pipe is connected and the write buffer is empty (even for read-only pipes).

#
val pipe_wait_connect : w32_pipe_server -> float -> bool

Wait until a client connects to this server. The float argument is the timeout in seconds. The function returns whether there is data to read or write. If not, a timeout has occurred.

#
val pipe_wait_rd : w32_pipe -> float -> bool
#
val pipe_wait_wr : w32_pipe -> float -> bool

Wait until the pipe becomes readable or writable. The float argument is the timeout in seconds. The function returns whether there is data to read or write. If not, a timeout has occurred.

#
val pipe_server_descr : w32_pipe_server -> Unix.file_descr

Returns the proxy descriptor for the pipe server. See lookup below for more on proxy descriptors. This function always returns the same descriptor. The user has to close this descriptor if this function is called.

#
val pipe_descr : w32_pipe -> Unix.file_descr

Returns the proxy descriptor for the pipe. See lookup below for more on proxy descriptors. This function always returns the same descriptor. The user has to close this descriptor if this function is called.

#
val pipe_name : w32_pipe -> string
#
val pipe_server_name : w32_pipe_server -> string

Returns the name of the pipe

#
val pipe_mode : w32_pipe -> pipe_mode
#
val pipe_server_mode : w32_pipe_server -> pipe_mode

Returns the pipe/server mode

#
val unpredictable_pipe_name : unit -> string

Returns a valid pipe name that can practically not be predicted

Shutting down pipes. The suggested model is that the client shuts down the pipe first. A pipe client ensures that all data are transmitted by waiting until the pipe becomes writable again, and then calling pipe_shutdown. The server then sees EOF when reading from the pipe, or gets an EPIPE error when writing to the pipe. The server should then also pipe_shutdown the endpoint.

When servers start the closure of connections, there is no clean way of ensuring that all written data are transmitted. There is the FlushFileBuffers Win32 function, but it is blocking.

I/O threads

I/O threads can be used to do read/write-based I/O in an asynchronous way for file handles that do not support asynchronous I/O by themselves, e.g. anonymous pipes.

I/O threads are only available if the application is compiled as multi-threaded program.

#
type w32_input_thread
#
val create_input_thread : Unix.file_descr -> w32_input_thread

Creates the input thread for this file descriptor. Data is being pumped from this handle to an internal buffer, and can be read from there by input_thread_read.

The thread continues to run until EOF is reached, an I/O error occurs, or until the thread is cancelled (cancel_input_thread).

After starting the input thread, the file descriptor must not be used anymore. It is now owned by the input thread.

#
val input_thread_event : w32_input_thread -> w32_event

This event is signaled when there is data to read, or the EOF is reached, or there is an error condition

#
val input_thread_read : w32_input_thread -> string -> int -> int -> int

input_thread_read t s pos len: Tries to read data from the buffer. If data is available, it is put into the len bytes at position pos of the string s, and the actual number of read bytes is returned.

If no data is available, the function fails with a Unix error of EAGAIN (non-blocking).

If the end of the data is reached, the function returns 0.

For cancelled requests, the function raises EPERM.

#
val cancel_input_thread : w32_input_thread -> unit

Stops the input thread. No more data will be pumped from the handle to the internal buffer. It is no error to cancel a thread that is already cancelled. There is no way to restart the thread later.

The thread is automatically cancelled by the GC finaliser. However, users are encouraged to call cancel_input_thread as soon as the thread is no longer needed, because a thread is an expensive resource.

Implementation note: Actually, cancellation is only fully implemented on Windows Vista. On XP the actual cancellation may be delayed indefinetely.

#
val input_thread_proxy_descr : w32_input_thread -> Unix.file_descr

Returns the proxy descriptor

#
type w32_output_thread
#
val create_output_thread : Unix.file_descr -> w32_output_thread

Creates the output thread for this file descriptor. Data is being pumped an internal buffer to this descriptor, and can be written there by output_thread_read.

The thread continues to run until it is explicitly closed, or an I/O error occurs, or until the thread is cancelled (cancel_output_thread).

After starting the output thread, the file descriptor must not be used anymore. It is now owned by the output thread.

#
val output_thread_event : w32_output_thread -> w32_event

This event is signaled when there is space in the buffer, or when there is an error condition

#
val output_thread_write : w32_output_thread -> string -> int -> int -> int

output_thread_write t s pos len: Tries to write data to the buffer. If this is possible, the substring starting at position pos of the string s with a length of len is appended to the buffer. The actual number of written bytes is returned.

If no space is available in the buffer, the function fails with a Unix error of EAGAIN (non-blocking).

For cancelled requests, the function raises EPERM.

#
val close_output_thread : w32_output_thread -> unit

Adds the EOF condition to the buffer. When the buffer is written to the descriptor, the descriptor will be closed.

Note that this is also an asynchronous operation, like output_thread_write. If closing is not possible at a certain moment, the Unix error EGAIN is raised. This ensures that all errors of previous writes can be reported.

The output thread terminates after a successful close.

For cancelled requests, the function raises EPERM.

#
val cancel_output_thread : w32_output_thread -> unit

Stops the output thread. This is different from closing as the data that is still in the buffer but not yet written may be dropped (if possible). Also, there is no error reporting.

It is no error to cancel a thread that is already cancelled or closed. There is no way to restart the thread later.

The thread is automatically cancelled by the GC finaliser. However, users are encouraged to call cancel_output_thread or close_output_thread as soon as the thread is no longer needed, because a thread is an expensive resource.

Implementation note: Actually, cancellation is only fully implemented on Windows Vista. On XP the actual cancellation may be delayed indefinetely.

#
val output_thread_proxy_descr : w32_output_thread -> Unix.file_descr

Returns the proxy descriptor

Processes

Keep in mind that Win32 distinguishes between two kinds of executables: console applications, and GUI applications. The kind is set at link time, and stored in the executable file. Years ago, these kinds meant different worlds, and a GUI application could not act like a console application, and vice versa. Nowaways, however, the distinction is mostly gone, and the application kind only affects defaults at program startup:

  • Console: A GUI application starts without console. However, it is possible to allocate a console later. A console application always starts with a console which is created by the OS if missing.
  • Standard handles: For a GUI application, stdin/stdout/stderr are initially set to the invalid file handle. Nevertheless, this feature of standard handles exists, and one can set these handles later. Also, the handles can be inherited by the parent process. For console applications, the standard handles are normally set to the console, and applications can redirect them.
  • Main program: Of course, there is also the difference which C function is called at program startup - hey, but this is O'Caml!
  • Waiting for completion: It is uncommon to wait for the completion of GUI applications. The command interpreter seems to implement a magic so that it is not waited until the program is finished when a GUI application is started. For console applications this is of course done. (Note that this feature is the main reason why programmers still have to link applications as console applications, and cannot simply get the same effect from a application that is linked as GUI and then opens a console.)
#
type create_process_option =
# | CP_change_directory of string
(*The initial working directory is set to this path. By default the new process starts with the current working directory of the caller.*)
# | CP_set_env of string
(*The process environment is set to this encoded array of environment variables. By default the current environment is passed down to the new process.
The string is created from an array of "name=value" settings by separating all elements by null bytes, and by putting two null bytes at the end.
*)
# | CP_std_handles of Unix.file_descr * Unix.file_descr * Unix.file_descr
(*Sets the standard handles of the new console process.*)
# | CP_create_console
(*Creates a new console window. The standard handles of the new process may also be modified - however, the exact effect is not well documented by Microsoft. I have the impression that the logic is this: handles pointing to the parent console are replaced by handles pointing to the new console. Also, invalid handles are replaced by handles of the new console. It does not matter how the standard handles are passed down - either implicitly or by CP_std_handles. So you cannot create a new console, and keep standard handles that are connected to the old console. Best practice is to avoid the combination of CP_std_handles and CP_create_console when there is already a console.
This flag does not have any effect when the started app is a GUI app.
*)
# | CP_detach_from_console
(*The new process detaches from the console at startup, even if it is a console application. Unless CP_std_handles is specified, the new process will initially not have standard handles (i.e. the standard handles are invalid handles)! GUI apps detach from the console anyway.*)
# | CP_inherit_console
(*The new console process inherits the console from the caller, if present. Otherwise the new console process starts without console. For GUI apps there is not any effect: They do not have a console anyway.*)
# | CP_inherit_or_create_console
(*If present, the console is inherited from the caller. If not present, a new console is created for console applications. This mode is the default.*)
# | CP_unicode_environment
(*Indicates that the environment is a Unicode environment*)
# | CP_ansi_environment
(*Indicates that the environment is an ANSI environment. This is the default.*)
# | CP_new_process_group
(*The new process is run in a new process group*)
# | CP_inherit_process_group
(*The new process is run in the same process group as the caller. This is the default*)
#
val cp_set_env : string array -> create_process_option

Returns the CP_set_env option for this array of environment variables (in the Unix.environment format)

#
val search_path : string option -> string -> string option -> string

search_path path_opt name ext_opt: Uses the SearchPath function to locate a file. If name does not end with ext_opt, this extension is added during the search. If path_opt is None, the default search path is used.

#
type w32_process

A handle to spawned processes

#
val create_process : string -> string -> create_process_option list -> w32_process

create_process cmd cmdline options: Spawns a new process that runs concurrently with the calling process. cmd is the command to execute (it is not searched by path, and the file suffix must be given). cmdline is the full command-line.

If the exit code of the new process does not play any role, it is ok to just ignore the returned process handle (which will be automatically closed by a GC finalizer).

#
val close_process : w32_process -> unit

Closes the handle in the w32_process value, if it is still open

#
val get_process_status : w32_process -> Unix.process_status option

Returns the process result if the process is finished, and None otherwise

#
val as_process_event : w32_process -> w32_event

Casts the process handle to an event handle. The process handle is in signaled state as soon as the spawned process is terminated. The event handle can be used in event_wait (above) and wsa_wait_for_multiple_events to wait for the termination of the process.

#
val emulated_pid : w32_process -> int

Returns the MSVCRT.DLL notion of the process identifier (pid). This kind of pid is used in the Unix library to refer to processes, especially in waitpid. Note that the pid is actually a handle, and it must be closed by calling Unix.waitpid.

Each call of emulated_pid returns a new handle.

#
val win_pid : w32_process -> int

Returns the Windows notion of the process identifier (pid)

#
val process_descr : w32_process -> Unix.file_descr

Returns the proxy descriptor of the process

#
val terminate_process : w32_process -> unit

Terminates the process

Consoles

#
val has_console : unit -> bool

True if there is a console

#
val is_console : Unix.file_descr -> bool

Tests whether the descriptor is the input or the output stream of the console.

#
val get_console_input : unit -> Unix.file_descr

Get the input stream of the console. If there is no console yet, a new one is opened.

The returned descriptor needs to be closed by the caller when done with it.

#
val get_console_output : unit -> Unix.file_descr

Get the output stream of the console. If there is no console yet, a new one is opened

The returned descriptor needs to be closed by the caller when done with it.

We use a simplified model of the console where only the visible part of the buffer is represented. All coordinates are relative to the visible part of the buffer.

#
type w32_console_attr = {
# mutable cursor_x
: int;(*from 0 (leftmost) to width-1 (rightmost)*)
# mutable cursor_y
: int;(*from 0 (topmost) to height-1 (bottommost)*)
# mutable cursor_size
: int;(*from 1 to 100*)
# mutable cursor_visible
: bool;
# mutable text_attr
: int;
}
#
type w32_console_info = {
# mutable width
: int;(*screen width of the console in chars*)
# mutable height
: int;(*screen height in lines*)
}
#
val get_console_attr : unit -> w32_console_attr
#
val set_console_attr : w32_console_attr -> unit

Get/set console attributes.

#
val get_console_info : unit -> w32_console_info

Get r/o console info.

#
val fg_blue : int
#
val fg_green : int
#
val fg_red : int
#
val fg_intensity : int
#
val bg_blue : int
#
val bg_green : int
#
val bg_red : int
#
val bg_intensity : int

Bits of text_attr

#
type w32_console_mode = {
# mutable enable_echo_input
: bool;
# mutable enable_insert_mode
: bool;
# mutable enable_line_input
: bool;
# mutable enable_processed_input
: bool;
# mutable enable_quick_edit_mode
: bool;
# mutable enable_processed_output
: bool;
# mutable enable_wrap_at_eol_output
: bool;
}

See the msdn docs for GetConsoleMode for details

#
val get_console_mode : unit -> w32_console_mode
#
val set_console_mode : w32_console_mode -> unit

Get/set the console mode.

#
val init_console_codepage : unit -> unit

Sets the code page of the console to the ANSI code page of the system. Unfortunately, the console uses the OEM code page by default (e.g. code page 437 instead of 1252). This function changes the code page back to the ANSI version.

Note, however, that the docs say: "If the current font is a raster font, SetConsoleOutputCP does not affect how extended characters are displayed." (grrmmpf) So you should also switch to a different font - otherwise you get input in the ANSI code page, and do output in the OEM code page.

For Windows novices: Historically, there were two types of 8 bit character sets. The older type is an IBM code page, and predates the ISO-8859 series of character sets. This code page was used at MS-DOS times. Microsoft calls this code page the "OEM" code page. Later, when ISO-8859 was created, Microsoft switched to code pages that are similar to this standard, but also do not fully match them. These newer code pages have names like "Windows-1252", and are now called ANSI code pages by Microsoft. The 8-bit versions of the Win32 calls (which are used by the Ocaml runtime)normally use the ANSI code page.

#
val clear_until_end_of_line : unit -> unit

Writes a space character from the current cursor position to the end of the line

#
val clear_until_end_of_screen : unit -> unit

Writes a space character from the current cursor position to the end of the screen

#
val clear_console : unit -> unit

Clears the screen and the buffer, and sets the cursor to (0,0).

Miscelleneous

#
val get_active_code_page : unit -> int

Get the active code page. See http://msdn.microsoft.com/en-us/library/dd317756%28v=VS.85%29.aspx for a list of codes. Also see Netconversion.win32_code_pages.

Proxy Descriptors

For a number of objects (w32_event, w32_pipe, and w32_pipe_server) it is possible to obtain proxy descriptors. These have type Unix.file_descr and they contain a real file handle. The purpose of these descriptors is to be used as proxy objects that can be passed to functions expecting file descriptors as input. However, you cannot do anything with the proxies except looking the corresponding real objects up. Proxy descriptors are used in interfaces that only allow to pass Unix.file_descr values in and out.

Proxy descriptors have to be closed by the caller once they have been handed out to the caller. Closing the proxy descriptor does not make the descriptor unusable (lookups still work), and the referenced object is also unaffected. It is up to the user when Unix.close is best called - it is even allowed to do it immediately after requesting the proxy descriptor, e.g. via pipe_descr. After closing the proxy, however, it is possible that the system generates another file descriptor that looks equal to the closed proxy. It is often best to close at the moment when one is really done with the proxy.

#
type w32_object =
# | W32_event of w32_event
# | W32_pipe of w32_pipe
# | W32_pipe_server of w32_pipe_server
# | W32_process of w32_process
# | W32_input_thread of w32_input_thread
# | W32_output_thread of w32_output_thread
#
val lookup : Unix.file_descr -> w32_object

Returns the real object behind a proxy descriptor, or raises Not_found. Note that the returned object needs not to be physically identical to the original object. It behaves, however, exactly the same way.

#
val lookup_event : Unix.file_descr -> w32_event
#
val lookup_pipe : Unix.file_descr -> w32_pipe
#
val lookup_pipe_server : Unix.file_descr -> w32_pipe_server
#
val lookup_process : Unix.file_descr -> w32_process
#
val lookup_input_thread : Unix.file_descr -> w32_input_thread
#
val lookup_output_thread : Unix.file_descr -> w32_output_thread

Returns the real object. If not found, or if the object is of unexpected type, Failure is raised.

#
val unregister : Unix.file_descr -> unit

Removes this descriptor from the lookup table. This should only be done after it is closed. Calling unregister is optional, and the removal will take place anyway when the descriptor is collected by the GC.

Miscelleneous

#
val test_close_on_exec : Unix.file_descr -> bool

Tests whether the handle is not inheritable

#
val modify_close_on_exec : Unix.file_descr -> bool -> unit

Sets the close-on-exec flag, i.e. whether the handle is not inheritable. Note that Unix.set_close_on_exec and Unix.clear_close_on_exec have a serious problem, and do not always work.

#
val is_crt_fd : Unix.file_descr -> int -> bool

Tests whether the descriptor has a certain CRT counterpart. E.g. use is_crt_fd 0 to check whether fd is Unix.stdin (physically)

#
val fill_random : string -> unit

Fills the string with random bytes. A cryptographically secure RNG is used

Debugging

#
module Debug : sig
#
val enable : bool Pervasives.ref

Enables [root:Netlog]-style debugging of the Ocaml wrapper

#
val debug_c_wrapper : bool -> unit

Sets whether to debug the C wrapper part. The debug messages are simply written to stderr

end
end