runtime/doc/channel.txt

*channel.txt*      For Vim version 8.1.  Last change: 2019 Sep 11


		  VIM REFERENCE MANUAL	  by Bram Moolenaar


		      Inter-process communication		*channel*

Vim uses channels to communicate with other processes.
A channel uses a socket or pipes.			*socket-interface*
Jobs can be used to start processes and communicate with them.
The Netbeans interface also uses a channel. |netbeans|

1. Overview				|job-channel-overview|
2. Channel demo				|channel-demo|
3. Opening a channel			|channel-open|
4. Using a JSON or JS channel		|channel-use|
5. Channel commands			|channel-commands|
6. Using a RAW or NL channel		|channel-raw|
7. More channel functions		|channel-more|
8. Channel functions details		|channel-functions-details|
9. Starting a job with a channel	|job-start|
10. Starting a job without a channel	|job-start-nochannel|
11. Job functions			|job-functions-details|
12. Job options				|job-options|
13. Controlling a job			|job-control|
14. Using a prompt buffer		|prompt-buffer|

{only when compiled with the |+channel| feature for channel stuff}
	You can check this with: `has('channel')`
{only when compiled with the |+job| feature for job stuff}
	You can check this with: `has('job')`

==============================================================================
1. Overview						*job-channel-overview*

There are four main types of jobs:
1. A daemon, serving several Vim instances.
   Vim connects to it with a socket.
2. One job working with one Vim instance, asynchronously.
   Uses a socket or pipes.
3. A job performing some work for a short time, asynchronously.
   Uses a socket or pipes.
4. Running a filter, synchronously.
   Uses pipes.

For when using sockets See |job-start|, |job-start-nochannel| and
|channel-open|.  For 2 and 3, one or more jobs using pipes, see |job-start|.
For 4 use the ":{range}!cmd" command, see |filter|.

Over the socket and pipes these protocols are available:
RAW	nothing known, Vim cannot tell where a message ends
NL	every message ends in a NL (newline) character
JSON	JSON encoding |json_encode()|
JS	JavaScript style JSON-like encoding |js_encode()|

Common combination are:
- Using a job connected through pipes in NL mode.  E.g., to run a style
  checker and receive errors and warnings.
- Using a daemon, connecting over a socket in JSON mode.  E.g. to lookup
  cross-references in a database.

==============================================================================
2. Channel demo				*channel-demo* *demoserver.py*

This requires Python.  The demo program can be found in
$VIMRUNTIME/tools/demoserver.py
Run it in one terminal.  We will call this T1.

Run Vim in another terminal.  Connect to the demo server with: >
	let channel = ch_open('localhost:8765')

In T1 you should see:
	=== socket opened === ~

You can now send a message to the server: >
	echo ch_evalexpr(channel, 'hello!')

The message is received in T1 and a response is sent back to Vim.
You can see the raw messages in T1.  What Vim sends is:
	[1,"hello!"] ~
And the response is:
	[1,"got it"] ~
The number will increase every time you send a message.

The server can send a command to Vim.  Type this on T1 (literally, including
the quotes):
	["ex","echo 'hi there'"] ~
And you should see the message in Vim. You can move the cursor a word forward:
	["normal","w"] ~

To handle asynchronous communication a callback needs to be used: >
	func MyHandler(channel, msg)
	  echo "from the handler: " . a:msg
	endfunc
	call ch_sendexpr(channel, 'hello!', {'callback': "MyHandler"})
Vim will not wait for a response.  Now the server can send the response later
and MyHandler will be invoked.

Instead of giving a callback with every send call, it can also be specified
when opening the channel: >
	call ch_close(channel)
	let channel = ch_open('localhost:8765', {'callback': "MyHandler"})
	call ch_sendexpr(channel, 'hello!')

When trying out channels it's useful to see what is going on.  You can tell
Vim to write lines in log file: >
	call ch_logfile('channellog', 'w')
See |ch_logfile()|.

==============================================================================
3. Opening a channel					*channel-open*

To open a channel: >
    let channel = ch_open({address} [, {options}])
    if ch_status(channel) == "open"
      " use the channel

Use |ch_status()| to see if the channel could be opened.

{address} has the form "hostname:port".  E.g., "localhost:8765".

{options} is a dictionary with optional entries:	*channel-open-options*

"mode" can be:						*channel-mode*
	"json" - Use JSON, see below; most convenient way. Default.
	"js"   - Use JS (JavaScript) encoding, more efficient than JSON.
	"nl"   - Use messages that end in a NL character
	"raw"  - Use raw messages
						*channel-callback* *E921*
"callback"	A function that is called when a message is received that is
		not handled otherwise.  It gets two arguments: the channel
		and the received message. Example: >
	func Handle(channel, msg)
	  echo 'Received: ' . a:msg
	endfunc
	let channel = ch_open("localhost:8765", {"callback": "Handle"})
<
		When "mode" is "json" or "js" the "msg" argument is the body
		of the received message, converted to Vim types.
		When "mode" is "nl" the "msg" argument is one message,
		excluding the NL.
		When "mode" is "raw" the "msg" argument is the whole message
		as a string.

		For all callbacks: Use |function()| to bind it to arguments
		and/or a Dictionary.  Or use the form "dict.function" to bind
		the Dictionary.

		Callbacks are only called at a "safe" moment, usually when Vim
		is waiting for the user to type a character.  Vim does not use
		multi-threading.

							*close_cb*
"close_cb"	A function that is called when the channel gets closed, other
		than by calling ch_close().  It should be defined like this: >
	func MyCloseHandler(channel)
<		Vim will invoke callbacks that handle data before invoking
		close_cb, thus when this function is called no more data will
		be passed to the callbacks.  However, if a callback causes Vim
		to check for messages, the close_cb may be invoked while still
		in the callback.  The plugin must handle this somehow, it can
		be useful to know that no more data is coming.
							*channel-drop*
"drop"		Specifies when to drop messages:
		    "auto"	When there is no callback to handle a message.
				The "close_cb" is also considered for this.
		    "never"	All messages will be kept.

							*channel-noblock*
"noblock"	Same effect as |job-noblock|.  Only matters for writing.

							*waittime*
"waittime"	The time to wait for the connection to be made in
		milliseconds.  A negative number waits forever.

		The default is zero, don't wait, which is useful if a local
		server is supposed to be running already.  On Unix Vim
		actually uses a 1 msec timeout, that is required on many
		systems.  Use a larger value for a remote server, e.g.  10
		msec at least.
							*channel-timeout*
"timeout"	The time to wait for a request when blocking, E.g. when using
		ch_evalexpr().  In milliseconds.  The default is 2000 (2
		seconds).

When "mode" is "json" or "js" the "callback" is optional.  When omitted it is
only possible to receive a message after sending one.

To change the channel options after opening it use |ch_setoptions()|.  The
arguments are similar to what is passed to |ch_open()|, but "waittime" cannot
be given, since that only applies to opening the channel.

For example, the handler can be added or changed: >
    call ch_setoptions(channel, {'callback': callback})
When "callback" is empty (zero or an empty string) the handler is removed.

After a callback has been invoked Vim will update the screen and put the
cursor back where it belongs.  Thus the callback should not need to do
`:redraw`.

The timeout can be changed: >
    call ch_setoptions(channel, {'timeout': msec})
<
							  *channel-close* *E906*
Once done with the channel, disconnect it like this: >
    call ch_close(channel)
When a socket is used this will close the socket for both directions.  When
pipes are used (stdin/stdout/stderr) they are all closed.  This might not be
what you want!  Stopping the job with job_stop() might be better.
All readahead is discarded, callbacks will no longer be invoked.

Note that a channel is closed in three stages:
  - The I/O ends, log message: "Closing channel". There can still be queued
    messages to read or callbacks to invoke.
  - The readahead is cleared, log message: "Clearing channel".  Some variables
    may still reference the channel.
  - The channel is freed, log message: "Freeing channel".

When the channel can't be opened you will get an error message.  There is a
difference between MS-Windows and Unix: On Unix when the port doesn't exist
ch_open() fails quickly.  On MS-Windows "waittime" applies.
*E898* *E901* *E902*

If there is an error reading or writing a channel it will be closed.
*E630* *E631* 

==============================================================================
4. Using a JSON or JS channel					*channel-use*

If mode is JSON then a message can be sent synchronously like this: >
    let response = ch_evalexpr(channel, {expr})
This awaits a response from the other side.

When mode is JS this works the same, except that the messages use
JavaScript encoding.  See |js_encode()| for the difference.

To send a message, without handling a response or letting the channel callback
handle the response: >
    call ch_sendexpr(channel, {expr})

To send a message and letting the response handled by a specific function,
asynchronously: >
    call ch_sendexpr(channel, {expr}, {'callback': Handler})

Vim will match the response with the request using the message ID.  Once the
response is received the callback will be invoked.  Further responses with the
same ID will be ignored.  If your server sends back multiple responses you
need to send them with ID zero, they will be passed to the channel callback.

The {expr} is converted to JSON and wrapped in an array.  An example of the
message that the receiver will get when {expr} is the string "hello":
	[12,"hello"] ~

The format of the JSON sent is:
    [{number},{expr}]

In which {number} is different every time.  It must be used in the response
(if any):

    [{number},{response}]

This way Vim knows which sent message matches with which received message and
can call the right handler.  Also when the messages arrive out of order.

A newline character is terminating the JSON text.  This can be used to
separate the read text.  For example, in Python:
	splitidx = read_text.find('\n')
	message = read_text[:splitidx]
	rest = read_text[splitidx + 1:]

The sender must always send valid JSON to Vim.  Vim can check for the end of
the message by parsing the JSON.  It will only accept the message if the end
was received.  A newline after the message is optional.

When the process wants to send a message to Vim without first receiving a
message, it must use the number zero:
    [0,{response}]

Then channel handler will then get {response} converted to Vim types.  If the
channel does not have a handler the message is dropped.

It is also possible to use ch_sendraw() and ch_evalraw() on a JSON or JS
channel.  The caller is then completely responsible for correct encoding and
decoding.

==============================================================================
5. Channel commands					*channel-commands*

With a JSON channel the process can send commands to Vim that will be
handled by Vim internally, it does not require a handler for the channel.

Possible commands are:				*E903* *E904* *E905*
    ["redraw", {forced}]
    ["ex",     {Ex command}]
    ["normal", {Normal mode command}]
    ["expr",   {expression}, {number}]
    ["expr",   {expression}]
    ["call",   {func name}, {argument list}, {number}]
    ["call",   {func name}, {argument list}]

With all of these: Be careful what these commands do!  You can easily
interfere with what the user is doing.  To avoid trouble use |mode()| to check
that the editor is in the expected state.  E.g., to send keys that must be
inserted as text, not executed as a command:
    ["ex","if mode() == 'i' | call feedkeys('ClassName') | endif"] ~

Errors in these commands are normally not reported to avoid them messing up
the display.  If you do want to see them, set the 'verbose' option to 3 or
higher.


Command "redraw" ~

The other commands do not explicitly update the screen, so that you can send a
sequence of commands without the cursor moving around.  A redraw can happen as
a side effect of some commands.  You must end with the "redraw" command to
show any changed text and show the cursor where it belongs.

The argument is normally an empty string:
	["redraw", ""] ~
To first clear the screen pass "force":
	["redraw", "force"] ~


Command "ex" ~

The "ex" command is executed as any Ex command.  There is no response for
completion or error.  You could use functions in an |autoload| script:
	["ex","call myscript#MyFunc(arg)"]

You can also use "call |feedkeys()|" to insert any key sequence.

When there is an error a message is written to the channel log, if it exists,
and v:errmsg is set to the error.


Command "normal" ~

The "normal" command is executed like with ":normal!", commands are not
mapped.  Example to open the folds under the cursor:
	["normal" "zO"]


Command "expr"  with response ~

The "expr" command can be used to get the result of an expression.  For
example, to get the number of lines in the current buffer:
	["expr","line('$')", -2] ~

It will send back the result of the expression:
	[-2, "last line"] ~
The format is:
	[{number}, {result}]

Here {number} is the same as what was in the request.  Use a negative number
to avoid confusion with message that Vim sends.  Use a different number on
every request to be able to match the request with the response.

{result} is the result of the evaluation and is JSON encoded.  If the
evaluation fails or the result can't be encoded in JSON it is the string
"ERROR".


Command "expr" without a response ~

This command is similar to "expr" above, but does not send back any response.
Example:
	["expr","setline('$', ['one', 'two', 'three'])"] ~
There is no third argument in the request.


Command "call" ~

This is similar to "expr", but instead of passing the whole expression as a
string this passes the name of a function and a list of arguments.  This
avoids the conversion of the arguments to a string and escaping and
concatenating them.  Example:
	["call", "line", ["$"], -2] ~

Leave out the fourth argument if no response is to be sent:
	["call", "setline", ["$", ["one", "two", "three"]]] ~

==============================================================================
6. Using a RAW or NL channel				*channel-raw*

If mode is RAW or NL then a message can be sent like this: >
    let response = ch_evalraw(channel, {string})

The {string} is sent as-is.  The response will be what can be read from the
channel right away.  Since Vim doesn't know how to recognize the end of the
message you need to take care of it yourself.  The timeout applies for reading
the first byte, after that it will not wait for anything more.

If mode is "nl" you can send a message in a similar way.  You are expected
to put in the NL after each message.  Thus you can also send several messages
ending in a NL at once.  The response will be the text up to and including the
first NL.  This can also be just the NL for an empty response.
If no NL was read before the channel timeout an empty string is returned.

To send a message, without expecting a response: >
    call ch_sendraw(channel, {string})
The process can send back a response, the channel handler will be called with
it.

To send a message and letting the response handled by a specific function,
asynchronously: >
    call ch_sendraw(channel, {string}, {'callback': 'MyHandler'})

This {string} can also be JSON, use |json_encode()| to create it and
|json_decode()| to handle a received JSON message.

It is not possible to use |ch_evalexpr()| or |ch_sendexpr()| on a raw channel.

A String in Vim cannot contain NUL bytes.  To send or receive NUL bytes read
or write from a buffer.  See |in_io-buffer| and |out_io-buffer|.

==============================================================================
7. More channel functions				*channel-more*

To obtain the status of a channel: ch_status(channel).  The possible results
are:
	"fail"		Failed to open the channel.
	"open"		The channel can be used.
	"buffered"	The channel was closed but there is data to read.
	"closed"	The channel was closed.

To obtain the job associated with a channel: ch_getjob(channel)

To read one message from a channel: >
	let output = ch_read(channel)
This uses the channel timeout.  To read without a timeout, just get any
message that is available: >
	let output = ch_read(channel, {'timeout': 0})
When no message was available then the result is v:none for a JSON or JS mode
channels, an empty string for a RAW or NL channel.  You can use |ch_canread()|
to check if there is something to read.

Note that when there is no callback, messages are dropped.  To avoid that add
a close callback to the channel.

To read all output from a RAW channel that is available: >
	let output = ch_readraw(channel)
To read the error output: >
	let output = ch_readraw(channel, {"part": "err"})

ch_read() and ch_readraw() use the channel timeout.  When there is nothing to
read within that time an empty string is returned.  To specify a different
timeout in msec use the "timeout" option:
	{"timeout": 123} ~
To read from the error output use the "part" option:
	{"part": "err"} ~
To read a message with a specific ID, on a JS or JSON channel:
	{"id": 99} ~
When no ID is specified or the ID is -1, the first message is returned. This
overrules any callback waiting for this message.

For a RAW channel this returns whatever is available, since Vim does not know
where a message ends.
For a NL channel this returns one message.
For a JS or JSON channel this returns one decoded message.
This includes any sequence number.

==============================================================================
8. Channel functions details			*channel-functions-details*

ch_canread({handle})						*ch_canread()*
		Return non-zero when there is something to read from {handle}.
		{handle} can be a Channel or a Job that has a Channel.

		This is useful to read from a channel at a convenient time,
		e.g. from a timer.

		Note that messages are dropped when the channel does not have
		a callback.  Add a close callback to avoid that.

		Can also be used as a |method|: >
			GetChannel()->ch_canread()

ch_close({handle})						*ch_close()*
		Close {handle}.  See |channel-close|.
		{handle} can be a Channel or a Job that has a Channel.
		A close callback is not invoked.

		Can also be used as a |method|: >
			GetChannel()->ch_close()

ch_close_in({handle})						*ch_close_in()*
		Close the "in" part of {handle}.  See |channel-close-in|.
		{handle} can be a Channel or a Job that has a Channel.
		A close callback is not invoked.

		Can also be used as a |method|: >
			GetChannel()->ch_close_in()


ch_evalexpr({handle}, {expr} [, {options}])			*ch_evalexpr()*
		Send {expr} over {handle}.  The {expr} is encoded
		according to the type of channel.  The function cannot be used
		with a raw channel.  See |channel-use|.
		{handle} can be a Channel or a Job that has a Channel.
								*E917*
		{options} must be a Dictionary.  It must not have a "callback"
		entry.  It can have a "timeout" entry to specify the timeout
		for this specific request.

		ch_evalexpr() waits for a response and returns the decoded
		expression.  When there is an error or timeout it returns an
		empty string.

		Note that while waiting for the response, Vim handles other
		messages.  You need to make sure this doesn't cause trouble.

		Can also be used as a |method|: >
			GetChannel()->ch_evalexpr(expr)


ch_evalraw({handle}, {string} [, {options}])		*ch_evalraw()*
		Send {string} over {handle}.
		{handle} can be a Channel or a Job that has a Channel.

		Works like |ch_evalexpr()|, but does not encode the request or
		decode the response.  The caller is responsible for the
		correct contents.  Also does not add a newline for a channel
		in NL mode, the caller must do that.  The NL in the response
		is removed.
		Note that Vim does not know when the text received on a raw
		channel is complete, it may only return the first part and you
		need to use |ch_readraw()| to fetch the rest.
		See |channel-use|.

		Can also be used as a |method|: >
			GetChannel()->ch_evalraw(rawstring)

ch_getbufnr({handle}, {what})				 *ch_getbufnr()*
		Get the buffer number that {handle} is using for {what}.
		{handle} can be a Channel or a Job that has a Channel.
		{what} can be "err" for stderr, "out" for stdout or empty for
		socket output.
		Returns -1 when there is no buffer.

		Can also be used as a |method|: >
			GetChannel()->ch_getbufnr(what)

ch_getjob({channel})						*ch_getjob()*
		Get the Job associated with {channel}.
		If there is no job calling |job_status()| on the returned Job
		will result in "fail".

		Can also be used as a |method|: >
			GetChannel()->ch_getjob()


ch_info({handle})						*ch_info()*
		Returns a Dictionary with information about {handle}.  The
		items are:
		   "id"		  number of the channel
		   "status"	  "open", "buffered" or "closed", like
				  ch_status()
		When opened with ch_open():
		   "hostname"	  the hostname of the address
		   "port"	  the port of the address
		   "sock_status"  "open" or "closed"
		   "sock_mode"	  "NL", "RAW", "JSON" or "JS"
		   "sock_io"	  "socket"
		   "sock_timeout" timeout in msec
		When opened with job_start():
		   "out_status"	  "open", "buffered" or "closed"
		   "out_mode"	  "NL", "RAW", "JSON" or "JS"
		   "out_io"	  "null", "pipe", "file" or "buffer"
		   "out_timeout"  timeout in msec
		   "err_status"	  "open", "buffered" or "closed"
		   "err_mode"	  "NL", "RAW", "JSON" or "JS"
		   "err_io"	  "out", "null", "pipe", "file" or "buffer"
		   "err_timeout"  timeout in msec
		   "in_status"	  "open" or "closed"
		   "in_mode"	  "NL", "RAW", "JSON" or "JS"
		   "in_io"	  "null", "pipe", "file" or "buffer"
		   "in_timeout"	  timeout in msec

		Can also be used as a |method|: >
			GetChannel()->ch_info()


ch_log({msg} [, {handle}])					*ch_log()*
		Write {msg} in the channel log file, if it was opened with
		|ch_logfile()|.
		When {handle} is passed the channel number is used for the
		message.
		{handle} can be a Channel or a Job that has a Channel.  The
		Channel must be open for the channel number to be used.

		Can also be used as a |method|: >
			'did something'->ch_log()


ch_logfile({fname} [, {mode}])					*ch_logfile()*
		Start logging channel activity to {fname}.
		When {fname} is an empty string: stop logging.

		When {mode} is omitted or "a" append to the file.
		When {mode} is "w" start with an empty file.

		Use |ch_log()| to write log messages.  The file is flushed
		after every message, on Unix you can use "tail -f" to see what
		is going on in real time.

		This function is not available in the |sandbox|.
		NOTE: the channel communication is stored in the file, be
		aware that this may contain confidential and privacy sensitive
		information, e.g. a password you type in a terminal window.

		Can also be used as a |method|: >
			'logfile'->ch_logfile('w')


ch_open({address} [, {options}])				*ch_open()*
		Open a channel to {address}.  See |channel|.
		Returns a Channel.  Use |ch_status()| to check for failure.

		{address} has the form "hostname:port", e.g.,
		"localhost:8765".

		If {options} is given it must be a |Dictionary|.
		See |channel-open-options|.

		Can also be used as a |method|: >
			GetAddress()->ch_open()


ch_read({handle} [, {options}])					*ch_read()*
		Read from {handle} and return the received message.
		{handle} can be a Channel or a Job that has a Channel.
		For a NL channel this waits for a NL to arrive, except when
		there is nothing more to read (channel was closed).
		See |channel-more|.

		Can also be used as a |method|: >
			GetChannel()->ch_read()


ch_readblob({handle} [, {options}])			*ch_readblob()*
		Like ch_read() but reads binary data and returns a |Blob|.
		See |channel-more|.

		Can also be used as a |method|: >
			GetChannel()->ch_readblob()


ch_readraw({handle} [, {options}])			*ch_readraw()*
		Like ch_read() but for a JS and JSON channel does not decode
		the message.  For a NL channel it does not block waiting for
		the NL to arrive, but otherwise works like ch_read().
		See |channel-more|.

		Can also be used as a |method|: >
			GetChannel()->ch_readraw()


ch_sendexpr({handle}, {expr} [, {options}])			*ch_sendexpr()*
		Send {expr} over {handle}.  The {expr} is encoded
		according to the type of channel.  The function cannot be used
		with a raw channel.
		See |channel-use|.				*E912*
		{handle} can be a Channel or a Job that has a Channel.

		Can also be used as a |method|: >
			GetChannel()->ch_sendexpr(expr)


ch_sendraw({handle}, {expr} [, {options}])		*ch_sendraw()*
		Send |String| or |Blob| {expr} over {handle}.
		Works like |ch_sendexpr()|, but does not encode the request or
		decode the response.  The caller is responsible for the
		correct contents.  Also does not add a newline for a channel
		in NL mode, the caller must do that.  The NL in the response
		is removed.
		See |channel-use|.

		Can also be used as a |method|: >
			GetChannel()->ch_sendraw(rawexpr)


ch_setoptions({handle}, {options})			*ch_setoptions()*
		Set options on {handle}:
			"callback"	the channel callback
			"timeout"	default read timeout in msec
			"mode"		mode for the whole channel
		See |ch_open()| for more explanation.
		{handle} can be a Channel or a Job that has a Channel.

		Note that changing the mode may cause queued messages to be
		lost.

		These options cannot be changed:
			"waittime"	only applies to |ch_open()|

		Can also be used as a |method|: >
			GetChannel()->ch_setoptions(options)


ch_status({handle} [, {options}])				*ch_status()*
		Return the status of {handle}:
			"fail"		failed to open the channel
			"open"		channel can be used
			"buffered"	channel can be read, not written to
			"closed"	channel can not be used
		{handle} can be a Channel or a Job that has a Channel.
		"buffered" is used when the channel was closed but there is
		still data that can be obtained with |ch_read()|.

		If {options} is given it can contain a "part" entry to specify
		the part of the channel to return the status for: "out" or
		"err".  For example, to get the error status: >
			ch_status(job, {"part": "err"})
<
		Can also be used as a |method|: >
			GetChannel()->ch_status()

==============================================================================
9. Starting a job with a channel			*job-start* *job*

To start a job and open a channel for stdin/stdout/stderr: >
    let job = job_start(command, {options})

You can get the channel with: >
    let channel = job_getchannel(job)

The channel will use NL mode.  If you want another mode it's best to specify
this in {options}.  When changing the mode later some text may have already
been received and not parsed correctly.

If the command produces a line of output that you want to deal with, specify
a handler for stdout: >
    let job = job_start(command, {"out_cb": "MyHandler"})
The function will be called with the channel and a message. You would define
it like this: >
    func MyHandler(channel, msg)

Without the handler you need to read the output with |ch_read()| or
|ch_readraw()|. You can do this in the close callback, see |read-in-close-cb|.

Note that if the job exits before you read the output, the output may be lost.
This depends on the system (on Unix this happens because closing the write end
of a pipe causes the read end to get EOF).  To avoid this make the job sleep
for a short while before it exits.

The handler defined for "out_cb" will not receive stderr.  If you want to
handle that separately, add an "err_cb" handler: >
    let job = job_start(command, {"out_cb": "MyHandler",
	    \			  "err_cb": "ErrHandler"})

If you want to handle both stderr and stdout with one handler use the
"callback" option: >
    let job = job_start(command, {"callback": "MyHandler"}) 

Depending on the system, starting a job can put Vim in the background, the
started job gets the focus.  To avoid that, use the `foreground()` function.
This might not always work when called early, put in the callback handler or
use a timer to call it after the job has started.

You can send a message to the command with ch_evalraw().  If the channel is in
JSON or JS mode you can use ch_evalexpr().

There are several options you can use, see |job-options|.
For example, to start a job and write its output in buffer "dummy": >
	let logjob = job_start("tail -f /tmp/log",
			     \ {'out_io': 'buffer', 'out_name': 'dummy'})
	sbuf dummy


Job input from a buffer ~
							*in_io-buffer*
To run a job that reads from a buffer: >
	let job = job_start({command},
	    \ {'in_io': 'buffer', 'in_name': 'mybuffer'})
<
							*E915* *E918*
The buffer is found by name, similar to |bufnr()|. The buffer must exist and
be loaded when job_start() is called.

By default this reads the whole buffer.  This can be changed with the "in_top"
and "in_bot" options.

A special mode is when "in_top" is set to zero and "in_bot" is not set: Every
time a line is added to the buffer, the last-but-one line will be sent to the
job stdin.  This allows for editing the last line and sending it when pressing
Enter.
							*channel-close-in*
When not using the special mode the pipe or socket will be closed after the
last line has been written.  This signals the reading end that the input
finished.  You can also use |ch_close_in()| to close it sooner.

NUL bytes in the text will be passed to the job (internally Vim stores these
as NL bytes).


Reading job output in the close callback ~
							*read-in-close-cb*
If the job can take some time and you don't need intermediate results, you can
add a close callback and read the output there: >

	func! CloseHandler(channel)
	  while ch_status(a:channel, {'part': 'out'}) == 'buffered'
	    echomsg ch_read(a:channel)
	  endwhile
	endfunc
	let job = job_start(command, {'close_cb': 'CloseHandler'})

You will want to do something more useful than "echomsg".

==============================================================================
10. Starting a job without a channel			*job-start-nochannel*

To start another process without creating a channel: >
    let job = job_start(command,
	\ {"in_io": "null", "out_io": "null", "err_io": "null"})

This starts {command} in the background, Vim does not wait for it to finish.

When Vim sees that neither stdin, stdout or stderr are connected, no channel
will be created.  Often you will want to include redirection in the command to
avoid it getting stuck.

There are several options you can use, see |job-options|.

							*job-start-if-needed*
To start a job only when connecting to an address does not work, do something
like this: >
	let channel = ch_open(address, {"waittime": 0})
	if ch_status(channel) == "fail"
	  let job = job_start(command)
	  let channel = ch_open(address, {"waittime": 1000})
	endif

Note that the waittime for ch_open() gives the job one second to make the port
available.

==============================================================================
11. Job functions					*job-functions-details*

job_getchannel({job})					 *job_getchannel()*
		Get the channel handle that {job} is using.
		To check if the job has no channel: >
			if string(job_getchannel()) == 'channel fail'
<
		Can also be used as a |method|: >
			GetJob()->job_getchannel()

job_info([{job}])					*job_info()*
		Returns a Dictionary with information about {job}:
		   "status"	what |job_status()| returns
		   "channel"	what |job_getchannel()| returns
		   "cmd"	List of command arguments used to start the job
		   "process"	process ID
		   "tty_in"	terminal input name, empty when none
		   "tty_out"	terminal output name, empty when none
		   "exitval"	only valid when "status" is "dead"
		   "exit_cb"	function to be called on exit
		   "stoponexit"	|job-stoponexit|

		   Only in Unix:
		   "termsig"	the signal which terminated the process
				(See |job_stop()| for the values)
				only valid when "status" is "dead"

		   Only in MS-Windows:
		   "tty_type"	Type of virtual console in use.
				Values are "winpty" or "conpty".
				See 'termwintype'.

		Without any arguments, returns a List with all Job objects.

		Can also be used as a |method|: >
			GetJob()->job_info()


job_setoptions({job}, {options})			*job_setoptions()*
		Change options for {job}.  Supported are:
		   "stoponexit"	|job-stoponexit|
		   "exit_cb"	|job-exit_cb|

		Can also be used as a |method|: >
			GetJob()->job_setoptions(options)


job_start({command} [, {options}])			*job_start()*
		Start a job and return a Job object.  Unlike |system()| and
		|:!cmd| this does not wait for the job to finish.
		To start a job in a terminal window see |term_start()|.

		If the job fails to start then |job_status()| on the returned
		Job object results in "fail" and none of the callbacks will be
		invoked.

		{command} can be a String.  This works best on MS-Windows.  On
		Unix it is split up in white-separated parts to be passed to
		execvp().  Arguments in double quotes can contain white space.

		{command} can be a List, where the first item is the executable
		and further items are the arguments.  All items are converted
		to String.  This works best on Unix.

		On MS-Windows, job_start() makes a GUI application hidden. If
		want to show it, Use |:!start| instead.

		The command is executed directly, not through a shell, the
		'shell' option is not used.  To use the shell: >
	let job = job_start(["/bin/sh", "-c", "echo hello"])
<		Or: >
	let job = job_start('/bin/sh -c "echo hello"')
<		Note that this will start two processes, the shell and the
		command it executes.  If you don't want this use the "exec"
		shell command.

		On Unix $PATH is used to search for the executable only when
		the command does not contain a slash.

		The job will use the same terminal as Vim.  If it reads from
		stdin the job and Vim will be fighting over input, that
		doesn't work.  Redirect stdin and stdout to avoid problems: >
	let job = job_start(['sh', '-c', "myserver </dev/null >/dev/null"])
<
		The returned Job object can be used to get the status with
		|job_status()| and stop the job with |job_stop()|.

		Note that the job object will be deleted if there are no
		references to it.  This closes the stdin and stderr, which may
		cause the job to fail with an error.  To avoid this keep a
		reference to the job.  Thus instead of: >
	call job_start('my-command')
<		use: >
	let myjob = job_start('my-command')
<		and unlet "myjob" once the job is not needed or is past the
		point where it would fail (e.g. when it prints a message on
		startup).  Keep in mind that variables local to a function
		will cease to exist if the function returns.  Use a
		script-local variable if needed: >
	let s:myjob = job_start('my-command')
<
		{options} must be a Dictionary.  It can contain many optional
		items, see |job-options|.

		Can also be used as a |method|: >
			BuildCommand()->job_start()


job_status({job})					*job_status()* *E916*
		Returns a String with the status of {job}:
			"run"	job is running
			"fail"	job failed to start
			"dead"	job died or was stopped after running

		On Unix a non-existing command results in "dead" instead of
		"fail", because a fork happens before the failure can be
		detected.

		If an exit callback was set with the "exit_cb" option and the
		job is now detected to be "dead" the callback will be invoked.

		For more information see |job_info()|.

		Can also be used as a |method|: >
			GetJob()->job_status()


job_stop({job} [, {how}])					*job_stop()*
		Stop the {job}.  This can also be used to signal the job.

		When {how} is omitted or is "term" the job will be terminated.
		For Unix SIGTERM is sent.  On MS-Windows the job will be
		terminated forcedly (there is no "gentle" way).
		This goes to the process group, thus children may also be
		affected.

		Effect for Unix:
			"term"	 SIGTERM (default)
			"hup"	 SIGHUP
			"quit"	 SIGQUIT
			"int"	 SIGINT
			"kill"	 SIGKILL (strongest way to stop)
			number	 signal with that number

		Effect for MS-Windows:
			"term"	 terminate process forcedly (default)
			"hup"	 CTRL_BREAK
			"quit"	 CTRL_BREAK
			"int"	 CTRL_C
			"kill"	 terminate process forcedly
			Others	 CTRL_BREAK

		On Unix the signal is sent to the process group.  This means
		that when the job is "sh -c command" it affects both the shell
		and the command.

		The result is a Number: 1 if the operation could be executed,
		0 if "how" is not supported on the system.
		Note that even when the operation was executed, whether the
		job was actually stopped needs to be checked with
		|job_status()|.

		If the status of the job is "dead", the signal will not be
		sent.  This is to avoid to stop the wrong job (esp. on Unix,
		where process numbers are recycled).

		When using "kill" Vim will assume the job will die and close
		the channel.

		Can also be used as a |method|: >
			GetJob()->job_stop()


==============================================================================
12. Job options						*job-options*

The {options} argument in job_start() is a dictionary.  All entries are
optional.  Some options can be used after the job has started, using
job_setoptions(job, {options}).  Many options can be used with the channel
related to the job, using ch_setoptions(channel, {options}).
See |job_setoptions()| and |ch_setoptions()|.

						*in_mode* *out_mode* *err_mode*
"in_mode"		mode specifically for stdin, only when using pipes
"out_mode"		mode specifically for stdout, only when using pipes
"err_mode"		mode specifically for stderr, only when using pipes
			See |channel-mode| for the values.

			Note: when setting "mode" the part specific mode is
			overwritten.  Therefore set "mode" first and the part
			specific mode later.

			Note: when writing to a file or buffer and when
			reading from a buffer NL mode is used by default.

						*job-noblock*
"noblock": 1		When writing use a non-blocking write call.  This
			avoids getting stuck if Vim should handle other
			messages in between, e.g. when a job sends back data
			to Vim.  It implies that when `ch_sendraw()` returns
			not all data may have been written yet.
			This option was added in patch 8.1.0350, test with: >
				if has("patch-8.1.350")
				  let options['noblock'] = 1
				endif
<
						*job-callback*
"callback": handler	Callback for something to read on any part of the
			channel.
						*job-out_cb* *out_cb*
"out_cb": handler	Callback for when there is something to read on
			stdout.  Only for when the channel uses pipes.  When
			"out_cb" wasn't set the channel callback is used.
			The two arguments are the channel and the message.

						*job-err_cb* *err_cb*
"err_cb": handler	Callback for when there is something to read on
			stderr.  Only for when the channel uses pipes.  When
			"err_cb" wasn't set the channel callback is used.
			The two arguments are the channel and the message.
						*job-close_cb*
"close_cb": handler	Callback for when the channel is closed.  Same as
			"close_cb" on |ch_open()|, see |close_cb|.
						*job-drop*
"drop": when		Specifies when to drop messages.  Same as "drop" on
			|ch_open()|, see |channel-drop|.  For "auto" the
			exit_cb is not considered.
						*job-exit_cb*
"exit_cb": handler	Callback for when the job ends.  The arguments are the
			job and the exit status.
			Vim checks up to 10 times per second for jobs that
			ended.  The check can also be triggered by calling
			|job_status()|, which may then invoke the exit_cb
			handler.
			Note that data can be buffered, callbacks may still be
			called after the process ends.
							*job-timeout*
"timeout": time		The time to wait for a request when blocking, E.g.
			when using ch_evalexpr().  In milliseconds.  The
			default is 2000 (2 seconds).
						*out_timeout* *err_timeout*
"out_timeout": time	Timeout for stdout.  Only when using pipes.
"err_timeout": time	Timeout for stderr.  Only when using pipes.
			Note: when setting "timeout" the part specific mode is
			overwritten.  Therefore set "timeout" first and the
			part specific mode later.

						*job-stoponexit*
"stoponexit": {signal}	Send {signal} to the job when Vim exits.  See
			|job_stop()| for possible values.
"stoponexit": ""	Do not stop the job when Vim exits.
			The default is "term".

						*job-term*
"term": "open"		Start a terminal in a new window and connect the job
			stdin/stdout/stderr to it.  Similar to using
			`:terminal`.
			NOTE: Not implemented yet!

"channel": {channel}	Use an existing channel instead of creating a new one.
			The parts of the channel that get used for the new job
			will be disconnected from what they were used before.
			If the channel was still used by another job this may
			cause I/O errors.
			Existing callbacks and other settings remain.

"pty": 1		Use a pty (pseudo-tty) instead of a pipe when
			possible.  This is most useful in combination with a
			terminal window, see |terminal|.
			{only on Unix and Unix-like systems}

				*job-in_io* *in_top* *in_bot* *in_name* *in_buf*
"in_io": "null"		disconnect stdin (read from /dev/null)
"in_io": "pipe"		stdin is connected to the channel (default)
"in_io": "file"		stdin reads from a file
"in_io": "buffer"	stdin reads from a buffer
"in_top": number	when using "buffer": first line to send (default: 1)
"in_bot": number	when using "buffer": last line to send (default: last)
"in_name": "/path/file"	the name of the file or buffer to read from
"in_buf": number	the number of the buffer to read from

				*job-out_io* *out_name* *out_buf*
"out_io": "null"	disconnect stdout (goes to /dev/null)
"out_io": "pipe"	stdout is connected to the channel (default)
"out_io": "file"	stdout writes to a file
"out_io": "buffer"	stdout appends to a buffer (see below)
"out_name": "/path/file" the name of the file or buffer to write to
"out_buf": number	the number of the buffer to write to
"out_modifiable": 0	when writing to a buffer, 'modifiable' will be off
			(see below)
"out_msg": 0		when writing to a new buffer, the first line will be
			set to "Reading from channel output..."

				*job-err_io* *err_name* *err_buf*
"err_io": "out"		stderr messages to go to stdout
"err_io": "null"	disconnect stderr  (goes to /dev/null)
"err_io": "pipe"	stderr is connected to the channel (default)
"err_io": "file"	stderr writes to a file
"err_io": "buffer"	stderr appends to a buffer (see below)
"err_name": "/path/file" the name of the file or buffer to write to
"err_buf": number	the number of the buffer to write to
"err_modifiable": 0	when writing to a buffer, 'modifiable' will be off
			(see below)
"err_msg": 0		when writing to a new buffer, the first line will be
			set to "Reading from channel error..."

"block_write": number	only for testing: pretend every other write to stdin
			will block

"env": dict		environment variables for the new process
"cwd": "/path/to/dir"	current working directory for the new process;
			if the directory does not exist an error is given


Writing to a buffer ~
							*out_io-buffer*
When the out_io or err_io mode is "buffer" and there is a callback, the text
is appended to the buffer before invoking the callback.

When a buffer is used both for input and output, the output lines are put
above the last line, since the last line is what is written to the channel
input.  Otherwise lines are appended below the last line.

When using JS or JSON mode with "buffer", only messages with zero or negative
ID will be added to the buffer, after decoding + encoding.  Messages with a
positive number will be handled by a callback, commands are handled as usual.

The name of the buffer from "out_name" or "err_name" is compared the full name
of existing buffers, also after expanding the name for the current directory.
E.g., when a buffer was created with ":edit somename" and the buffer name is
"somename" it will use that buffer.

If there is no matching buffer a new buffer is created.  Use an empty name to
always create a new buffer.  |ch_getbufnr()| can then be used to get the
buffer number.

For a new buffer 'buftype' is set to "nofile" and 'bufhidden' to "hide".  If
you prefer other settings, create the buffer first and pass the buffer number.
					*out_modifiable* *err_modifiable*
The "out_modifiable" and "err_modifiable" options can be used to set the
'modifiable' option off, or write to a buffer that has 'modifiable' off.  That
means that lines will be appended to the buffer, but the user can't easily
change the buffer.
					*out_msg* *err_msg*
The "out_msg" option can be used to specify whether a new buffer will have the
first line set to "Reading from channel output...".  The default is to add the
message.  "err_msg" does the same for channel error.

When an existing buffer is to be written where 'modifiable' is off and the
"out_modifiable" or "err_modifiable" options is not zero, an error is given
and the buffer will not be written to.

When the buffer written to is displayed in a window and the cursor is in the
first column of the last line, the cursor will be moved to the newly added
line and the window is scrolled up to show the cursor if needed.

Undo is synced for every added line.  NUL bytes are accepted (internally Vim
stores these as NL bytes).


Writing to a file ~
							*E920*
The file is created with permissions 600 (read-write for the user, not
accessible for others).  Use |setfperm()| to change this.

If the file already exists it is truncated.

==============================================================================
13. Controlling a job					*job-control*

To get the status of a job: >
	echo job_status(job)

To make a job stop running: >
	job_stop(job)

This is the normal way to end a job. On Unix it sends a SIGTERM to the job.
It is possible to use other ways to stop the job, or even send arbitrary
signals.  E.g. to force a job to stop, "kill it": >
	job_stop(job, "kill")

For more options see |job_stop()|.

==============================================================================
14. Using a prompt buffer				*prompt-buffer*

If you want to type input for the job in a Vim window you have a few options:
- Use a normal buffer and handle all possible commands yourself.
  This will be complicated, since there are so many possible commands.
- Use a terminal window.  This works well if what you type goes directly to
  the job and the job output is directly displayed in the window.
  See |terminal-window|.
- Use a prompt window. This works well when entering a line for the job in Vim
  while displaying (possibly filtered) output from the job.

A prompt buffer is created by setting 'buftype' to "prompt". You would
normally only do that in a newly created buffer.

The user can edit and enter one line of text at the very last line of the
buffer.  When pressing Enter in the prompt line the callback set with
|prompt_setcallback()| is invoked.  It would normally send the line to a job.
Another callback would receive the output from the job and display it in the
buffer, below the prompt (and above the next prompt).

Only the text in the last line, after the prompt, is editable. The rest of the
buffer is not modifiable with Normal mode commands.  It can be modified by
calling functions, such as |append()|.  Using other commands may mess up the
buffer.

After setting 'buftype' to "prompt" Vim does not automatically start Insert
mode, use `:startinsert` if you want to enter Insert mode, so that the user
can start typing a line.

The text of the prompt can be set with the |prompt_setprompt()| function.

The user can go to Normal mode and navigate through the buffer.  This can be
useful see older output or copy text.

The CTRL-W key can be used to start a window command, such as CTRL-W w to
switch to the next window.  This also works in Insert mode (use Shift-CTRL-W
to delete a word). When leaving the window Insert mode will be stopped.  When
coming back to the prompt window Insert mode will be restored.

Any command that starts Insert mode, such as "a", "i", "A" and "I", will move
the cursor to the last line.  "A" will move to the end of the line, "I" to the
start of the line.


 vim:tw=78:ts=8:noet:ft=help:norl: