Execute external programs asynchronously under Windows
From Vim Tips Wiki
created 2008 · complexity basic · author Don Reba · version 7.0
Note: Most features from this tip were incorporated into the AsyncCommand plugin that lets you define how to launch and display results from your own asynchronous commands.
 Running external commands
Under Windows, Vim provides two commands for running external programs:
!start. The first suspends Vim, runs the program, waits for a keypress, then returns to Vim. The second runs the program without suspending Vim or waiting for a keypress.
The first method has the drawback of not letting you work while the program is running, but with the second you do not see the result of the program's execution. Here is how to get the best of both worlds:
:!start cmd /c "my app.exe" & pause
The following is an alternative that will leave the command window open until you close it manually:
:!start cmd /k "my app.exe"
Using this knowledge, the following mapping allows you to press F5 to run the compiled program corresponding to the current source file:
:nnoremap <silent> <F5> :!start cmd /c "%:p:r:s,$,.exe," & pause<CR>
Under Windows, if the command begins with "
!start ", Vim creates a process to run the program without invoking a shell, so the "Hit any key to close this window..." message does not appear. :help :!start :help win32-vimrun
This tip assumes that
cmd.exe is your shell (Windows command prompt :help 'shell'), and that you are running a console program (not GUI). Most console programs (and batch scripts) will not pause at the end of the output, so just calling the command by itself with
:!start is not helpful because you will not see the output. Vim's
:!start command starts a single process and additional shell commands like
pause cannot be fed to that process. The mapping works around these issues by invoking the shell directly and telling it to run a command like
"my app.exe" & pause. The cmd shell will execute
"my app.exe" (the quotes are required if the path contains a space), and will then execute
&& is used instead of
pause will only be executed if
"my app.exe" returns errorlevel 0).
%:p:r:s,$,.exe, is replaced with the current file name (
%) modified to include the full path (
:p), without the extension (
:r), and with the end of the string (
$) replaced with
.exe. The comma (
,) is a delimiter for the substitute (
:s). :help filename-modifiers
The alternate method works in the same way, but uses the
/k cmd.exe switch, which executes a command and leaves the window open, as opposed to the
/c switch, which runs a command and exits.
GUI applications do not need to use this method. For example, if you enter a command like
:!calc (to run calc.exe for the Windows calculator), Vim will open a shell window, then launch a new
calc window from that shell. When
calc is closed, the prompt
Hit any key to close this window... appears, and Vim is frozen until a key is pressed. But Calculator is a GUI program so it runs in a new process, in a new window. So, if you want to asynchronously run a GUI program like Calculator, you can simply enter:
 Some notes on cmd.exe quoting
Microsoft's cmd.exe shell behaves in very unexpected ways regarding quoting. For example, the following will not work:
cmd /c "my app.exe" "quoted arg to my app"
cmd.exe will automatically strip the first and last quote character from the command if the first character is a quote, so what actually gets executed is:
my app.exe" "quoted arg to my app
Obviously, this doesn't work.
One solution is therefore to add an extra pair of quotes, which looks weird, but works in most cases:
cmd /c ""my app.exe" "quoted arg to my app""
cmd.exe will automatically remove the extra quotes in this case.
But another problem presents itself if you need to use a special character such as & in an argument. Normally you would do this by quoting:
cmd /c my_app.exe "arg&moreArg"
But if you need all of quoted args, quoted application, and special characters which should not be interpreted as special characters, you will need to escape the special characters for the sake of the cmd.exe shell:
cmd /c ""my app.exe" "arg^&moreArg""
Or, you can surround your command in parentheses and escape ALL special characters, including quotes:
cmd /c (^"my app.exe^" ^"arg^&moreArg^")
See the discussions on vim_dev about proper settings of :help 'shellxquote' for details:
- discussion of 7.3.443, surrounding external commands in quotes
- discussion of 7.3.445, surrounding external commands in parentheses
- discussion of 7.3.446, escaping special characters with ^ when surrounding with parentheses
 See also
 !start vs. Windows start command
The following mapping launches the currently edited file with the external application that is associated with it, using the file type's default handler. This is like entering the filename in the Start > Run... dialog, or double-clicking in Windows Explorer. For example,
example.html will be opened in your default browser.
:nmap <Leader>x :silent ! start "1" "%:p"<CR>
In the Vim command, ':silent' avoids the Hit-Enter prompt, ':p' makes this launch independent from the CWD, the surrounding "" make it handle spaces. The Windows
start command opens the passed file in a separate window, so that Vim doesn't block; the Windows shell locates the application that is associated with the type of file. The "1" is the optional "title" argument to 'start'; without this dummy, 'start' will take the passed file as the title, because it is enclosed in double quotes.
 Getting results back into Vim
In many cases, you may want to run a process in a shell, but then do something with the results in Vim. Normally you could just do this with the system() function or by reading in the output with :r!. But, if you want to do this while continuing to use Vim, you can use the methods from above to launch the process asynchonously, and then use Vim's clientserver features to send the results back into Vim. For example, here is a simple "hello world" sort of command, that will just echo what is passed into it, but do so by first sending it through an asynchronous shell.
command! -nargs=+ AsyncHello call SendAsyncText(<q-args>) function! SendAsyncText(hello_text) let temp_file = tempname() exec 'silent !start cmd /c "echo '.a:hello_text.' > '.temp_file \ ' & vim --servername '.v:servername.' --remote-expr "GetAsyncText('."'".temp_file."')\"". \ ' & pause"' endfunction function! GetAsyncText(temp_file_name) echomsg readfile(a:temp_file_name) call delete(a:temp_file_name) endfunction
This command and associated functions work by invoking an asynchronous external process that writes to a temporary file that is then passed back to Vim. If you passed the text, "hello Vim" into AsyncHello, and the current Vim server is named GVIM3, the command that is actually executed in the shell window is:
echo hello Vim > some_temporary_file_name & vim --servername GVIM3 --remote-expr "GetAsyncText('some_temporary_file_name')" & pause
Note how our use of use of
--servername v:servername ensures that the results will be sent back to the correct Vim. We use
--remote-expr to make it easy to write and invoke a function to process the results of the command in any way we desire. In our case it is a simple
readfile call, but it could be pretty much anything, including use of a command like
cgetfile to run an asynchronous search, for example. The basic idea is the same, however: call a shell command that writes to a known file name before invoking a Vim function on the results.
You may have realized that the
pause is unnecessary in this case, but it serves to demonstrate the fact that the newly spawned window is opened in the foreground. If your goal is to continue working in Vim, sending the new console window to the background would be useful. In most versions of Windows, this can be done using
:!start /min instead of just
:!start in the command.
One final thing to note is the final call to the delete() function in our processing function. This is important, otherwise every time we invoke our command, we will leave behind the temporary file.
 Extended example: asynchronous grep
Here is an extended example of getting the asynchronous process to communicate back to Vim. The following will perform a grep (well, a findstr) asynchronously on Windows. This example uses all the same ideas in our "hello world" example, to do something that is actually useful. Something extra to note in this function is the use of escape() to make sure the user can pass in a search pattern containing a '!' character. According to :help :!, any such characters in a shell command will be expanded to the previously used shell command unless escaped with a backslash. Note that if your search does not need to be asynchronous, it is easy to grep from within Vim without any tricks like this.
if has('win32') " basic async search command! -bar -complete=file -nargs=+ GrepAsync call AsyncGrep('',<f-args>) " same, but recursive command! -bar -complete=file -nargs=+ RGrepAsync call AsyncGrep('/S',<f-args>) function! AsyncGrep(flags, pattern, ...) " find a place to store the results to allow Vim to read them in let grep_temp_file = tempname() " get the file list into a format findstr can read, or search all files if " no file list was passed in if a:0 > 0 let grep_file_list = substitute(join(a:000, '" "'),'/','\','g') else let grep_file_list = '*' endif " Need to escape '!' characters according to :help :! let flags = escape(a:flags, '!') let pattern = escape(a:pattern, '!') let grep_file_list = escape(grep_file_list, '!') call writefile(['Grep results for "'.pattern.'" in "'.grep_file_list.'"'], grep_temp_file) " execute the search in a new process and redirect to a temporary file, then " send the result file back to this Vim instance exec 'silent !start /min cmd /c "echo Searching for "'.pattern.'" in "'.grep_file_list.'", do not close this window... & '. \ 'findstr /R /N '.flags.' "'.pattern.'" "'.grep_file_list.'" >> "'.grep_temp_file.'" & '. \ 'vim --servername '.v:servername.' --remote-expr "ParseAsyncGrep('."'".grep_temp_file."')".'""' endfunction function! ParseAsyncGrep(tempfile) " set up the errorformat so Vim can parse the output let oldefm = &errorformat let &errorformat = &grepformat " parse the results into the quickfix window, but don't jump to the first " search hit exec 'cgetfile '.a:tempfile " restore the previous state let &errorformat = oldefm call delete(a:tempfile) " the echomsg is mostly debug, but could be used instead of the botright " copen to alert the user without interfering echomsg "got grep results file ".a:tempfile " open the quickfix window but don't interfere with the user; jump back to " the current window after opening the quickfix window botright copen wincmd p redraw endfunction endif
 See also
- Avoiding the "Hit ENTER to continue" prompts
- Marc Weber's python-based background-cmd plugin on github to accomplish this on any system
- AsyncCommand is a pure-vimscript implementation based on this tip that works on Windows and Unix/Linux.
- shellasync is another python-based implementation for Unix
- Shougo's vimproc requires an external library, but supports all major operating systems
- Ask on vim_use for some async execution techniques for Unix-based systems because :help :!start indicates that
:!startis for Windows only.
- Generally, you can just add '&' to the end of a command on Unix. It won't pop open a terminal like Windows, so it's even more important to redirect the output to a file. --Pydave 23:01, August 8, 2011 (UTC)
I replaced the first of the following commands with the second:
:nnoremap <silent> <buffer> <F5> :!start cmd /c "%<.exe" && pause<CR> :nnoremap <silent> <F5> :!start cmd /c "%:p:r:s,$,.exe," & pause<CR>
- I removed the
<buffer>since you would need to apply the mapping to every appropriate buffer which is too hard, particularly for someone trying a simple tip.
- The "
%<.exe" does not work as hoped: the "
.exe" does nothing (and "
%<" is deprecated). Also, it does not include the path (editing
- While it's not really necessary to append ".exe", the weird ':s,$,.exe,' expression does it correctly without requiring an
- Using '&&', the cmd shell executes 'pause' only if the first program returns "success" (errorlevel 0). Using '&', 'pause' is always executed (documented as after the first program).
- I actually didn't know the distinction. This is pretty much what every script in this tip was intended to do, so I have made the change there, too and will update my own config. Thanks! --Fritzophrenic 15:57, January 31, 2010 (UTC)
This is just a temp comment to explain my thinking. JohnBeckett 03:43, January 31, 2010 (UTC)