VIM USER MANUAL - by Bram Moolenaar

		       Write larger plugins

When plugins do more than simple things, they tend to grow big.  This file
explains how to make sure they still load fast and how to split them up in
smaller parts.

|52.1|	Export and import
|52.2|	Autoloading
|52.3|	Autoloading without import/export
|52.4|	Other mechanisms to use
|52.5|	Using a Vim9 script from legacy script

     Next chapter: |usr_90.txt|  Installing Vim
 Previous chapter: |usr_51.txt|  Create a plugin
Table of contents: |usr_toc.txt|

==============================================================================


Vim9 script was designed to make it easier to write large Vim scripts.  It
looks more like other script languages, especially Typescript.  Also,
functions are compiled into instructions that can be executed quickly.  This
makes Vim9 script a lot faster, up to a 100 times.

The basic idea is that a script file has items that are private, only used
inside the script file, and items that are exported, which can be used by
scripts that import them.  That makes very clear what is defined where.

Let's start with an example, a script that exports one function and has one
private function: >

	vim9script

	export def GetMessage(count: string): string
	   var nr = str2nr(count)
	   var result = 


To {nr} we say '
	   result ..= GetReply(nr)
	   return result
	enddef

	def GetReply(nr: number): string
	  if nr == 42
	     return 'yes'
	  elseif nr = 22
	     return 'maybe'
	  else
	     return 'no'
	  endif
	enddef

The `vim9script` command is required, `export` only works in a |Vim9| script.

The `export def GetMessage(...` line starts with `export`, meaning that this
function can be called by other scripts.  The line `def GetReply(...` does not
start with `export`, this is a script-local function, it can only be used
inside this script file.

Now about the script where this is imported.  In this example we use this
layout, which works well for a plugin below the "pack" directory:
	.../plugin/theplugin.vim
	.../lib/getmessage.vim

Assuming the "..." directory has been added to 'runtimepath', Vim will look
for plugins in the "plugin" directory and source "theplugin.vim".  Vim does
not recognize the "lib" directory, you can put any scripts there.

The above script that exports GetMessage() goes in lib/getmessage.vim.  The
GetMessage() function is used in plugin/theplugin.vim: >

	vim9script

	import "../lib/getmessage.vim"
	command -nargs=1 ShowMessage echomsg getmessage.GetMessage(<f-args>)

The `import` command uses a relative path, it starts with "../", which means
to go one directory up.  For other kinds of paths see the `:import` command.

How we can try out the command that the plugin provides: >
	ShowMessage 1
<	To 1 we say no ~
>
	ShowMessage 22
<	To 22 we say maybe ~

Notice that the function GetMessage() is prefixed with the imported script
name "getmessage".  That way, for every imported function used, you know what
script it was imported from.  If you import several scripts each of them could
define a GetMessage() function: >

	vim9script

	import "../lib/getmessage.vim"
	import "../lib/getother.vim"
	command -nargs=1 ShowMessage echomsg getmessage.GetMessage(<f-args>)
	command -nargs=1 ShowOther echomsg getother.GetMessage(<f-args>)

If the imported script name is long or you use it in many places, you can
shorten it by adding an "as" argument: >
	import "../lib/getmessage.vim" as msg
	command -nargs=1 ShowMessage echomsg msg.GetMessage(<f-args>)


RELOADING

One thing to keep in mind: the imported "lib/getmessage.vim" script will be
sourced only once.  When it is imported a second time sourcing it will be
skipped, since the items in it have already been created.  It does not matter
if this import command is in another script, or in the same script that is
sourced again.

This is efficient when using a plugin, but when still developing a plugin it
means that changing "lib/getmessage.vim" after it has been imported will have
no effect.  You need to quit Vim and start it again. (Rationale: the items
defined in the script could be used in a compiled function, sourcing the
script again may break those functions).


USING GLOBALS

Sometimes you will want to use global variables or functions, so that they can
be used anywhere.  A good example is a global variable that passes a
preference to a plugin.  To avoid other scripts using the same name, use a
prefix that is very unlikely to be used elsewhere.  For example, if you have a
"mytags" plugin, you could use: >

	g:mytags_location = '$HOME/project'
	g:mytags_style = 'fast'

==============================================================================


After splitting your large script into pieces, all the lines will still be
loaded and executed the moment the script is used.  Every `import` loads the
imported script to find the items defined there.  Although that is good for
finding errors early, it also takes time.  Which is wasted if the
functionality is not often used.

Instead of having `import` load the script immediately, it can be postponed
until needed.  Using the example above, only one change needs to be made in
the plugin/theplugin.vim script: >
	import autoload "../lib/getmessage.vim"

Nothing in the rest of the script needs to change.  However, the types will
not be checked.  Not even the existence of the GetMessage() function is
checked until it is used.  You will have to decide what is more important for
your script: fast startup or getting errors early.  You can also add the
"autoload" argument later, after you have checked everything works.


AUTOLOAD DIRECTORY

Another form is to use autoload with a script name that is not an absolute or
relative path: >
	import autload "monthlib.vim"

This will search for the script "monthlib.vim" in the autoload directories of
'runtimepath'.  With Unix one of the directories often is "~/.vim/autoload".
It will also search under 'packpath', under "start".

The main advantage of this is that this script can be easily shared with other
scripts.  You do need to make sure that the script name is unique, since Vim
will search all the "autoload" directories in 'runtimepath', and if you are
using several plugins with a plugin manager, it may add a directory to
'runtimepath', each of which might have an "autoload" directory.

Without autoload: >
	import "monthlib.vim"

Vim will search for the script "monthlib.vim" in the import directories of
'runtimepath'.  Note that in this case adding or removing "autoload" changes
where the script is found.  With a relative or absolute path the location does
not change.

==============================================================================


						*write-library-script*
A mechanism from before import/export is still useful and some users may find
it a bit simpler.  The idea is that you call a function with a special name.
That function is then in an autoload script.  We will call that one script a
library script.

The autoload mechanism is based on a function name that has "#" characters: >

	mylib#myfunction(arg)

Vim will recognize the function name by the embedded "#" character and when
it is not defined yet search for the script "autoload/mylib.vim" in
'runtimepath'.  That script must define the "mylib#myfunction()" function.
Obviously the name "mylib" is the part before the "#" and is used as the name
of the script, adding ".vim".

You can put many other functions in the mylib.vim script, you are free to
organize your functions in library scripts.  But you must use function names
where the part before the '#' matches the script name.  Otherwise Vim would
not know what script to load.  This is where it differs from the import/export
mechanism.

If you get really enthusiastic and write lots of library scripts, you may
want to use subdirectories.  Example: >

	netlib#ftp#read('somefile')

Here the script name is taken from the function name up to the last "#". The
"#" in the middle are replaced by a slash, the last one by ".vim".  Thus you
get "netlib/ftp.vim".  For Unix the library script used for this could be:

	~/.vim/autoload/netlib/ftp.vim

Where the function is defined like this: >

	def netlib#ftp#read(fname: string)
		#  Read the file fname through ftp
	enddef

Notice that the name the function is defined with is exactly the same as the
name used for calling the function.  And the part before the last '#'
exactly matches the subdirectory and script name.

You can use the same mechanism for variables: >

	var weekdays = dutch#weekdays

This will load the script "autoload/dutch.vim", which should contain something
like: >

	var dutch#weekdays = ['zondag', 'maandag', 'dinsdag', 'woensdag',
		\ 'donderdag', 'vrijdag', 'zaterdag']

Further reading: |autoload|.

==============================================================================


Some may find the use of several files a hassle and prefer to keep everything
together in one script.  To avoid this resulting in slow startup there is a
mechanism that only defines a small part and postpones the rest to when it is
actually used.  *write-plugin-quickload*

The basic idea is that the plugin is loaded twice.  The first time user
commands and mappings are defined that offer the functionality.  The second
time the functions that implement the functionality are defined.

It may sound surprising that quickload means loading a script twice.  What we
mean is that it loads quickly the first time, postponing the bulk of the
script to the second time, which only happens when you actually use it.  When
you always use the functionality it actually gets slower!

This uses a FuncUndefined autocommand.  This works differently from the
|autoload| functionality explained above.

The following example shows how it's done: >

	" Vim global plugin for demonstrating quick loading
	" Last Change:	2005 Feb 25
	" Maintainer:	Bram Moolenaar <Bram@vim.org>
	" License:	This file is placed in the public domain.

	if !exists("s:did_load")
		command -nargs=* BNRead  call BufNetRead(<f-args>)
		map <F19> :call BufNetWrite('something')<CR>

		let s:did_load = 1
		exe 'au FuncUndefined BufNet* source ' .. expand('<sfile>')
		finish
	endif

	function BufNetRead(...)
		echo 'BufNetRead(' .. string(a:000) .. ')'
		" read functionality here
	endfunction

	function BufNetWrite(...)
		echo 'BufNetWrite(' .. string(a:000) .. ')'
		" write functionality here
	endfunction

When the script is first loaded "s:did_load" is not set.  The commands between
the "if" and "endif" will be executed.  This ends in a |:finish| command, thus
the rest of the script is not executed.

The second time the script is loaded "s:did_load" exists and the commands
after the "endif" are executed.  This defines the (possible long)
BufNetRead() and BufNetWrite() functions.

If you drop this script in your plugin directory Vim will execute it on
startup.  This is the sequence of events that happens:

1. The "BNRead" command is defined and the <F19> key is mapped when the script
   is sourced at startup.  A |FuncUndefined| autocommand is defined.  The
   ":finish" command causes the script to terminate early.

2. The user types the BNRead command or presses the <F19> key.  The
   BufNetRead() or BufNetWrite() function will be called.

3. Vim can't find the function and triggers the |FuncUndefined| autocommand
   event.  Since the pattern "BufNet*" matches the invoked function, the
   command "source fname" will be executed.  "fname" will be equal to the name
   of the script, no matter where it is located, because it comes from
   expanding "<sfile>" (see |expand()|).

4. The script is sourced again, the "s:did_load" variable exists and the
   functions are defined.

Notice that the functions that are loaded afterwards match the pattern in the
|FuncUndefined| autocommand.  You must make sure that no other plugin defines
functions that match this pattern.

==============================================================================


In some cases you have a legacy Vim script where you want to use items from a
Vim9 script.  For example in your .vimrc you want to initialize a plugin.  The
best way to do this is to use `:import`.  For example: >

	import 'myNicePlugin.vim'
	call myNicePlugin.NiceInit('today')

This finds the exported function "NiceInit" in the Vim9 script file and makes
it available as script-local item "myNicePlugin.NiceInit". `:import` always
uses the script namespace, even when "s:" is not given.  If "myNicePlugin.vim"
was already sourced it is not sourced again.

Besides avoiding putting any items in the global namespace (where name clashes
can cause unexpected errors), this also means the script is sourced only once,
no matter how many times items from it are imported.

In some cases, e.g. for testing, you may just want to source the Vim9 script.
That is OK, but then only global items will be available.  The Vim9 script
will have to make sure to use a unique name for these global items. Example: >
	source ~/.vim/extra/myNicePlugin.vim
	call g:NicePluginTest()

==============================================================================

Next chapter: |usr_90.txt|  Installing Vim


Copyright: see |manual-copyright|  vim:tw=78:ts=8:noet:ft=help:norl: