Riml, a relaxed VimL
Riml is a subset of VimL with some added features, and it compiles to plain VimL. Some of the added features include classes, string interpolation, heredocs, default case-sensitive string comparison, default parameters in functions, and other things programmers tend to take for granted. To see how Riml is compiled to VimL, just take a look at this README. The left side is Riml, and the right side is the VimL after compilation.
Variables
count = 1 let s:count = 1
while count < 5 while s:count < 5
source other.vim source other.vim
count += 1 let s:count += 1
end endwhileIf you don't specify a scope modifier (or 'namespace' in Vimspeak), it's script local (s:) by default in the global scope. Within a function, variables without scope modifiers are plain old local variables.
globally
a = 3 let s:a = 3locally (within function or for loop)
def exampleFunc(msg) function! s:exampleFunc(msg)
a = 3 let a = 3
echo msg echo a:msg
end endfunction
for i in expr() for i in s:expr()
echo i echo i
end endforNotice that within a function, it's unnecessary to prefix argument variables with 'a:'. This is, of course, unless we shadow the argument variable by creating our own local variable called 'msg'. In that case, we'd have to refer to the argument variable as 'a:msg' explicitly. Shadowing variables in Riml is considered bad practice, as it's much easier to just come up with unique variable names across a scope.
Assignment as Expression
let a = b = c = 0 let s:c = 0
let s:b = s:c
let s:a = s:bThe 'let' is optional in Riml for all assignments. Without it, the results are the same.
Multiple Assignment Statement
a = 0, b = 1 let s:a = 0
let s:b = 1Checking for existence
unless callcount? if !exists("s:callcount")
callcount = 0 let s:callcount = 0
end endif
callcount += 1 let s:callcount += 1
echo "called #{callcount} times" echo "called " . s:callcount . " times"Notice in the last line of Riml there's string interpolation. This works in double-quoted strings and heredocs, which we'll encounter later.
In Riml, you can choose to end any block with 'end', or with whatever you used to do in VimL ('endif', 'endfunction', etc...). Also, 'if' and 'unless' can now be used as statement modifiers:
callcount = 0 unless callcount?
callcount += 1
echo "called #{callcount} times"Here, the compiled output is the same as the previous example's. Both 'if' and 'unless' can be used this way.
True and False
a = true let s:a = 1
b = false let s:b = 0Operators And Case Sensitivity
if "hi" == greeting if "hi" ==# s:greeting
echo greeting echo s:greeting
end endComparisons compile to case-sensitive by default. To get case-insensitive comparisons, you have to explicitly use the form ending in '?' (ex: '==?'). The only operators that don't add a '#' even though the forms exist are the 'is' and 'isnot' operators. This is because 'is' is much different from its cousin 'is#', and the same is true of 'isnot'.
=== Operator
Oh no, not another of those operators! Well, this one is pretty sweet actually. In VimL, automatic type conversion can be a pain. For example:
echo 4 ==# "4"
=> 1To mitigate this, we can wrap each side in a list, since lists are more strict regarding equality:
echo [4] ==# ["4"]
=> 0The '===' operator wraps both operands in lists:
echo 4 === "4" echo [4] ==# ["4"]
=> 0Heredocs
msg = <<EOS let s:msg = "a vim heredoc! " . s:cryForJoy() . "!\nHooray!"
A vim heredoc! #{cryForJoy()}!
Hooray!
EOSRiml heredocs must have the ending pattern ('EOS' in this case, but can be anything) start at the beginning of the line. Interpolated expressions are allowed in heredocs.
Functions
def fillSearchPat function! s:fillSearchPat()
@/ = getSearchPat() let @/ = s:getSearchPat()
return @/ return @/
end endfunctionFunctions are by default prepended by 's:' unless explicitly prepended with a different scope modifier. Of course, you can use the old form ('function! Name()') for defining functions if you want, as Riml aims to be as compatible as possible with VimL. There are a few exceptions where Riml and VimL aren't compatible, and these differences are explained in the section 'Incompatibilities with VimL'.
When defining a function with no parameters and no keywords (such as 'dict' or 'abort'), the parens after the function name are optional.
Default Parameters
def fillSearchPat(pat = getDefaultSearchPat()) function! s:fillSearchPat(...)
@/ = pat let __splat_var_cpy = a:000
return @/ if !empty(__splat_var_cpy)
end let pat = remove(__splat_var_cpy, 0)
else
let pat = s:getDefaultSearchPat()
endif
@/ = pat
return @/
endfunctionDefault parameters must be the last parameters given to a function, but there can be more than one default parameter. Also, a splat parameter (... or *argName) can come after default parameters. Splats parameters and splat arguments will be explained in the next section.
We can now call the function 'fillSearchPat' without any arguments and it will use the default parameter.
Splat Parameters
A splat parameter is a formal parameter to a function starting with a '*'. It must be the last parameter given to a function.
def my_echo(*words) function! s:my_echo(...)
echo words echo a:000
end endfunction
my_echo("hello", "world") call my_echo("hello", "world")
=> ["hello", "world"]Splat Arguments
Splat arguments are splats used in a function call context. With the function my_echo defined in the
previous section, here's an example at work:
words = ["hello", "world"] let s:words = ["hello", "world"]
echo 'without splat:' echo 'without splat:'
my_echo(words) call my_echo(s:words)
echo 'with splat:' echo 'with splat:'
my_echo(*words) call call('my_echo', s:words)
=> without splat:
=> [["hello", "world"]]
=> with splat:
=> ["hello", "world"]The way to think about how splat arguments work is that the list variable that's
splatted is broke into its constituent parts, just like in Ruby. So, my_echo(*words)
is equivalent to my_echo("hello", "world") in this example. Notice how you can now
pass the exact same arguments your function received to another function by splatting a
parameter and passing it on to another function.
def some_func(*args)
other_func(*args) " passes same arguments along to other function.
endClasses
Basic Class
function! s:SID()
if exists('s:SID_VALUE')
return s:SID_VALUE
endif
let s:SID_VALUE = matchstr(expand('<sfile>'), '<SNR>\zs\d\+\ze_SID$')
return s:SID_VALUE
endfunction
class MyClass function! s:MyClassConstructor(data, otherData, ...)
def initialize(data, otherData, *options) let myClassObj = {}
self.data = data let myClassObj.data = a:data
self.otherData = otherData let myClassObj.otherData = a:otherData
self.options = options let myClassObj.options = a:000
end let myClassObj.getData = function('<SNR>' . s:SID() . '_s:MyClass_getData')
let myClassObj.getOtherData = function('<SNR>' . s:SID() . '_s:MyClass_getOtherData')
defm getData return myClassObj
return self.data endfunction
end
function! <SID>s:MyClass_getdata() dict
defm getOtherData return self.data
return self.otherData endfunction
end
end function! <SID>s:MyClass_getOtherData() dict
return self.otherData
endfunctionIn this basic example of a class, we see a *splat variable. This is just a convenient way to refer to 'a:000' in the body of a function. Splat variables are optional parameters and get compiled to '...'.
Classes can only be defined once, and cannot be reopened. Public member functions are defined with 'defm'. If you want to create a non-public function inside a class, use 'def'. To create an instance of this class, simply:
obj = new MyClass('someData', 'someOtherData')Classes have a default scope modifier of 's:'. That is, they cannot be instantiated outside the script in which they are defined. In order to allow them to be instantiated in any script file, you must declare the class with the explicit scope modifier of 'g:'. For example:
class g:MyClass
...
endIn any one Riml-generated VimL file, there is at most 1 generated s:SID() function.
This generated function will be omitted in the rest of this README for brevity, but
every Riml file that has a class with a member function will have this code in the
compiled output.
Class Inheritance
class Translation function! s:TranslationConstructor(input)
def initialize(input) let translationObj = {}
self.input = input let translationObj.input = a:input
end return translationObj
end endfunction
class FrenchToEnglishTranslation < Translation function! s:FrenchToEnglishTranslationConstructor(input)
defm translate let frenchToEnglishTranslationObj = {}
if (self.input == "Bonjour!") let translationObj = s:TranslationConstructor(a:input)
echo "Hello!" call extend(frenchToEnglishTranslationObj, translationObj)
else let frenchToEnglishTranslationObj.translate = function('<SNR>' . s:SID() . '_s:FrenchToEnglishTranslation_translate')
echo "Sorry, I don't know that word." return frenchToEnglishTranslationObj
end endfunction
end
end function! <SID>s:FrenchToEnglishTranslation_translate() dict
if (self.input ==# "Bonjour!")
translation = new echo "Hello!"
\ FrenchToEnglishTranslation("Bonjour!") else
translation.translate() echo "Sorry, I don't know that word."
endif
endfunction
let s:translation = s:TranslationConstructor("Bonjour!")
call s:translation.translate()
=> "Hello!"Classes that inherit must have their superclass defined before inheritance takes place. In this example, 'Translation' is defined first, which is legal. Since 'Translation' has an initialize function and 'FrenchToEnglishTranslation' doesn't, 'FrenchToEnglishTranslation' instances use the initialize function from 'Translation', and new instances must be provided with an 'input' argument on creation. Basically, if a class doesn't provide an initialize function, it uses its superclass's.
If you look at the last line of Riml in the previous example, you'll see that it doesn't use VimL's builtin 'call' function for calling the 'translate' method on the translation object. Riml can figure out when 'call' is necessary, and will add it to the compiled VimL.
A base class and inheriting class can have different scope modifiers. For example, if you had a script-local base class and wanted to extend it but have the extending class be global, this is not a problem. Simply:
class g:GlobalClass < ScriptLocalClass
...
endUsing 'super'
class Car function! s:CarConstructor(make, model, color)
def initialize(make, model, color) let carObj = {}
self.make = make let carObj.make = a:make
self.model = model let carObj.model = a:model
self.color = color let carObj.color = a:color
end endfunction
end
class HotRod < Car function! s:HotRodConstructor(make, model, color, topSpeed)
def initialize(make, model, color, topSpeed) let hotRodObj = {}
self.topSpeed = topSpeed let hotRodObj.topSpeed = a:topSpeed
super(make, model, color) let carObj = s:CarConstructor(a:make, a:model, a:color)
end call extend(hotRodObj, carObj)
let hotRodObj.drive = function('<SNR>' . s:SID() . '_s:HotRod_drive')
defm drive return hotRodObj
if self.topSpeed > 140 endfunction
echo "Ahhhhhhh!"
else function! <SID>s:HotRod_drive() dict
echo "Nice" if self.topSpeed ># 140
end echo "Ahhhhhhh!"
end else
end echo "Nice"
endif
newCar = new HotRod("chevy", "mustang", "red", 160) endfunction
newCar.drive()
let s:newCar = s:HotRodConstructor("chevy", "mustang", "red", 160)
call s:newCar.drive()
=> "Ahhhhhhh!"Use of 'super' is legal only within subclasses. If arguments are given, these arguments are sent to the superclass's function of the same name. If no arguments are given and parentheses are omitted, ('super' as opposed to 'super()'), every single argument is passed to the superclass's function. This mirrors Ruby's approach.
Super can be called from an initialize (constructor) function, a public member function ('defm'), or a non-public function ('def'). An error is given during compilation if no superclass function with that name is defined.
Compiling Riml
To compile a riml file named 'example.riml' that resides in the current directory:
$ riml -c example.rimlThis will create a new VimL file named 'example.vim' in the current directory.
riml_source
It's useful to split a project into many files. For example, imagine we're creating a plugin called 'awesome' that does something totally awesome, and it relies on another library we wrote called 'my_framework' that's also written in Riml.
Somewhere in 'awesome.riml', we have the line:
riml_source 'my_framework.riml'This will compile the file 'my_framework.riml' and create a VimL file named 'my_framework.vim'.
In 'awesome.riml', that line will be compiled to:
source 'my_framework.vim'This process is recursive, meaning that if 'my_framework.riml' riml_source's other files, then those files will be compiled as well.
The previous example would only work if 'my_framework.riml' were in the current working directory where the compilation command 'riml -c' was issued. In order to tell the compiler where to look for files that are being riml_source'd, you can provide an environment variable or a command-line option of colon-separated paths. For example:
riml -c awesome.riml -S 'first_dir:second_dir'With this command, the compiler will look for files that are riml_source'd first in ./first_dir, then in ./second_dir if not found in first_dir. The same can be achieved by:
riml -c awesome.riml RIML_SOURCE_PATH='first_dir:second_dir'Paths can either be relative or absolute. The sourced files that are compiled will end up in the same directory in which they were found.
riml_include
Sometimes it's useful to have many files in development, but to include a file's contents into another file during the build process. This is much like the C preprocessor's #include directive.
To include a file named 'my_lib.riml':
riml_include 'my_lib.riml'This compiles the file and includes its content in place of the riml_include directive itself. Much like riml_sourcing, the process is recursive. If 'my_lib.riml' includes files, these files are also compiled and will be part of the inclusion. Note that riml_include does not create a new file like riml_source does.
Like riml_source, riml_include looks in a set of ordered paths to find the files to include, with the default being the working directory in which the compilation command was issued. Here are some examples of commands that set the include path:
riml -c 'main.riml' -I 'lib:helpers:debug'
riml -c 'plugin_name.riml' RIML_INCLUDE_PATH='lib:modules'Since riml_include acts as a sort of preprocessor, it cannot be issued inside of a conditional, function, or anything dynamic. It must be at the top-level (non-nested).
Also, if you are included lots of files, you don't have to worry about multiple-inclusion, or which file should be included before which, because during the compile phase the class dependencies are resolved and the inclusions are re-ordered based on this info. It's easiest to include all the dependencies you need to include for each file without worrying about repetitions or ordering.
Take a look at https://github.com/dsawardekar/portkey/blob/develop/lib/includes/portkey_inc.riml to see how you can treat a list of inclusions as a kind of header file that other files include.
Incompatibilities with VimL
Riml aims to be as compatible with VimL as possible, therefore any legal VimL should be legal Riml as well. Unfortunately, this is not 100% possible as Vim is an old and cryptic beast, and trying to create grammar for every possible VimL construct would be a nightmare. In practice, however, when I've transformed plain old VimL plugins to be Riml-compatible, only a couple of ':' needed to be placed in strategic locations for it to be valid Riml. This is explained below.
Ex-literals
Fortunately, there are some pretty simple rules to follow to get valid Riml.
When doing anything with autocommands, normal, commands, set, ranges, etc... simply do:
:autocmd BufEnter * blahblah...That is, prepend ':' to the line. When a line starts with ':', it passes directly through the compiler and no transformations occur. This includes string interpolation, which is ignored as well.
In Riml, when a line starts with ':' it's called an ex-literal.
Ex-literals are necessary for the following:
- autocommands
- command definitions
- set
- ranges (:h cmdline-ranges)
- normal (:h normal)
- mappings
- augroups
Basically anything that isn't a class, number, string, list, dict, function call, function definition, loop or if construct, variable definition or unlet, etc... needs to be an ex-literal in Riml.
Note that, like 'echo' which isn't a builtin function (:h functions) but is still legal Riml, 'execute' is also allowed as it takes a string. This is extremely useful, as we can now use execute with a string that allows interpolation!
Imagine having to write a grammar rule for the following:
set statusline+=[%{strlen(&fenc)?&fenc:'none'}, " File encodingSince there's no string after the '+=', it makes it very hard. So when the compiler can't parse a file correctly, prepend those lines with ':' and all should be well.
Abbreviations
In VimL, there are abbreviations for everything; even "keywords" like 'function' can be abbreviated. In Riml, abbreviations are not allowed. This makes Riml much easier to read and understand.
Everything Else That Works
Everything not mentioned above as illegal is legal Riml. Here's a short (non-comprehensive) list of constructs which are legal Riml but not mentioned in any of the examples:
- try/catch/finally blocks and throw
- curly-brace variable and function names
- while (and until) loops
- ternary operators
- exponents
- line continuations
- autoloadable variables and functions (:h autoload)
- let unpack (:h let-unpack)
- Much more!
Tools
- vim-riml - Syntax highlighting and automatic indentation for Riml files.