Skip to content

Latest commit

 

History

History
1055 lines (820 loc) · 40.4 KB

language-guide.md

File metadata and controls

1055 lines (820 loc) · 40.4 KB

Language guide

Overview

The mgmt tool has various frontends, each of which may produce a stream of between zero or more graphs that are passed to the engine for desired state application. In almost all scenarios, you're going to want to use the language frontend. This guide describes some of the internals of the language.

Theory

The mgmt language is a declarative (immutable) functional, reactive programming language. It is implemented in golang. A longer introduction to the language is available as a blog post here!

Types

All expressions must have a type. A composite type such as a list of strings ([]str) is different from a list of integers ([]int).

There is a variant type in the language's type system, but it is only used internally and only appears briefly when needed for type unification hints during static polymorphic function generation. This is an advanced topic which is not required for normal usage of the software.

The implementation of the internal types can be found in lang/types/.

bool

A true or false value.

str

Any "string!" enclosed in quotes.

int

A number like 42 or -13. Integers are represented internally as golang's int64.

float

A floating point number like: 3.1415926. Float's are represented internally as golang's float64.

list

An ordered collection of values of the same type, eg: [6, 7, 8, 9,]. It is worth mentioning that empty lists have a type, although without type hints it can be impossible to infer the item's type.

map

An unordered set of unique keys of the same type and corresponding value pairs of another type, eg: {"boiling" => 100, "freezing" => 0, "room" => 25, "house" => 22, "canada" => -30,}. That is to say, all of the keys must have the same type, and all of the values must have the same type. You can use any type for either, although it is probably advisable to avoid using very complex types as map keys.

struct

An ordered set of field names and corresponding values, each of their own type, eg: struct{answer => "42", james => "awesome", is_mgmt_awesome => true,}. These are useful for combining more than one type into the same value. Note the syntactical difference between these and map's: the key's in map's have types, and as a result, string keys are enclosed in quotes, whereas struct fields are not string values, and as such are bare and specified without quotes.

func

An ordered set of optionally named, differently typed input arguments, and a return type, eg: func(s str) int or: func(bool, []str, {str: float}) struct{foo str; bar int}.

Expressions

Expressions, and the Expr interface need to be better documented. For now please consume lang/interfaces/ast.go. These docs will be expanded on when things are more certain to be stable.

Statements

There are a very small number of statements in our language. They include:

  • bind: bind's an expression to a variable within that scope without output

    • eg: $x = 42
  • if: produces up to one branch of statements based on a conditional expression

     if <conditional> {
     	<statements>
     } else {
     	# the else branch is optional for if statements
     	<statements>
     }
    
  • resource: produces a resource

     file "/tmp/hello" {
     	content => "world",
     	mode => "o=rwx",
     }
    
  • edge: produces an edge

     File["/tmp/hello"] -> Print["alert4"]
    
  • class: bind's a list of statements to a class name in scope without output

     class foo {
     	# some statements go here
     }
    

    or

     class bar($a, $b) { # a parameterized class
     	# some statements go here
     }
    
  • include: include a particular class at this location producing output

     include foo
    
     include bar("hello", 42)
     include bar("world", 13) # an include can be called multiple times
    
  • import: import a particular scope from this location at a given namespace

     # a system module import
     import "fmt"
    
     # a local, single file import (relative path, not a module)
     import "dir1/file.mcl"
    
     # a local, module import (relative path, contents are a module)
     import "dir2/"
    
     # a remote module import (absolute remote path, contents are a module)
     import "git://github.com/purpleidea/mgmt-example1/"
    

    or

     import "fmt" as *	# contents namespaced into top-level names
     import "foo.mcl"	# namespaced as foo
     import "dir1/" as bar	# namespaced as bar
     import "git://github.com/purpleidea/mgmt-example1/"	# namespaced as example1
    

All statements produce output. Output consists of between zero and more edges and resources. A resource statement can produce a resource, whereas an if statement produces whatever the chosen branch produces. Ultimately the goal of executing our programs is to produce a list of resources, which along with the produced edges, is built into a resource graph. This graph is then passed to the engine for desired state application.

Bind

This section needs better documentation.

If

This section needs better documentation.

Resource

Resources express the idempotent workloads that we want to have apply on our system. They correspond to vertices in a graph which represent the order in which their declared state is applied. You will usually want to pass in a number of parameters and associated values to the resource to control how it behaves. For example, setting the content parameter of a file resource to the string hello, will cause the contents of that file to contain the string hello after it has run.

Undefined parameters

For some parameters, there is a distinction between an unspecified parameter, and a parameter with a zero value. For example, for the file resource, you might choose to set the content parameter to be the empty string, which would ensure that the file has a length of zero. Alternatively you might wish to not specify the file contents at all, which would leave that property undefined. If you omit listing a property, then it will be undefined. To control this property programmatically, you need to specify an is-defined value, as well as the value to use if that boolean is true. You can do this with the resource-specific elvis operator.

$b = true # change me to false and then try editing the file manually
file "/tmp/mgmt-elvis" {
	content => $b ?: "hello world\n",
	state => $const.res.file.state.exists,
}

This example is static, however you can imagine that the $b value might be chosen in a programmatic way, even one in which that value varies over time. If it evaluates to true, then the parameter will be used. If no elvis operator is specified, then the parameter value will also be used. If the parameter is not specified, then it will obviously not be used.

Meta parameters

Resources may specify meta parameters. To do so, you must add them as you would a regular parameter, except that they start with Meta and are capitalized. Eg:

file "/tmp/f1" {
	content => "hello!\n",

	Meta:noop => true,
	Meta:delay => $b ?: 42,
	Meta:autoedge => false,
}

As you can see, they also support the elvis operator, and you can add as many as you like. While it is not recommended to add the same meta parameter more than once, it does not currently cause an error, and even though the result of doing so is officially undefined, it will currently take the last specified value.

You may also specify a single meta parameter struct. This is useful if you'd like to reuse a value, or build a combined value programmatically. For example:

file "/tmp/f1" {
	content => "hello!\n",

	Meta => $b ?: struct{
		noop => false,
		retry => -1,
		delay => 0,
		poll => 5,
		limit => 4.2,
		burst => 3,
		sema => ["foo:1", "bar:3",],
		autoedge => true,
		autogroup => false,
	},
}

Remember that the top-level Meta field supports the elvis operator, while the individual struct fields in the struct type do not. This is to be expected, but since they are syntactically similar, it is worth mentioning to avoid confusion.

Please note that at the moment, you must specify a full metaparams struct, since partial struct types are currently not supported in the language. Patches are welcome if you'd like to add this tricky feature!

Resource naming

Each resource must have a unique name of type str that is used to uniquely identify that resource, and can be used in the functioning of the resource at that resources discretion. For example, the file resource uses the unique name value to specify the path.

Alternatively, the name value may be a list of strings []str to build a list of resources, each with a name from that list. When this is done, each resource will use the same set of parameters. The list of internal edges specified in the same resource block is created intelligently to have the appropriate edge for each separate resource.

Using this construct is a veiled form of looping (iteration). This technique is one of many ways you can perform iterative tasks that you might have traditionally used a for loop for instead. This is preferred, because flow control is error-prone and can make for less readable code.

The single str variation, may only be used when it is possible for the compiler to determine statically that the value is of that type. Otherwise, it will assume it to be a list of strings. Programmers should explicitly wrap their variables in a string by interpolation to force this static str determination, or in square brackets to force a list. The former is generally preferable because it generates a smaller function graph since it doesn't need to build a list.

Internal edges

Resources may also declare edges internally. The edges may point to or from another resource, and may optionally include a notification. The four properties are: Before, Depend, Notify and Listen. The first two represent normal edge dependencies, and the second two are normal edge dependencies that also send notifications. You may have multiples of these per resource, including multiple Depend lines if necessary. Each of these properties also supports the conditional inclusion elvis operator as well.

For example, you may write:

$b = true # for example purposes
if $b {
	pkg "drbd" {
		state => "installed",

		# multiple properties may be used in the same resource
		Before => File["/etc/drbd.conf"],
		Before => Svc["drbd"],
	}
}
file "/etc/drbd.conf" {
	content => "some config",

	Depend => $b ?: Pkg["drbd"],
	Notify => Svc["drbd"],
}
svc "drbd" {
	state => "running",
}

There are two unique properties about these edges that is different from what you might expect from other automation software:

  1. The ability to specify multiples of these properties allows you to avoid having to manage arrays and conditional trees of these different dependencies.
  2. The keywords all have the same length, which means your code lines up nicely.

Edge

Edges express dependencies in the graph of resources which are output. They can be chained as a pair, or in any greater number. For example, you may write:

Pkg["drbd"] -> File["/etc/drbd.conf"] -> Svc["drbd"]

to express a relationship between three resources. The first character in the resource kind must be capitalized so that the parser can't ascertain unambiguously that we are referring to a dependency relationship.

Edge naming

Each edge must have a unique name of type str that is used to uniquely identify that edge, and can be used in the functioning of the edge at its discretion.

Alternatively, the name value may be a list of strings []str to build a list of edges, each with a name from that list.

Using this construct is a veiled form of looping (iteration). This technique is one of many ways you can perform iterative tasks that you might have traditionally used a for loop for instead. This is preferred, because flow control is error-prone and can make for less readable code.

The single str variation, may only be used when it is possible for the compiler to determine statically that the value is of that type. Otherwise, it will assume it to be a list of strings. Programmers should explicitly wrap their variables in a string by interpolation to force this static str determination, or in square brackets to force a list. The former is generally preferable because it generates a smaller function graph since it doesn't need to build a list.

Class

A class is a grouping structure that bind's a list of statements to a name in the scope where it is defined. It doesn't directly produce any output. To produce output it must be called via the include statement.

Defining classes follows the same scoping and shadowing rules that is applied to the bind statement, although they exist in a separate namespace. In other words you can have a variable named foo and a class named foo in the same scope without any conflicts.

Classes can be both parameterized or naked. If a parameterized class is defined, then the argument types must be either specified manually, or inferred with the type unification algorithm. One interesting property is that the same class definition can be used with include via two different input signatures, although in practice this is probably fairly rare. Some usage examples include:

A naked class definition:

class foo {
	# some statements go here
}

A parameterized class with both input types being inferred if possible:

class bar($a, $b) {
	# some statements go here
}

A parameterized class with one type specified statically and one being inferred:

class baz($a str, $b) {
	# some statements go here
}

Classes can also be nested within other classes. Here's a contrived example:

import "fmt"
class c1($a, $b) {
	# nested class definition
	class c2($c) {
		test $a {
			stringptr => fmt.printf("%s is %d", $b, $c),
		}
	}

	if $a == "t1" {
		include c2(42)
	}
}

Defining polymorphic classes was considered but is not currently allowed at this time.

Recursive classes are not currently supported and it is not clear if they will be in the future. Discussion about this topic is welcome on the mailing list.

A class names can contain colons to indicate it is nested inside of the class in the same scope which is named with the prefix indicated by colon separation. Instead of needing to repeatedly indent the child classes, we can instead prefix them at the definition site (where created with the class keyword) with the name of the parent class, followed by a colon, to get the desired embedded sugar.

For example, instead of writing:

class base() {
	class inner() {
		class deepest() {
		}
	}
}

You can instead write:

class base() {
}
class base:inner() {
}
class base:inner:deepest() {
}

Of course, you can only access any of the inner classes by first including (with the include keyword) a parent class, and then subsequently including the inner one.

Include

The include statement causes the previously defined class to produce the contained output. This statement must be called with parameters if the named class is defined with those.

The defined class can be called as many times as you'd like either within the same scope or within different scopes. If a class uses inferred type input parameters, then the same class can even be called with different signatures. Whether the output is useful and whether there is a unique type unification solution is dependent on your code.

Classes can be included under a new scoped prefix by using the as field and an identifier. When used in this manner, the captured scope of the class at its definition site are made available in the scope of the include. Variables, functions, and child classes are all exported.

Variables are available in the include scope:

import "fmt"

class c1 {
	test "t1" {}	# gets pulled out
	$x = "hello"	# gets exported
}
include c1 as i1

test "print0" {
	anotherstr => fmt.printf("%s", $i1.x),	# hello
	onlyshow => ["AnotherStr",], # displays nicer
}

Classes are also available in the new scope:

import "fmt"
class c1 {
	test "t1" {}	# gets pulled out
	$x = "hello"	# gets exported

	class c0 {
		test "t2" {}
		$x = "goodbye"
	}
}
include c1 as i1
include i1.c0 as i0

test "print0" {
	anotherstr => fmt.printf("%s", $i1.x),	# hello
	onlyshow => ["AnotherStr",], # displays nicer
}
test "print1" {
	anotherstr => fmt.printf("%s", $i0.x),	# goodbye
	onlyshow => ["AnotherStr",], # displays nicer
}

Of course classes can be parameterized too, and those variables specified during the include.

Import

The import statement imports a scope into the specified namespace. A scope can contain variable, class, and function definitions. All are statements. Furthermore, since each of these have different logical uses, you could theoretically import a scope that contains an int variable named foo, a class named foo, and a function named foo as well. Keep in mind that variables can contain functions (they can have a type of function) and are commonly called lambdas.

There are a few different kinds of imports. They differ by the string contents that you specify. Short single word, or multiple-word tokens separated by zero or more slashes are system imports. Eg: math, fmt, or even math/trig. Local imports are path imports that are relative to the current directory. They can either import a single mcl file, or an entire well-formed module. Eg: file1.mcl or dir1/. Lastly, you can have a remote import. This must be an absolute path to a well-formed module. The common transport is git, and it can be represented via an FQDN. Eg: git://github.com/purpleidea/mgmt-example1/.

The namespace that any of these are imported into depends on how you use the import statement. By default, each kind of import will have a logic namespace identifier associated with it. System imports use the last token in their name. Eg: fmt would be imported as fmt and math/trig would be imported as trig. Local imports do the same, except the required .mcl extension, or trailing slash are removed. Eg: foo/file1.mcl would be imported as file1 and bar/baz/ would be imported as baz. Remote imports use some more complex rules. In general, well-named modules that contain a final directory name in the form: mgmt-whatever/ will be named whatever. Otherwise, the last path token will be converted to lowercase and the dashes will be converted to underscores. The rules for remote imports might change, and should not be considered stable.

In any of the import cases, you can change the namespace that you're imported into. Simply add the as whatever text at the end of the import, and whatever will be the name of the namespace. Please note that whatever is not surrounded by quotes, since it is an identifier, and not a string. If you'd like to add all of the import contents into the top-level scope, you can use the as * text to dump all of the contents in. This is generally not recommended, as it might cause a conflict with another identifier.

Stages

The mgmt compiler runs in a number of stages. In order of execution they are:

All of the above needs to be done every time the source code changes. After this point, the function engine runs and produces events. On every event, we "interpret" which produces a resource graph. This series of resource graphs are passed to the engine as they are produced.

What follows are some notes about each step.

Lexing

Lexing is done using nex. It is a pure-golang implementation which is similar to Lex or Flex, but which produces golang code instead of C. It integrates reasonably well with golang's yacc which is used for parsing. The token definitions are in: lang/lexer.nex. Lexing and parsing run together by calling the LexParse method.

Parsing

The parser used is golang's implementation of yacc. The documentation is quite abysmal, so it's helpful to rely on the documentation from standard yacc and trial and error. One small advantage yacc has over standard yacc is that it can produce error messages from examples. The best documentation is to examine the source. There is a short write up available here. The yacc file exists at: lang/parser.y. Lexing and parsing run together by calling the LexParse method.

Interpolation

Interpolation is used to transform the AST (which was produced from lexing and parsing) into one which is either identical or different. It expands strings which might contain expressions to be interpolated (eg: "the answer is: ${foo}") and can be used for other scenarios in which one statement or expression would be better represented by a larger AST. Most nodes in the AST simply return their own node address, and do not modify the AST.

This stage also implements the class nesting when it finds class names with colons that should be nested inside of a base class. Currently this does modify the AST for efficiency and simplicity.

Scope propagation

Scope propagation passes the parent scope (starting with the top-level, built-in scope) down through the AST. This is necessary so that children nodes can access variables in the scope if needed. Most AST node's simply pass on the scope without making any changes. The ExprVar node naturally consumes scope's and the StmtProg node cleverly passes the scope through in the order expected for the out-of-order bind logic to work.

This step typically calls the ordering algorithm to determine the correct order of statements in a program.

Type unification

Each expression must have a known type. The unpleasant option is to force the programmer to specify by annotation every type throughout their whole program so that each Expr node in the AST knows what to expect. Type annotation is allowed in situations when you want to explicitly specify a type, or when the compiler cannot deduce it, however, most of it can usually be inferred.

For type inference to work, each Stmt node in the AST implements a TypeCheck method which is able to return a list of invariants that must hold true. This starts at the top most AST node, and gets called through to it's children to assemble a giant list of invariants. The invariants all have the same form. They specify that a particular expression corresponds to two particular types which may both contain unification variables.

Each Expr node in the AST implements an Infer and Check method. The Infer method returns the type of that node along with a list of invariants as described above. Unification variables can of course be used throughout. The Check method always uses a generic check implementation and generally doesn't need to be implemented by the user.

Once the list of invariants has been collected, they are run through an invariant solver. The solver can return either return successfully or with an error. If the solver returns successfully, it means that it has found a single mapping between every expression and it's corresponding type. At this point it is a simple task to run SetType on every expression so that the types are known. During this stage, each SetType method verifies that it's a compatible type that it can use. If either that method or if the solver returns in error, it is usually due to one of two possibilities:

  1. Ambiguity

    The solver does not have enough information to make a definitive or unique determination about the expression to type mappings. The set of invariants is ambiguous, and we cannot continue. An error will be returned to the programmer. In this scenario the user will probably need to add a type annotation, possibly because of a design bug in the user's program.

  2. Conflict

    The solver has conflicting information that cannot be reconciled. In this situation an explicit conflict has been found. If two invariants are found which both expect a particular expression to have different types, then it is not possible to find a valid solution. This almost always happens if the user has made a type error in their program.

Only one solver currently exists, but it is possible to easily plug in an alternate implementation if someone wants to experiment with the art of solver design and would like to propose a more logical or performant variant.

Function graph generation

At this point we have a fully typed AST. The AST must now be transformed into a directed, acyclic graph (DAG) data structure that represents the flow of data as necessary for everything to be reactive. Note that this graph is different from the resource graph which is produced and sent to the engine. It is just a coincidence that both happen to be DAG's. (You aren't surprised when you see a list data structure show up in more than one place, do you?)

To produce this graph, each node has a Graph method which it can call. This starts at the top most node, and is called down through the AST. The edges in the graphs must represent the individual expression values which are passed from node to node. The names of the edges must match the function type argument names which are used in the definition of the corresponding function. These corresponding functions must exist for each expression node and are produced as the vertices of this returned graph. This is built for the function engine.

Function engine creation and validation

Finally we have a graph of the data flows. The function engine must first initialize which creates references to each of the necessary function implementations, and gets information about each one. It then needs to be type checked to ensure that the data flows all correctly match what is expected. If you were to pass an int to a function expecting a bool, this would be a problem. If all goes well, the program should get run shortly.

Function engine running and interpret

At this point the function engine runs. It produces a stream of events which cause the Output() method of the top-level program to run, which produces the list of resources and edges. These are then transformed into the resource graph which is passed to the engine.

Function API

If you'd like to create a built-in, core function, you'll need to implement the function API interface named Func. It can be found in lang/interfaces/func.go. Your function must have a specific type. For example, a simple math function might have a signature of func(x int, y int) int. The simple functions have their types known before compile time. You may also include unification variables in the function signature as long as the top-level type is a function.

A separate discussion on this matter can be found in the function guide.

What follows are each of the method signatures and a description of each. Failure to implement the API correctly can cause the function graph engine to block, or the program to panic.

Info

Info() *Info

The Info method must return a struct containing some information about your function. The struct has the following type:

type Info struct {
	Sig  *types.Type // the signature of the function, must be KindFunc
}

You must implement this correctly. Other fields in the Info struct may be added in the future. This method is usually called before any other, and should not depend on any other method being called first. Other methods must not depend on this method being called first.

If you use any unification variables in the function signature, then your function will not be made available for use inside templates. This is a limitation of the golang templating library. In the future if this limitation proves to be significantly annoying, we might consider writing our own template library.

Example

func (obj *FooFunc) Info() *interfaces.Info {
	return &interfaces.Info{
		Sig: types.NewType("func(a str, b int) float"),
	}
}

Example

This example contains unification variables.

func (obj *FooFunc) Info() *interfaces.Info {
	return &interfaces.Info{
		Sig: types.NewType("func(a ?1, b ?2, foo [?3]) ?1"),
	}
}

Init

Init(*Init) error

Init is called by the function graph engine to create an implementation of this function. It is passed in a struct of the following form:

type Init struct {
	Hostname string // uuid for the host
	Input  chan types.Value // Engine will close `input` chan
	Output chan types.Value // Stream must close `output` chan
	World  resources.World
	Debug  bool
	Logf   func(format string, v ...interface{})
}

These values and references may be used (wisely) inside your function. Input will contain a channel of input structs matching the expected input signature for your function. Output will be the channel which you must send values to whenever a new value should be produced. This must be done in the Stream() function. You may carefully use World to access functionality provided by the engine. You may use Logf to log informational messages, however there is no guarantee that they will be displayed to the user. Debug specifies whether the function is running in a user-requested debug mode. This might cause you to want to print more log messages for example. You will need to save references to any or all of these info fields that you wish to use in the struct implementing this Func interface. At a minimum you will need to save Output as a minimum of one value must be produced.

Example

Please see the example functions in
[lang/core/](https://github.com/purpleidea/mgmt/tree/master/lang/core/).

Stream

Stream(context.Context) error

Stream is called by the function engine when it is ready for your function to start accepting input and producing output. You must always produce at least one value. Failure to produce at least one value will probably cause the function engine to hang waiting for your output. This function must close the Output channel when it has no more values to send. The engine will close the Input channel when it has no more values to send. This may or may not influence whether or not you close the Output channel. You must shutdown if the input context cancels.

Example

Please see the example functions in
[lang/core/](https://github.com/purpleidea/mgmt/tree/master/lang/core/).

BuildableFunc Function API

For some functions, it might be helpful to have a function which needs a "build" step which is run after type unification. This step can be used to build the function using the determined type, but it may also just be used for checking that unification picked a valid solution.

Suppose you want to implement a function which can assume different type signatures. The mgmt language does not support polymorphic types-- you must use static types throughout the language, however, it is legal to implement a function which can take different specific type signatures based on how it is used. For example, you might wish to add a math function which could take the form of func(x int, y int) int or func(x float, y float) float depending on the input values. For this case you could use a signature containing unification variables, eg: func(x ?1, y ?1) ?1. At the end the buildable function would need to check that it received a ?1 type of either int or float, since this function might not support doing math on strings. Remember that type unification can only return zero or one solutions, it's not possible to return more than one, which is why this secondary validation step is a brilliant way to filter out invalid solutions without needing to encode them as algebraic conditions during the solver state, which would otherwise make it exponential.

InferableFunc Function API

You might also want to implement a function which takes an arbitrary number of input arguments (the number must be statically fixed at the compile time of your program though) and which returns a string or something else.

The InferableFunc interface adds ad additional FuncInfer method which you must implement to satisfy such a function implementation. This lets you dynamically generate a type signature (including unification variables) and a list of invariants before running the type unification solver. It takes as input a list of the statically known input types and input values (if any) and as well the number of input arguments specified. This is usually enough information to generate a fixed type signature of a fixed size.

Using this API should generally be pretty rare, but it is how certain special functions such as fmt.printf are built. If you'd like to implement such a function, then please notify the project authors as we're curious about your use case.

Frequently asked questions

(Send your questions as a patch to this FAQ! I'll review it, merge it, and respond by commit with the answer.)

Why am I getting a deploy.readfile error when the file actually exists?

You may be seeing an error like:

readfile: open /*/files/foo: file does not exist can't read file /files/foo?

If you look, the foo file is indeed in the files/ directory. The problem is that the files/ directory won't be seen if you didn't specify to include it as part of your deploy. To do so, chances are that all you need to do is add a metadata.yaml file into the parent directory to that files folder. This will be used as the entrypoint instead of the naked main.mcl file that you have there, and with that metadata entrypoint, you get a default files/ directory added. You can of course change the files/ path by setting a key in the metadata.yaml file, but we recommend you leave it as the default.

What is the difference between ExprIf and StmtIf?

The language contains both an if expression, and and if statement. An if expression takes a boolean conditional and it must contain exactly two branches (a then and an else branch) which each contain one expression. The if expression will return the value of one of the two branches based on the conditional.

Example:

# this is an if expression, and both branches must exist
$b = true
$x = if $b {
	42
} else {
	-13
}

The if statement also takes a boolean conditional, but it may have either one or two branches. Branches must only directly contain statements. The if statement does not return any value, but it does produce output when it is evaluated. The output consists primarily of resources (vertices) and edges.

Example:

# this is an if statement, and in this scenario the else branch was omitted
$b = true
if $b {
	file "/tmp/hello" {
		content => "world",
	}
}

What is the difference types.Value.Str() and types.Value.String()?

In the lang/types library, there is a types.Value interface. Every value in our type system must implement this interface. One of the methods in this interface is the String() string method. This lets you print a representation of the value. You will probably never need to use this method.

In addition, the types.Value interface implements a number of helper functions which return the value as an equivalent golang type. If you know that the value is a bool, you can call x.Bool() on it. If it's a string you can call x.Str(). Make sure not to call one of those type methods unless you know the value is of that type, or you will trigger a panic!

I created a &ListValue{} but it's not working!

If you create a base type like bool, str, int, or float, all you need to do is build the &BoolValue and set the V field. Eg:

someBool := &types.BoolValue{V: true}

If you are building a container type like list, map, struct, or func, then you also need to specify the type of the contained values. This is because a list has a type of []str, or []int, or even [][]foo. Eg:

someListOfStrings := &types.ListValue{
	T: types.NewType("[]str"),	# must match the contents!
	V: []types.Value{
		&types.StrValue{V: "a"},
		&types.StrValue{V: "bb"},
		&types.StrValue{V: "ccc"},
	},
}

If you don't build these properly, then you will cause a panic! Even empty lists have a type.

Is the class statement a singleton?

Not really, but practically it can be used as such. The class statement is not a singleton since it can be called multiple times in different locations, and it can also be parameterized and called multiple times (with include) using different input parameters. The reason it can be used as such is that statement output (from multple classes) that is compatible (and usually identical) will be automatically collated and have the duplicates removed. In that way, you can assume that an unparameterized class is always a singleton, and that parameterized classes can often be singletons depending on their contents and if they are called in an identical way or not. In reality the de-duplication actually happens at the resource output level, so anything that produces multiple compatible resources is allowed.

Are recursive class definitions supported?

Recursive class definitions where the contents of a class contain a self-referential include, either directly, or with indirection via any other number of classes is not supported. It's not clear if it ever will be in the future, unless we decide it's worth the extra complexity. The reason is that our FRP actually generates a static graph which doesn't change unless the code does. To support dynamic graphs would require our FRP to be a "higher-order" FRP, instead of the simpler "first-order" FRP that it is now. You might want to verify that I got the nomenclature correct. If it turns out that there's an important advantage to supporting a higher-order FRP in mgmt, then we can consider that in the future.

I realized that recursion would require a static graph when I considered the structure required for a simple recursive class definition. If some "depth" value wasn't known statically by compile time, then there would be no way to know how large the graph would grow, and furthermore, the graph would need to change if that "depth" value changed.

I don't like the mgmt language, is there an alternative?

Yes, the language is just one of the available "frontends" that passes a stream of graphs to the engine "backend". While it is the recommended way of using mgmt, you're welcome to either use an alternate frontend, or write your own. To write your own frontend, you must implement the GAPI interface.

I'm an expert in FRP, and you got it all wrong; even the names of things!

I am certainly no expert in FRP, and I've certainly got lots more to learn. One thing FRP experts might notice is that some of the concepts from FRP are either named differently, or are notably absent.

In mgmt, we don't talk about behaviours, events, or signals in the strict FRP definitons of the words. Firstly, because we only support discretized, streams of values with no plan to add continuous semantics. Secondly, because we prefer to use terms which are more natural and relatable to what our target audience is expecting. Our users are more likely to have a background in Physiology, or systems administration than a background in FRP.

Having said that, we hope that the FRP community will engage with us and help improve the parts that we got wrong. Even if that means adding continuous behaviours!

This is brilliant, may I give you a high-five?

Thank you, and yes, probably. "Props" may also be accepted, although patches are preferred. If you can't do either, donations to support the project are welcome too!

Where can I find more information about mgmt?

Additional blog posts, videos and other material is available!.

Suggestions

If you have any ideas for changes or other improvements to the language, please let us know! We're still pre 1.0 and pre 0.1 and happy to change it in order to get it right!