RTC is no longer maintained. It has been replaced by RDL (github.com/plum-umd/rdl), which provides a more practal approach to types for Ruby.
Rtc (Ruby Type Checker) can be loaded via a simple
require 'rtc'
at the beginning of a file. Doing this give you access to the following functions (specific usage is given below):
- rtc_typesig
-
annotate a class as polymorphic (see “Polymorphic Classes”)
- rtc_annotated
-
gain access to annotation functions within a class (see “Annotating a class”)
- rtc_no_subtype
-
tell Rtc that a class should not be considered a subtype of it’s parent class (introduces a break in the inheritance chain). Example usage:
class MyClass rtc_no_subtype end
this is equivalent to writing:
class MyClass rtc_annotated no_subtype end
This let’s you break the inheritance chain without opting into function annotation.
If you place rtc_annotated in a class definition, this opens up access to the following functions:
- typesig
-
annotate a function type
- no_subtype
-
see the above explanation of rtc_no_subtype
- define_iterator
-
used in the definition of polymorphic classes (see “Polymorphic Classes” below)
- define_iterators
-
used in the definition of polymorphic classes (see “Polymorphic Classes” below)
This function let’s you annotate the type of instance fields and instance methods, as well as class methods and fields. It’s basic form is:
typesig("name: type")
This can be used to annotate a method, or a field
Method annotations take the form:
typesig("foo: (argument type list) -> return type")
In this example, foo is the name of the function being annotated. If the function contains special characters (e.g., *, [], ?, !, etc.) then the name must be enclosed in single quotes, e.g.
typesig("'*': (Fixnum) -> Fixnum")
would be the annotation on the * method of a Fixnum object. The argument type list is a (possibly empty) comma separated list of type expressions. The most basic form of a type expression is a simple class name, for example, Fixnum, String, etc. The resolution of these names to actual classes is subject to the rules in “Class Name Resolution” below. If a polymorphic class is used, then you can specify the type of the type parameters using the ‘<>’ syntax familiar to C++ and Java programmers. For example, if you wished to say a function takes an Array of Fixnums, you would write Array<Fixnum>. You may also use more complex type expressions as type parameters, specifically Union Types and Symbol Types (the parser will accept Optional types and Vararg types as type paremeters, but doing so will not work the way you want). There are several other type expressions available:
- Union types
-
Indicates a union of different type expressions. For example, the type signature:
typesig("foo: (Fixnum or String) -> Fixnum")
is a method that can take either a Fixnum or a String, and returns a Fixnum. Any type expression may be a part of a union. To construct a union, separate several type expressions with “or”, i.e., type expr 1 or type expr 2 or …
- Symbol types
-
Rtc lifts Symbols into the type system. For example, the symbol
:foohas type :foo. Currently Rtc only supports symbols that do not contain spaces, so that symbols like:'complex symbol'are not supported.
- Optional types
-
You may say an argument is optional by prefixing the type expression with a single ?
- Vararg types
-
You may say that a function takes a rest argument with type t by writing *t. For example, *Fixnum or even *Fixnum or String
- Structural types
-
An ‘anonymous’ class specification. For example, if you wish to specify that a function takes an object that supports at least the addition operator, you would write
typesig("foo: [ '+': (Fixnum) -> Fixnum ]")
Multiple types are separated by a comma. Both field annotations (see below) and method annotations are supported.
- Top Type
-
The top type is a special type for which everything in the type system of Rtc is a subtype. It is the equivalent of “Object” in Java. The top type is expressed with .?, so
typesig("takes_any_type: (.?) -> .?")
You may specify multiple annotations on a single function. For instance, if you want to say a function takes a Fixnum and returns a Fixnum, AND it takes a String and returns a String you could write:
typesig("my_method: (Fixnum) -> Fixnum") typesig("my_method: (String) -> String")
Note that this is different from type signature:
typesig("my_method: (Fixnum or String) -> (Fixnum or String)")
as this would denote a method that could take a Fixnum and then return a String. If multiple type signatures are applied to a method, they are automatically intersected (see below on the handling of intersection types).
You may also specify nameless type signatures. This is especially useful when several type signatures are to be applied to a method. A nameless type signature is the same as a regular type signature, except that it lacks a name, e.g.
typesig("(Fixnum) -> Fixnum")
Nameless type signatures are applied to the next defined method in a class. So in the following example, all the nameless type signatures would be applied to the bar method:
class MyClass # ... typesig("(Fixnum) -> Fixnum") typesig("(?Fixnum,?String) -> Float") def bar(...) # ... end end
In some cases you may not care particularly about the return type of a function. The initialize method of a class is one such example. In this case, you may simply write
typesig("my_method: (arguments)")
This is actually just shorthand for
typesig("my_method: (arguments) -> .?")
A field may be annotated using the following:
typesig("@number_field: Fixnum")
From a type checking stand point this is shorthand for the following:
typesig("number_field: () -> Fixnum") typesig("'number_field=': (Fixnum) -> Fixnum")
However, annotating a field using the above syntax allows the type of a field to be queried at runtime (see the “Using Annotations” below).
Like method annotations, multiple annotation may be applied to a field. For example, the following:
typesig("@foo: String") typesig("@foo: Fixnum")
From a type checking stand point, this is shorthand for:
typesig("foo: () -> String") typesig("foo: () -> Fixnum") typesig("'foo=': (Fixnum) -> Fixnum") typesig("'foo=': (String) -> String")
Recall that multiple annotations are intersected automatically. Note that the type of field is automatically unioned at runtime, so that if one were to query the type of foo field at runtime (see below “Using Annotations”) you would get back “Fixnum or String”.
Rtc also allows annotation of class methods. To annotate a class method, prefix the method name with “self.”. Note that if you need to define a class method that contains special characters, the quotes must only be applied to the method name, not the “self.” portion. For example, typesig(“self.‘*’: (Fixnum) -> Fixnum”) is correct, NOT typesig(“‘self.*’: (Fixnum) -> Fixnum”). Like instance methods, class method annotations may be given before the defintion of the actual class method.
All the advanced functionality of method annotations, with the exception of nameless method annotations, are available also for class method annotations.
Class fields operate almost exactly the same as instance fields, but intead the name is prefixed @@ instead of @.
If you annotate a method two things happen: the type of the method is recorded and the method is wrapped to perform type checking on the arguments and the return type.
The more complex rules and corner cases used in type checking are detailed below (see “Type Checking”), but a brief overview is given here. If, for example, you annotate the method foo as follows:
class MyClass rtc_annotated typesig("foo: (Fixnum) -> String") # definition of foo follows... end
And then attempt to call foo with a String, the type checker will catch this error and complain. By default, if a type error occurs, then the exception Rtc::TypeMismatchException is thrown, although other behavior on type mismatches is possible (see “Controlling Rtc” below). Note that the method need not be defined before the annotation is written, the type signature may be specified before the definition of a method. However, you may not specify a type signature of a method defined in a subclass in a superclass, although support for “abstract type signatures” may be added in the future. Note further that Rtc will perform return type checking as well, this checks thatyou return what you say you are, as well as ensuring people are adhering to your API. Type checking is implemented by wrapping the original method in another method. This method has it’s own parameters, so code that relies on the parameters property of a method will no longer work. However, after wrapping, instance methods may be bound, passed around, called indirectly as would a regular function. The wrapping may introduce some overhead in calling, if this is a problem for you, see the “Disabling Rtc Globally” section in “Controlling Rtc” below.
The type of a method may be queried at runtime by calling rtc_typeof on an instance of an object (this method is accessible for all objects which derive from Object). This method accepts a symbol or a string. For example, you could do:
some_object.rtc_typeof("some_method_name")
or
some_object.rtc_typeof(:some_method_name)
By default, this method will return the type signature that is defined farthest down the inheritance chain. For instance, suppose you have the following class hierarchy:
Baz < Bar < Foo < Object
If Bar defines a type for the method “m” and Baz also defines a type for the method “m”, then if you call rtc_typeof(“m”) on an instance of Baz, then the type signature in Baz will be returned. If you wish to get the type signature as it appears in a specific class, you can use rtc_instance_typeof (see below) or by passing a second arugment to rtc_typeof. The second argument must be an instance of Class, and if specified only that class will be checked for the type signature. Continuing with our example, if you wanted to get the type signature for “m” as it appears in Bar and you have an instance of Baz, you would make the following call:
baz_instance.rtc_typeof("m", Bar)
rtc_typeof also allows the querying of field field type. This is done by prefixing the string or symbol passed to rtc_typeof with a @. Thus, to get the type of the field foo on some object you would do:
some_object.rtc_typeof(:@foo)
or
some_object.rtc_typeof("@foo")
As with methods, the class in which to look for the type may be specified with a second argument.
If you wish to know the type of a method or field and you do not have an instance of that class, you can use the rtc_instance_typeof, which is defined on all Class objects. The usage is fairly predictable, i.e.
SomeClass.rtc_instance_typeof("instance_method_name")
Note that
SomeClass.rtc_typeof("instance_method_name")
will NOT work, as it will try to find the type signature of the class method with the name instance_method_name. As with rtc_typeof, this method accepts both symbols and strings, and field types may be retrieved by prefixing a field name with @.
By default, rtc_instance_typeof will look up the inheritance chain for a type signature of a method (provided, of course, there is no a rtc_no_subtype that breaks the inheritance chain). If you wish for rtc_instance_typeof to only look in the current class, then you may specify false as a second argument to rtc_instance_typeof.
If no type signature is found, either using rtc_typeof and rtc_instance_typeof, then nil is returned.
Types of objects are accessible at runtime via the rtc_type method that is defined on all objects that derive from Object. The object returned by this method contains many internal functions, however two very important methods are provided for public use (in this section, the term “type object” refers to the object returned by the rtc_type method). The <= operator is defined on all type objects, and it can be used to test whether one type is a subtype of another. the rtc_typeof and rtc_instance_typeof methods detailed above also return type objects, allowing you to write code like the following:
my_object.rtc_typeof("foo") <= other_object.rtc_typeof("bar")
You may test if two objects are subtypes by doing the following:
my_object.rtc_type <= other_object.rtc_type
The rtc_type method also defines a to_s method that allows you to see a human readable description of the type of an object. For instance:
puts [1,2,3].rtc_typeof # prints "Array<Fixnum>"
This can be very helpful in debugging.
Ruby does not have in it’s type system a notion of polymorphic classes, but Rtc does. For instance, Array is polymorphic with one type parameter, and the Hash is parameteric with two type parameters. This section details how to make your own polymorphic classes, and to use type signatures that use polymorphism.
Defining your own polymorphic class requires two steps. First, you must annotate the type parameters your polymorphic class uses. Like in Java, these tend to be single letters, but unlike Java, they are usually lower case. To do apply this annotation, you must write something like the following:
rtc_typesig("class MyArray<t>") class MyArray rtc_annotated end
This would tell Rtc that the class MyArray is parameterized over the type parameter t. If your class is parameterized over multiple types, use a comma separated list:
rtc_typesig("class MyHash<k,v>") class MyHash rtc_annotated end
Note the use of the rtc_annotated in the class body. When annotating a class as polymorphic the class need not be defined. However, internally, Rtc remembers that you’ve said that a class is polymorphic and then waits for you to use rtc_annotated. At the call to rtc_annotated, it checks to see if the class in which the call occured has been annotated, and then applies the deferred annotation. Therefore, if you wish to annotate a class as polymorphic before it’s declaration, the declaration must include rtc_annotated.
The name resolution rules for class names inside polymorphic annotations are not the same as those in “Class Name Resolution”. Due to the method used for remembering forward annotations detailed above, the rules are stricter. Absolute class names operate under the same rules as those given in “Class Name Resolution” but the handling of relative class names is different. When resolving the relative name, ONLY the current module or class in which the call to rtc_typesig is searched for the class. Therefore, the following wil not work:
module Foo module Bar rtc_typesig("class MyHash<k,v>") end class MyHash rtc_annotated # ... end end
The above would tell Rtc that the class Foo::Bar::MyHash was polymorphic, which is not what is happening. However, generally speaking the polymorphic annotation will appear immediately before the definition of the polymorphic class, which means you will not have to worry about this rule.
As mentioned, defining the type parameter names is only the first step. You must also tell Rtc how to infer at runtime the type of each type parameter. That is, Rtc must be able to figure out that [1,2] is an Array<Fixnum> not and Array<String> or Array<Float or String>. To do this you must use the define_iterator or define_iterators function. The best explanation of the use of these functions is via example:
rtc_typesig("class Hash<k,v>") class Hash rtc_annotated define_iterators :k => :each_key, :v => :each_value end
When Rtc needs to figure out the type of k, it calls the each_key function with no arguments. When called with no arguments, the each_key returns an Enumerator over all the keys in the hash, which Rtc then iterates over to find the type of keys. More specifically, define_iterators accepts a hash that maps type parameters to method names. When Rtc wishes to find out the type of each type parameter, it calls the method name specified with no arguments. This method must return an enumerator that provides access to ALL values which will have the type assignd to k, e.g. all the keys in the hash, all the values in the hash, etc. The define_iterator method is similar, but rather accepts two arguments, the first is the type parameter, the second is the method that will provide the enumerator.
When annotating the methods of a polymorphic class, you may freely use the type parameters given in the class annotation in method signatures. For instance, you could annotated a push method of an Array<t> class as follows:
typesig("push: (t) -> Array<t>")
This means that push takes a single argument that is a subtype of the type parameter t, and returns an Array of t. An interesting feature of Rtc is that it performs replacement on these type parameters for you when you get the type of a method. For instance, if you wrote the following:
puts [1,2,3].rtc_typeof("push")
“push: (Fixnum) -> Array<Fixnum>” would be printed. Given this type signature, you might expect a call like the following:
[1,2,3].push("foo")
to fail, but this is not the case see the “Type Checking” section below for why.
You may also use the type parameter of one class as the type for another parameterized class. For example, suppose the Set class is Parameterized over k, and Arrays are parameterized over t. You can write the to_a method of a set as:
typesig("to_a: () -> Array<k>")
In this section we refer to the context as the Class or Module in which typesig is called. There are two ways to specify class names in Rtc: absolute and relative names.
Absolute names begin with a double colon “::”. When resolving the an absolute class name, searching begins from the Object context. Absolute names are the most explicit, but tend to be unwieldly and unnecessary.
Relative names are the more commonly used names. They are differentiated from absolute names in that they do not begin with a ::. When resolving a relative name, first the current context is searched. If the name is not found in that context, then the current context’s parent (that is, the class or module that contains the current context) is searched. This proceeds until the object context is reached. If the name is not found there, then the search fails. For instance, suppose we have the following Module layout:
module Foo module Bar class B; end class C; end end class B; end end
In the context of Foo::Bar::B, the relative name “B” would resolve to Foo::Bar::B (after the lookup in class B failed, the lookup in module Bar would succeed). If you wished to refer to the class Foo::B, then you would have to write “Foo::B”.
In the context of Foo::B, if you wish to reference the class Foo::Bar::B, the relative name “Bar::B” would be necessary, and similarly if the class Foo::Bar::C was to be referenced. In the case of ambiguity, it is always best to use absolute names.
Currently by default, Rtc will throw a TypeMismatchException if you pass incorrect arguments to a method. However, there are several other possibilities for the error behavior. These are controlled by assigning to Rtc.on_type_error. The following are the current error behaviors
- exception
-
The default. The exception Rtc::TypeMismatchException is thrown if a type mismatch occurs. Selected by assiging :exception
- ignore
-
Silently ignores all type errors. Selected by assigning nil or :ignore
- exit
-
Exits immediately with a non-zero error code on type error. Selected by assignin :exit
- callback
-
Calls the provided callback with the error message. Selected by assigning any object with a ‘call’ method.
- log
-
Writes the error message to a log file. Specifically, it calls the write method of an object (which may or may not be a file). Selected by assigning any object wih a ‘write’ method.
- console
-
Writes the error message to standard out. Selected by assigning :console
Rtc currently provides the setup hook Rtc.setup, which calls the selected block with the Rtc module as a start up. You can then write code like this in your app’s configuration:
Rtc.setup do |rtc| rtc.on_error = :ignore end
The method wrapping approach used by Rtc imposes some overhead to a program. If this is unacceptable, you can globally disable Rtc by setting the environmental variable RTC_DISABLE. This environmental variable must be set before Rtc is loaded for the first time. If set, all other functionality of Rtc will be available (rtc_typeof, rtc_type, etc.) but type checking will be completely disabled. You can use this to select type checking in development mode, and when you are confident you’ve caught all type errors, you can deploy your app with Rtc disabled.
This section is somewhat heavy on theoretical material. Familiarity with the concept of subtyping and polymorphism is assumed. As mentioned previously, when a function is the types of passed arguments are checked against the argument types in the type signature, and the type returned from the function is checked against the return type in the type signature. This section details how the subtype checks are performed.
The most trivial case involves subtyping two different classes. The rule adopted is simple, if some class derives from another class, it is considered a subtype of that class. The derivation need not be immediate, grandparent classes, great-grandparent classes, etc. are checked. If this behavior is not desirable, this chain can be broken with the rtc_no_subtype function. It says that the current class should not be considered a subtype of any of it’s ancestor classes. Note however, if a class B derives from a class A that has rtc_no_subtype specified, class B is still considered a subtype of class A, unless of course, class B also has rtc_no_subtype specified.
A type t considered a subtype of a union if there exists some type in the union for which t is a subtype. Therefore, Fixnum is a subtype of (Fixnum or String). A Union is a subtype of another type if every type in the union is a subtype of that type. Combining these two rules, we get the subtype rule for two unions. A union A is a subtype of another union B if for every type t in A, there is another type in B for which t is a subtype.
The subtyping of methods is contravariant for arguments, and covariant for the return type. In simple english, this means that a method m is a subtype of another method n, if all of n’s arguments are subtypes of m’s arguments, and m’s return type is a subtype of n’s return type.
Before getting into the subtype rules for polymorphic objects, we have to first discuss the approach Rtc takes to type parameters. Unlike statically typed languages like C++, Java, or ML, you can have homogenous data structures. For example, in Java, and List<Integer> can ONLY contain integers. However, Ruby has no such restriction. Therefore, the type of an object can change from one line to the next, as the following snippet shows:
my_array = [1] puts my_array.rtc_type # prints "Array<Fixnum>" my_array.push("foo") puts my_array.rtc_type # prints "Array<Fixnum or String>" my_array.delete_at(0) puts my_array.rtc_type # prints "Array<String>"
across these three lines (not counting the puts statement) the type changed twice. To accomodate this dynamic behavior, we say that a type parameter is either dynamic or fixed. This is something of a abuse of terminology, at runtime, type parameters are substituted with type stand-ins, but that is an implementation detail and only confuses the issue. Type parameter here can refer either to the formal type parameter of a polymorphic class, e.g. the t in Array<t>, or a specific instance of that formal type parameter, e.g. the Fixnum in Array<Fixnum> (context will differentiate which meaning is to be used).
A dynamic type parameter displays the above behavior, it’s type changes based on the operations performed on the polymorphic object. By default, all type parameters are dynamic as this more closely models the Ruby type system. However, in some cases you may wish to restrict the type of a type parameter. For example, you may wish to enforce that an array may only contain Fixnum (the actual method for performing this enforcing is detailed below in the section “Constraining Type Parameters”). In this case, the type parameter’s type may no longer change over time, and we say that the type parameter is “fixed”. Whether or not a type parameter is fixed or dynamic affects the subtype rules used.
If the type parameters of a polymorphic class p are dynamic, then the following conditions must be met if it is to be a subtype of another polymorphic class q:
-
p’s actual class must match the actual class of q
-
All the type parameters of p must be a subtype of q’s type parameters
When subtyping some other type against the type parameter itself (for instance, checking that the passed argument is a subtype of t in the type signature “push: (t) -> Array<t>”) then the subtype check always succeeds. If this was not the case, then the following perfectly safe code:
[1,2,3].push("foo")
would be rejected by the type checker.
If the type parameters of a polymorphic class p are fixed, then the following conditions must be met if it is to be a subtype of another polymorphic class q:
-
p’s actual class must match the actual class of q
-
All the type of parameters of q must be exactly q’s type parameters
A type is only a subtype of a type parameter only if it is a subtype of the type the to which the type parameter has been fixed. For example, if you fixed the type parameter of an array to be a Fixnum, then only Fixnums (and subtype of Fixnum) would be considered a subtype of that array’s type parameter.
You can fix the type of a type parameter by using the rtc_annotate method, which is available on all classes that derive from Object. The usage is best explained by an example:
[1,2,3].rtc_annotate("Array<Fixnum>")
This would fix the type parameter of the array to be a Fixnum. If you attempt to fix a type parameter that is narrower than the current type of the type parameter, a Rtc::TypeNarrowingError is thrown. For example, the following would throw the Rtc::TypeNarrowingError:
[1,"2"].rtc_annotate("Array<Fixnum>")
The rtc_annotate function returns the object being annotated, which means you can write code like
foo([1,2].rtc_annotate("Array<Fixnum>"))
Beware aliasing! Rtc will try to infer the most general type for a polymorphic type. This comes into play for nested polymorphic types. For example, in the following:
arr = [[1], [1,"foo"], [1,"foo",1.0]]
arr has the type Array<Array<String or Fixnum or Float>> NOT Array<Array<Fixnum> or Array<Fixnum or String> or …> This can lead to a problem if you write code like this:
arr = [[1].rtc_annotate("Array<Fixnum>"), [1,"foo"], [1, "foo", 1.0]]
and then pass arr to a method that takes a Array<Array<Fixnum or String or Float>> which tries to push a Float onto the first sub-array. Doing so will throw a Rtc::TypeMismatchException within the body of the method, not at the call to the method. Annotating the arr as being of type Array<Array<String or Fixnum or Float>> will not fix the problem either. Smarter handling of aliasing may be added, but for now be careful.