Declarations
Declarations
The rest of PureScript AST except for Binder is all here.
First, something not usual
data DeclarationRef
= TypeRef (ProperName 'TypeName) (Maybe [ProperName 'ConstructorName])
| ...
-- From Names.hs
newtype ProperName (a :: ProperNameType)
= ProperName { runProperName :: String }
Ok, WTF is 'TypeName?
It is defined here
data ProperNameType = TypeName | ConstructorName | ClassName | Namespace
Actually, this is a use case of DataKinds extension. The ProperNameType is a kind, and TypeName is kind constructor. So you can put things like Int (which is a type) in the kind constructor, for example
x :: DeclarationRef
x = TypeRef (ProperName "x" :: ProperName TypeName) (Maybe [ProperName "x" :: ProperName ConstructorName])
Basically, it is attaching name with type-level information, so you won't mess the code up.
See http://stackoverflow.com/questions/20558648/what-is-the-datakinds-extension-of-haskell for an excellent explanation.
DeclarationRef
One thing to note is the fact that import and export are essentially symmetrical, so it can be represented uniformly.
findDuplicateRefs: First, I found use of stripPosInfo pretty annoying - maybe there is a way of programming without such concern? Second, why should we find such duplicates? In Sugar/Names/Export and Sugar/Names/Import, you can see it is just used for warning.
Declaration
Interestingly, DataDeclType can be both Data and Newtype. But shouldn't newtype have just one constructor which takes just one argument? Is it a good design to make them orthogonal? My idea: by unifying this, newtype is essentially a syntax sugar which is only known to parser. It becomes easier to do transformations later on with a single type of data. Optimization can be done uniformly.
About NameKind:
data NameKind
-- A private value introduced as an artifact of code generation (class instances, class member
-- accessors, etc.)
= Private
-- A public value for a module member or foreing import declaration
| Public
-- A name for member introduced by foreign import
| External
Note: Guard is also an Expr, but its type will be checked to be Bool.
Note: qualified seems deprecated. It seems that the PureScript team wants to make module first-class, so a unified renaming could be used rather than a extra qualified.
Note: DerivedInstance is not much useful here. Since template PureScript is not ready.
Note: The isFixityDecl etc. look very boilerplate ....
Expr
The NumericLiteral looks not very ideal at least to me ... It is a decision made halfway by PS team and now you have to do casting manually.
Another thing -- Something will be "removed during desugaring", such as ObjectGetter String, so it can be a problem if you are reusing this AST somewhere else.
About TypeClassDictionary
Take a look at http://okmij.org/ftp/Computation/typeclass.html.
So, type class compilation has basically two methods:
- Dictionary pass, or Monomorphization, or type classes as macros
- Intensional type analysis
In JavaScript, the dictionary can be implemented with Object. For every class or constraint, the dictionary will be indexed by type name. Interestingly, here the type is imprecise -- Although author certainly only want ObjectLiteral, but he has to write Expr here.
type Constraint = (Qualified (ProperName 'ClassName), [Type])
Note: Language.PureScript.TypeClassDictionaries has other definitions about type class.
The point is to understand "placeholder" -- a compile time hole to be filled with simple object later. The placeholder contains information necessary to determine how to access the object.
I am not very clear about the process, but it seems like:
SuperClass -> TypeClass -> TypeClassAccessor -> ObjectLiteral
About CaseAlternative
Two kinds:
- Destruction:
case ... of - Guard:
| a == b = c | ...
During the case expansion phase of desugaring, top-level binders will get desugared into case expressions, hence the need for guards and multiple binders per branch here.
data X = A Int | B Int
example :: X -> Int
example x = case x of
A y | y == 1 -> 2
B z | z == 2 -> 1
| otherwise -> 0