Bartosz Witkowski - Blog.

Typeclasses and show

In this part we’ll start implementing our evaluator. As a first step we’re going to just print out what was parsed.

New concepts

Typeclass

I’ve written about typeclasses in scala before so if you do know scala but don’t know typeclasses you can take a peek.

We won’t cover every nook and cranny of typeclasses but I’ll try to present the general gist of them. Typeclasses are a mechanism for compile time predicate dispatch. That sounds loaded but is actually quite simple - typeclasses allow you to define a set of predicates and functions that can be reused with multiple types - example application: operator overloading (though operator overloading is not implemented that way in mercury).

Concretely, let’s consider this simple typeclass:

:- typeclass show(T) where [
  pred show(T, string),
  mode show(in, out) is det
].

The syntax for typeclasses is the :- token which start almost all “type” related declarations the typeclass keyword followed by a comma separated list of parameters, followed by the where keyword. Surrounded by the [ and ] tokens is a comma departed list of predicates or functions.

A typeclass is a promise that some types may wish to hold. In our example you may want enforce the show promise. The show promise is simple - for some type T we want its string representation.

Let’s implement such a promise for the simples case of all - string. showing a string will unify the string with that string quoted:

:- instance show(string) where [
  (show(String, "\"" +++ String ++ "\""))
]

Here you can see the instance declaration - sometimes called the evidence that string has show (or is/supports show). The other possibility is to create the implementation in a separate predicate e.g show_string and use:

:- instance show(string) where [
  pred(show/2) is show_string.
].

The pred(show/2) is mercury’s way to disambiguate between predicates, and functions and also between functions and predicates of different arity. We will use this notation foo/2 to always mean a foo predicate with 2 arguments - unless explicitly stated it is a function.

We can use the show typeclass by implementing a simple show function (typeclasses, functions, and predicates have a separate namespace so there’s no ambiguity):

:- func show(T) = string <= show(T).
show(T) = String :- show(T, String).

The function declaration can be read as forall T such that T is show define a function from T to string. At this stage it is important to discern between the usage and the declaration - the forall word is key here. This predicate “works” for all types that support show - even those that are not written yet!

If you’re familiar with java or c++ or language that support interfaces the above concept is one of the key to understanding the differences between typeclasses and interfaces. The other big difference is that the there is no “dynamic dispatch if the compiler can’t find the evidence that some type supports a given typeclass it won’t compile. Example:

main(IO_1, IO_Last) :-
  io.write_string(show("1"), IO_1, IO_2),
  io.write_string(show(1), IO_2, IO_Last).

Because there’s no evidence that int supports the show typeclass mercury will fail to compile this class.

Typeclasses and Modules

Public typeclasses (i.e those that should be visible outside of the module that they are declared) should be declared in the interface section.

If you want to include a typeclass instance only a “stub” should be exported i.e:

:- instance show(string).

will export the show instance for string. The actual instance should be implemented in the implementation section.

Matching and Multiple Predicate Rules

Previously we discussed matching in this form:

 ( X = option_1
   ...
 ; X = option_2
   ...
 ; X = option_N
   ...
 )

But there’s also an alternative that is sometimes useful. For example consider this adt:

:- type foo ---> bar ; baz.

And the predicate

:- pred show_foo(foo::in, string::out) is det.
show_foo(Foo, String) :-
  ( Foo = bar
    String = "bar"
  ; Foo = baz
    String = "baz"
  ).

We can also write it using multiple rules this way:

:- pred show_foo(foo::in, string::out) is det.
show_foo(bar, "bar").
show_foo(baz, "baz").

Which at least for me is somewhat nicer. This will only work if the match is complete (you cover all cases) - otherwise the predicate will not be infered as det.

Goal

1) Create the show module:

:- func unwords(list(string)) = string.
:- pred unwords(list(string)::in, string::out) is det. 

2) In the adts module add an instance of show(lisp_val) - to make it easier create a predicate:

:- pred show_lisp_val(lisp_val::in, string::out) is det. 

On:

3) In the main module:

Testing

Check out if this works:

$ ./main a
a
$ ./main '$'
$
$ ./main '(a$ b c . d)'
(a$ b c . d)
$ ./main '"foo"'
"foo"

Implementation

Full code for this part can be found here.

Let’s start from the changes in main module - first we have to add two imports to adts and show, then in the main predicate modify what happens after string.to_char_list(First, CharList):

( parser.top_level_expression(Exp, CharList, Rest) ->
  ( Rest = [] -> 
    io.write_string(show(Exp) ++ "\n", IO_2, IO_Last)
  ; io.write_string("Cannot parse expression!\n", IO_2, IO_Last))
; io.write_string("Cannot parse expression!!\n", IO_2, IO_Last))

Just a couple of changes and nothing fancy going on here - let’s tackle the show module starting with the interface section.

:- module show.

:- interface.

:- import_module string, integer, list.

:- typeclass show(T) where [
  pred show(T, string),
  mode show(in, out) is det
].

:- func unwords(list(string)) = string.
:- pred unwords(list(string)::in, string::out) is det.

:- func show(T) = string <= (show(T)).

:- instance show(integer).
:- instance show(string).

After importing string, integer and list modules we export the show typeclasss declaration, the unwords function and predicate along with instances of show for integer and string types.

In the implementation section we can add instances of show(integer) and show(string) and add the implementation of the show(T) function:

:- implementation.

:- instance show(integer) where [
  (show(Int, String) :- String = integer.to_string(Int))
].

:- instance show(string) where [
  (show(String, "\"" ++ String ++ "\""))
].

show(T) = String :- show(T, String).

The only kind of complicated thing is the unwords predicate let’s break it down:

unwords([], "").

Ok, so this is trivial - a empty list produces an empty string now for an not empty list:

unwords([Head|Tail], String) :-
  unwords_aux(Tail, Head, String).

:- pred unwords_aux(list(string)::in, string::in, string::out) is det.
unwords_aux([], Res, Res).
unwords_aux([Head|Tail], Acc, Res) :-
  unwords_aux(Tail, Acc ++ " " ++ Head, Res).

We use an auxiliary predicate with a accumulator passing style recursive rule where in the iterative step we append the current element to the accumulator.

Note that this may not be terribly efficient - it depends how string is implemented in the worst case (where for example string is backed by a string) this can be very inefficient.

Having done the show module, let’s finish up the changes to the adts module. In the interface section we only add one declaration:

:- instance show(lisp_val).

And in the implementation section we add the instance:

:- instance show(lisp_val) where [
  pred(show/2) is show_lisp_val
].

Let’s go over show_lisp_val part by part:

show_lisp_val(lisp_string(String), show(String)).
show_lisp_val(lisp_number(Integer), show(Integer)).

For lisp_string and lisp_number we just delegate to show(string) and show(integer) respectively, going forwards:

show_lisp_val(lisp_atom(Name), Name).

For lisp_atom we print out the atom name. Now lisp_bool:

show_lisp_val(lisp_bool(yes), "#t").
show_lisp_val(lisp_bool(no),  "#f").

Note how we match both on lisp_val and bool here - we could rewrite it so that matching on the boolean value be an inner match but this is also possible.

Now for the trickier part:

show_lisp_val(lisp_list(List), String) :-
  lisp_val_list_to_string_list(List, String_List),
  unwords(String_List, Single_String),
  String = "(" ++ Single_String ++ ")".

:- pred lisp_val_list_to_string_list(list(lisp_val)::in, list(string)::out) is det.
lisp_val_list_to_string_list(Val_List, String_List) :- 
  lisp_val_list_to_string_list_aux(Val_List, [], String_List).

:- pred lisp_val_list_to_string_list_aux(list(lisp_val)::in, list(string)::in, list(string)::out) is det.
lisp_val_list_to_string_list_aux([], Acc, list.reverse(Acc)).
lisp_val_list_to_string_list_aux([H | T], Acc, String_List) :- 
  show_lisp_val(H, String),
  lisp_val_list_to_string_list_aux(T, [String | Acc], String_List).

Let’s go over lisp_val_list_to_string_list first - again we recursively match on the input list this time a list of lisp_vals - in each iteration step we convert the Head to a string using show_lisp_val and we append the String to the accumulator. In the base case we reverse the list so that it comes “the right way”.

Now show_lisp_val(lisp_list(...)) unwords the list of strings into a single list and adds parentheses to both sides.

show_lisp_val for dotted_list is not much different:

show_lisp_val(lisp_dotted_list(List, Last), String) :-
  lisp_val_list_to_string_list(List, String_List),
  unwords(String_List, List_String),
  show_lisp_val(Last, Last_String),
  String = "(" ++ List_String ++ " . " ++ Last_String ++ ")".

The only thing that differs is that we add the dot . and the last element (also converted to string using show_lisp_val.

In the next part we will learn about higher order predicates and anonymous predicates and functions, so that we can start building our evaluator!

Next