Description
Learning Objectives
• Write a non-trivial program in Haskell.
• Understand an implementation of mutable data in a pure functional programming paradigm.
• Use types and functions to add new program structure to a language.
• Implement an instance of a Monad.
• The starter code contains three files, AList.hs, Mutation.hs and MutationUser.hs.
Part 0: Warm-up with association lists
One step above the primitive data types we’ve been studying in Haskell so far is the association list, a poor
man’s dictionary, storing key-value pairs:
type AList a b = [ ( a , b ) ]
In this section, you’ll write some utility functions to practice programming in Haskell, and because you might
find them handy later on. Note that Haskell provides much better ways of storing maps (e.g., via hashing),
but the point of this exercise is to get you writing your own code!
1. Complete each of the following functions inside AList.hs.
(a) lookupA :: Eq a => AList a b -> a -> b: returns the value in the association list corresponding to the given key. Assumes that the key is in the list.
(b) insertA :: Eq a => AList a b -> (a, b) -> AList a b: returns a new association list which
is the old one, except with the new key-value pair inserted. However, it returns the same list if
the key already exists in the list.
(c) updateA :: Eq a => AList a b -> (a, b) -> AList a b: returns a new association list which
is the old one, except with the value corresponding to the given key changed to the given new
value. However, it returns the same list if the key does not appear in the list.
Part 1: Starter code
We have given you starter code in Mutation.hs already. Read through this section to make sure you
understand what the different types defined there represent.
Memory
Our first step is to represent mutable values in Haskell (remember: representing is not the same as actually
being). To do this, we’ll follow a common approach of using a dictionary-like data structure to map names
to values, representing what ”variables” are stored in ”memory.”
First, we create a data type Value to represent the different types of data we’ll store and allow the user to
change.
data Value = In tV al I n t e g e r |
BoolVal Bool |
d e r i v i n g Show
1
Then, we’ll define the Memory data type as simply an association list mapping numeric keys to Values.
type Memory = AList I n t e g e r Value d e r i v i n g Show
You might notice that it’s quite cumbersome to explicitly have different constructors for each Value type,
but this is how we must proceed using what we currently know about Haskell’s type system, since lists must
contain values of the same data type. For ways of achieving heterogeneous lists (lists containing values of
different types) in Haskell, see https://wiki.haskell.org/Heterogenous collections
Pointers
You might think that we’re going to the helpers in Part 0 directly to access and modify this memory. Indeed,
that’s not a bad idea, except we want the extra guarantee that when we access a particular value in memory,
we know what type it’s supposed to be, so that we don’t, for example, accidentally access a Bool and try
to treat it like an Integer. There are many ways of getting around this problem. We’re going to use a one
that draws inspiration from C: typed pointers. Traditionally, a pointer is a memory address, specifying the
location of a particular stored value; in our implementation, it will just be a unique integer.
data P oi n t e r a = P I n t e g e r
This is a little funny – we’ve defined a pointer with a type parameter, but not actually used that parameter
in the constructor of that type. We use this to enable the Haskell compiler to statically check the types of
our pointers in our code.
Our final ingredient to enable static type checking is to make the built-in types Integer and Bool instances
of a type class that can interact with our memory representation. We need to define a new type class, which
supports three actions. Notice that the type signatures ensure that pointers of a given type will only ever
be used to store values of that same type!
c l a s s Mutable a where
g e t : : Memory −> P oi n t e r a −> a
s e t : : Memory −> P oi n t e r a −> a −> Memory
d e f : : Memory −> I n t e g e r −> a −> ( P oi n t e r a , Memory )
The functions are essentially wrappers around the association list functions you implemented earlier, except
that they constrain the types of the values manipulated in the memory. Note that def is a little special, in
that it both returns a new Pointer, as well as new Memory.
2. Inside Mutation.hs, make Integer and Bool instances of the Mutable typeclass by implementing the
three functions for each type (you may need to look up the syntax for making a type an instance of a
type class).
You can use case expressions (http://learnyouahaskell.com/syntax-in-functions#case-expressions)
here to pattern match on the Values stored in the memory. For all erroneous cases (get a key which
doesn’t exist, def on a key which already exists, etc.) your functions should use the error function to
raise a runtime error. Note that your code for Integer and Bool will probably be quite similar, other
than the particular Value constructor you pattern match against.
3. To see whether you are following along, implement the following function inside MutationUser.hs.
You will probably want to use let here.
−− | Takes a number and memory , and s t o r e s two new v al u e s i n memory :
−− − the i n t e g e r ( n + 3 ) a t l o c a t i o n 100
−− − the b o ole an ( n > 0 ) a t l o c a t i o n 500
−− Return the p oi n t e r t o each s t o r e d value , and the new memory .
−− You may assume t h e s e l o c a t i o n s a r e not al r e a d y used by the memory .
p oi n t e r T e s t : : I n t e g e r −> Memory −> ( ( P oi n t e r I n t e g e r , P oi n t e r Bool ) , Memory )
2
> l e t ( ( p1 , p2 ) , mem) = p oi n t e r T e s t 5 [ ]
> g e t mem p1
8
> g e t mem p2
True
Even though you haven’t written much code yet, it’s quite important that you understand what you’re doing
at this stage, as the rest of the assignment builds on this.
Part 2: Chaining
Next, we’re going to modify our type class to allow us to chain together stateful operations. This strategy
exactly follows what we develop in lecture for Stack, so it is recommended that you do not do this part until
we have reached that point in lecture, or you have read the corresponding section in the course notes.
4. In Mutation.hs, introduce a new type. Notice that we’re using not using a type synonym here, so it’s
a little different from StackOp in lecture.
data StateOp a = StateOp (Memory −> ( a , Memory ) )
runOp : : StateOp a −> Memory −> ( a , Memory )
runOp ( StateOp op ) mem = op mem
Change the definitions of get, set, and def so that their types are all written in terms of StateOp.
(Remember currying, and reorder the arguments.)
After making this change, your pointerTest function will no longer compile. Don’t worry, you’ll
return to it soon.
Then, define the chaining operations ”then”, (>>>) :: StateOp a -> StateOp b -> StateOp b,
and ”bind”, (>~>) :: StateOp a -> (a -> StateOp b) -> StateOp b, so that you can chain together invocations of the three basic operations. Once you are done this, you should be able to write
code like:
f : : I n t e g e r −> StateOp Bool
f x =
d e f 1 4 >˜> \p1 −>
d e f 2 True >˜> \p2 −>
s e t p1 ( x + 5 ) >>>
g e t p1 >˜> \y −>
s e t p2 ( y > 3 ) >>>
g e t p2
5. In Mutation.hs, implement returnVal :: a -> StateOp a, which is a function that takes a value,
then creates a new StateOp which doesn’t interact with the memory at all, and instead just returns
the value as the first element in the tuple. Example usage:
g : : I n t e g e r −> StateOp I n t e g e r
g x =
d e f 1 ( x + 4 ) >˜> \p −>
g e t p >˜> \y −>
r e t u r nV al ( x ∗ y )
> runOp ( g 1 0 ) [ ]
( 1 4 0 , [ ( 1 , In tV al 1 4 ) ] )
3
Then in MutationUser.hs, use your work from this section to modify the type annotation and implementation of pointerTest so that it, too, is now a StateOp. After this, you should be able to compile
your program with this function.
Part 3: Calling with references
With our creation of pointers, we have achieved essentially the same semantics as C function calls. We still
pass all arguments by value, but now we have a type of value which really represents a reference to another
value. If we pass such a reference to a function, we can then change the value it points to!
6. To illustrate this idea, implement the following functions in MutationUser.hs:
(a) swap :: Mutable a => Pointer a -> Pointer a -> StateOp (), which takes two pointers and
swaps the values they refer to. Reminder that this is not something we knew how to do otherwise
in either Racket (without mutation) or Haskell.
(b) swapCycle :: Mutable a => [Pointer a] -> StateOp (), which takes a list of pointers p1,
…, pn, with corresponding values v1, …, vn, and sets p1’s value to v2, p2’s value to v3.,
etc., and pn’s value to v1. This function should not change anything if its argument has length
less than 2.
Part 4: Safety Improvements
The implementation of memory and pointers that you have completed in the first parts of this assignment
has a number of safety and memory concerns. Let’s fix two of them.
Allocation
Our current def function specifies allows the user to specify an integer ”address” for the value that will be
stored in memory. This is pretty ridiculous: not only does it allow the user to accidentally try to claim
some memory which is already allocated, the actual memory-access functions get and set operate on pointer
values, not the raw integers specified by the user.
7. In Mutation.hs, write a function alloc :: Mutable a => a -> StateOp (Pointer a), which is similar to def, except that the function automatically generates a fresh (i.e., unused) number to bind in
the value in memory, rather than accepting a number as a parameter.
Deallocation
A part of what we mean when we say that memory lacks scope is that names never go out of scope – once
we use def (or alloc) to create a new storage location in memory, this is a global name. This is particularly
egregious when we realize that the pointers themselves are simply Haskell values, and so are local to whatever
funtion they’re defined in.
8. Define a function free :: Mutable a => Pointer a -> StateOp () which takes a pointer, and removes the corresponding name-value binding from the memory. You should add to AList.hs to do
this.
4
Part 5: Compound mutable values
Now that we are able to mutate booleans and integers, our final task will be mutating a simple compound
data type (parallel to a struct in C). First, copy-and-paste your code from Mutation.hs into a new file,
CompoundMutation.hs. You will do all your work for this part in this file, to allow us to mark your work
from the previous parts separately. Copy-and-paste the following type into CompoundMutation.hs.
−− A type r e p r e s e n t i n g a pe r s on with two a t t r i b u t e s :
−− age and whether they a r e a s t u d e n t o r not .
data Person = Person I n t e g e r Bool d e r i v i n g Show
Your task is to make Person an instance of the Mutable type class, WITHOUT adding an extra constructor
to Value, but instead adding another constructor to Pointer. Think about this as the amount of data stored
at each ”address” in memory being small, and so we can’t store a Person at one location; instead, a Pointer
Person should contain one pointer to each attribute. This has the nice property that we should be able to
access pointers to the attributes of a Person separately. We will leave the implementation open-ended, other
than the requirement that you do not change Value. However, we constrain the interface you must provide:
once you are finished, the following function must compile with your code.
pe r s onTe s t : : Person −> I n t e g e r −> StateOp ( I n t e g e r , Bool , Person )
pe r s onTe s t pe r s on x =
−− not u si n g a l l o c , but we c o ul d
d e f 1 pe r s on >˜> \ p e r s o nP oi n t e r −>
g e t ( p e r s o nP oi n t e r @@ age ) >˜> \ oldAge −>
s e t ( p e r s o nP oi n t e r @@ age ) x >>>
g e t ( p e r s o nP oi n t e r @@ i s S t u d e n t ) >˜> \ s t u −>
g e t ( p e r s o nP oi n t e r @@ age ) >˜> \newAge −>
s e t p e r s o nP oi n t e r ( Person ( 2 ∗ newAge ) ( not s t u ) ) >>>
g e t p e r s o nP oi n t e r >˜> \newPerson −>
g e t ( p e r s o nP oi n t e r @@ i s S t u d e n t ) >˜> \newStu −>
r e t u r nV al ( oldAge , newStu , newPerson )
> f s t ( runOp ( pe r s onTe s t ( Person 2 True ) 1 0 ) [ ] )
( 2 , F al se , Person 20 F al s e )
Hint: (@@), age, and isStudent should be implemented as functions.
5
