Overview
This assignment gets you acclimated to programming with pattern matching and types. These features are found in such languages as Standard ML, Haskell, Elm, OCaml, Reason, F#, Scala, Idris, Agda, and Coq/Gallina. These languages excel at analysis and implementation of other languages; they are the technology of choice for applications like compilers, verification, and other static analysis, including of security properties. In the next few weeks, you will explore such applications by using Standard ML to implement type systems and lambda calculus.
The assignment has three parts:
 To begin, you will answer some questions about reading.
 On your own, you will write many small functions.
 Possibly working with a partner, you will make a small change to the μScheme interpreter that is written in ML (in Chapter 5).
After completing this assignment, you will be ready to tackle serious programming tasks in Standard ML.
Prelude
Setup
COMP 105 uses two different implementations of Standard ML. For the small problems, we recommend Moscow ML, which is in /usr/sup/bin/mosml
. To start Moscow ML, use
ledit mosml P full I /comp/105/lib
If everything is working correctly, you should see this prompt:
Moscow ML version 2.103 (Tufts University, April 2012)
Enter `quit();' to quit.

If you don’t see the Tufts name, send an immediate email of complaint to staff@cs.tufts.edu
, with a copy to comp105grades@cs.tufts.edu
.
For the large problem, we recommend a nativecode compiler called mlton
(pronounced “Milton”).
ML’s initial basis
As in the hofs
and continuations
assignments, we expect you to use the initial basis, which is properly known as the Standard ML Basis Library. By the standards of popular languages, the basis is small, but it is still much more than you can learn in a week. Fortunately, you only have to learn a few key parts:
 Type constructors
list
,option
,bool
,int
,string
, andorder
 Modules
List
andOption
, includingList.filter
,List.exists
,List.find
, and others  Other module functions
Int.toString
,Int.compare
, andString.compare
 Toplevel functions
o
,print
(for debugging),map
,app
,foldr
,foldl
 Our own
Unit
module, which is not part of the Basis Library but is described in our ML Learning Guide.
The most convenient guide to the basis is the Moscow ML help system; type
 help "";
at the mosml
interactive prompt. The help file is badged incorrectly, but as far as I know, it is up to date.
If you have Jeff Ullman’s book, you need to know that Chapter 9 describes the 1997 basis, which is out of date: today’s compilers use the 2004 basis, which is a standard. But there are only a few differences, primarily in I/O and arrays. The most salient difference is in the interface to TextIO.inputLine
.
Unit testing
Although Standard ML does not have checkexpect
and friends built in, we can do a lot with higherorder functions. Here are some examples of tests written with our Unit
module:
val () =
Unit.checkExpectWith Int.toString "2 means the third"
(fn () => List.nth ([1, 2, 3], 2))
3
val () = (* this test fails *)
Unit.checkExpectWith Bool.toString "2 is false"
(fn () => List.nth ([true, false, true], 2))
false
val () = Unit.reportWhenFailures ()
If you include these tests in your warmup.sml
file, you can run them on the Unix shell command line, using mosmlc
(with a “c”):
$ mosmlc o a.out I /comp/105/lib warmup.sml && ./a.out
In test '2 is false', expected value false but got true
One of two internal Unit tests passed.
$
You’ll use Unit.checkExpectWith
to write your own unit tests. You’ll also use Unit.checkAssert
and Unit.checkExnWith
. The details are in the ML Learning Guide.
Things you need to review before starting
We provide a guide to Learning Standard ML. Learning Standard ML will guide you to other reading.
We also provide a handout on applying the 9step design process to Standard ML. This handout includes the key code excerpts needed to design and program with standard type constructors list
, option
bool
, int
, string
, and order
, as well as the tree
constructor on the homework and the exp
constructor from μScheme.
What we expect from your submission
We expect you will submit code that compiles, has the types given in the assignment, is acceptably styled, is tested, avoids redundant case analysis, and avoids forbidden functions and constructs. Code that does not compile, that has the wrong types, or that uses forbidden functions or constructs will earn No Credit. Code that is untested or has redundant case analysis may earn disappointing grades for structure and organization.
We expect the right types
As always, your code is assessed in part through automated testing. To be testable, each function must not only have the correct name; it must also have the correct type. Your type definitions must also match the type definitions given in the assignment.
To help you check types, I provide a script that analyzes your solution and reports on any type mismatches that it finds. On the command line, run
mlsanitycheck warmup.sml
The “sanity check” reports both missing and illtyped functions. Any function that does not pass the sanity check will earn No Credit. The sanity check looks at the types listed here:
(* first declaration for sanity check *)
val firstVowel : char list > bool
val mynull : 'a list > bool
val reverse : 'a list > 'a list
val minlist : int list > int
exception Mismatch
val zip : 'a list * 'b list > ('a * 'b) list
val ziptoo : 'a list * 'b list > ('a * 'b) list
val pairfoldrEq : ('a * 'b * 'c > 'c) > 'c > 'a list * 'b list > 'c
val concat : 'a list list > 'a list
type 'a env = string > 'a
exception NotFound of string
val emptyEnv : 'a env
val bindVar : string * 'a * 'a env > 'a env
val lookup : string * 'a env > 'a
val isBound : string * 'a env > bool
datatype nat
= ZERO
 TIMES10PLUS of nat * int
val times10plus : nat * int > nat
val intOfNat : nat > int
val natOfInt : int > nat
val natString : nat > string
val carryIntoNat : nat * int > nat
val addWithCarry : nat * nat * int > nat
val addNats : nat * nat > nat
exception Negative
val borrowFromNat : nat * int > nat
val subWithBorrow : nat * nat * int > nat
val subNats : nat * nat > nat
val mulNats : nat * nat > nat
val singletonOf : 'a > 'a flist
val atFinger : 'a flist > 'a
val fingerLeft : 'a flist > 'a flist
val fingerRight : 'a flist > 'a flist
val deleteLeft : 'a flist > 'a flist
val deleteRight : 'a flist > 'a flist
val insertLeft : 'a * 'a flist > 'a flist
val insertRight : 'a * 'a flist > 'a flist
val ffoldl : ('a * 'b > 'b) > 'b > 'a flist > 'b
val ffoldr : ('a * 'b > 'b) > 'b > 'a flist > 'b
(* last declaration for sanity check *)
We don’t promise to have remembered all the functions and their types here. The mlsanitycheck
script will help you, but making sure that every function has the right type is your job, not ours.
We expect wise, wellformatted unit tests
By this time, we expect that you understand the value of unit tests. Grading will focus on your code; except where specifically requested below (naturalnumber arithmetic, freevariable analysis), your unit tests won’t be graded. But we still expect the following:
You will indent all unit tests by eight spaces. This indentation will enable graders to focus on your code.
You will use unit tests wisely. If a function is simple, do take a minute to validate it with a unit test. If a function is not so simple, develop unit tests in the same way you have done for the past three assignments: one unit test per case in the code.
If you need debugging help during office hours, we expect that your code will be accompanied by failing unit tests. (If you cannot get your code to typecheck, we will help you do this without unit tests. But if you need help getting code to produce right answers, we will demand to see your unit tests.)
We expect case analysis only when necessary.
Case analysis is the enemy. All the more so when it is not necessary. Redundant case analysis is a problem in all levels of programming, but as you are learning ML, it is especially easy to fall into. Redundant case analysis typically manifests in one of two ways:
Two cases are present in a
fun
, orcase
, but one is completely subsumed by the other. The most common example is one case to handle the empty list and another case that handles all lists. The emptylist case is often redundant.Example:
fun append ([], ys) = ys  append (xs, ys) = foldr op :: ys xs
In this code, the first case is subsumed by the second. It can be eliminated without changing the meaning of the code, and eliminating it typically improves performance.
A case analysis is performed where no case analysis is needed.
fun sum [] = 0  sum (n :: ns) = foldl op + n ns
These two cases should be replaced by a single case:
fun sum ns = foldl op + 0 ns
We expect you to examine your code carefully and to remove all redundant case analyses.
We don’t expect written algebraic laws
We expect you to continue using a systematic design process, but because ML code is so close to algebraic laws, we don’t expect you to write algebraic laws separately. If you come to office hours, however, we do expect you to be able to talk about algebraic laws and to write them on the board.
We expect an acceptable style
Nobody can learn good style in a week. But you can learn to imitate somebody else’s style, and we expect you to be judicious about what style you imitate. You have access to books by Ullman, Ramsey, and Harper, and to a technical report by Tofte. These sources are not equally good:
Ullman provides the most gentle introduction to ML, and he provides the most information about ML. His book is especially good for programmers whose primary experience is in Clike languages. But, to put it politely, Ullman’s code is not idiomatic. Much of what you see from Ullman should not be imitated.
Ramsey’s code, starting in Chapter 5, is a better guide to what ML should look like. Harper’s code is also very good, and Tofte’s code is reasonable.
On this assignment, we expect you to devote a little effort to good style. Focus on getting your code working first. Then submit it. Then pick up our “Style Guide for Standard ML Programmers”, which contains many examples of good and bad style. Edit your code lightly to conform to the style guide, and submit it again.
In the long run, we expect you to master and follow the guidelines in the style guide.
We expect you to avoid forbidden functions and constructs
While not everybody can learn good style quickly, everybody can learn to avoid the worst faults. In ML, you must avoid these functions and idioms:
Unlike μScheme, Standard ML provides a
length
function in the initial basis. It is banned. The entire assignment must be solved without usinglength
.Solutions that use
length
will earn No Credit.Use function definition by pattern matching. Do not use the functions
null
,mynull
,hd
, andtl
; use patterns instead.Solutions that use
hd
ortl
will earn No Credit.Except for functions given below, do not define auxiliary functions at top level. Use
local
orlet
. You will find it useful to uselocal
to define functions for use in unit tests.Solutions that define auxiliary functions at top level will earn No Credit.
Do not use
open
; if needed, use short abbreviations for common structures. For example, if you want frequent access to theListPair
structure, you can writestructure LP = ListPair
and from there on you can refer to, e.g.,
LP.map
.Solutions that use
open
may earn No Credit for your entire assignment.Unless the problem explicitly says it is OK, do not use any imperative features.
Unless explicitly exempted, solutions that use imperative features will earn No Credit.
Reading comprehension (10%)
These problems will help guide you through the reading. We recommend that you complete them before starting the other problems below. You can download the questions.
Read section 5.1 of Harper about tuple types and tuple patterns. Also look at the list examples in sections 9.1 and 9.2 of Harper.
Now consider the pattern
(x::y::zs, w)
. For each of the following expressions, tell whether the pattern matches the value denoted. If the pattern matches, say what values are bound to the four variablesx
,y
,zs
, andw
. If it does not match, explain why not.([1, 2, 3], ("COMP", 105))
(("COMP", 105), [1, 2, 3])
([("COMP", 105)], (1, 2, 3))
(["COMP", "105"], true)
([true, false], 2.718281828)
Answers here:
You are starting to be ready to use pattern matching.
Read the descriptions of patterns and example values (steps 3 and 2) in the handout “Program Design with ML Types and Pattern Matching.” In the same handout, look at Table 1, including the Types of parts column. Using the ideas you find there, prepare to answer questions about this expression:
case f (x, y, z) of [] => raise Empty  w :: ws => if p w then SOME w else NONE
You are told that the subexpression
f (x, y, z)
has type'a list
. Using that information, give the type of each of these code fragments, which are built from parts of patterns:The type of the pattern
w :: ws
→
The type of the variable
ws
→
The type of the expression
SOME w
→
You are ready to use pattern matching on common ML types.
Read the section on unit testing in the guide to learning ML. Read about infix function names in step 3 of the “design steps” section of the handout “Program Design with ML Types and Pattern Matching”. And read the section on unit testing in this homework.
Now, using the interpreter to be sure your answer is well typed, translate the following failing unit test into ML:
(checkexpect (foldl + 0 '(1 2 3)) 7)
Your translation:
→
You are ready to write unit tests.
In chapter 5 of Programming Languages: Build, Prove, and Compare, the
eval
code for applying a function appears in code chunk 371d. In evaluatingAPPLY (f, args)
, if expressionf
does not evaluate to either a primitive function or a closure, the code raises theRuntimeError
exception.Show a piece of μScheme code that would, when evaluated, cause chunk 371d to raise the
RuntimeError
exception. We are asking for μScheme code that the interpreter runs, not for ML code.→
When exception
RuntimeError
is raised, what happens from the user’s point of view? That is, if a person is interacting with the μScheme interpreter at the read/eval/print loop, and evaluating an expression raises theRuntimeError
exception, what uservisible actions will the interpreter take?
You are ready to write
zip
and to write environment functions that use exceptions.“Free” variables are those that are not bound to a value in the current scope. You can find a longer discussion and precise definition in section 5.11 of Build, Prove, and Compare, which starts on page 382. Read the section and identify the free variables of the expressions below. Remember that a name occurring in function position can be a free variable.
Free variables of
(lambda (x) (lambda (y) (equal? x y)))
→
Free variables of
(lambda (y) (equal? x y))
→
Free variables of
(lambda (s1 s2) (if (or (atom? s1) (atom? s2)) (= s1 s2) (and (equal? (car s1) (car s2)) (equal? (cdr s1) (cdr s2)))))
→
You are ready to improve the μScheme interpreter. You may do this with a partner. You and your partner will turn your answers to parts (a) and (b) into unit tests.
Programming problems to solve individually (75%)
Working on your own, please solve the problems below. Place your solutions in file warmup.sml
. At the start of each problem, please place a short comment, like
(***** Problem A *****)
At the very end of your warmup.sml
, please place the following line:
val () = Unit.reportWhenFailures () (* put me at the _end_ *)
This placement will ensure that if a unit test fails, you are alerted.
To receive credit, your warmup.sml
file must compile and execute in the Moscow ML system, and your functions must have the right types. At minimum, your code must compile without warnings or errors, and it must pass our sanity tests:
% /usr/sup/bin/mosmlc toplevel I /comp/105/lib c warmup.sml
% mlsanitycheck warmup.sml
Defining functions using clauses and patterns
Related Reading for problems A and B: In Learning Standard ML read about Expressions (sections I, II, and III), Data (I, II, and II), Inexhaustive pattern matches, Types (I), Definitions (III, IV), and Expressions (VIII).
A. Define a function mynull : 'a list > bool
, which when applied to a list tells whether the list is empty. Avoid if
, and make sure the function takes constant time. Do not use any functions from the Standard Basis. Make sure your function has the same type as the null
in the Standard Basis.
B. Define a function firstVowel : char list > bool
that takes a list of lowercase letters and returns true
if the first character is a vowel (aeiou) and false
if the first character is not a vowel or if the list is empty. Use the wildcard symbol _
whenever possible, and avoid if
.
Lists
Related Reading for problems C to F: In Learning Standard ML, in addition to the section noted above, read about Types (III), and Exceptions. You will need to understand lists and pattern matching on lists (see Data III). You may also wish to read the section on Curried Functions.
C. Functions foldl
and foldr
are predefined with type
('a * 'b > 'b) > 'b > 'a list > 'b
They are like the μScheme versions except the ML versions are Curried.
Define
reverse : 'a list > 'a list
usingfoldl
orfoldr
.
(In ML, the reverse function is in the initial basis asrev
.)Implement
minlist : int list > int
, which returns the smallest element of a nonempty list of integers. Usefoldl
orfoldr
.If given an empty list of integers, your solution must fail (e.g., by
raise Match
).Your solution should work regardless of the representation of integers: it should not matter how many bits are used to represent a value of type
int
. (Hint: The course solutionmax*
from the hofs homework works regardless of the representation of integers. Perhaps you can steal an idea from it.)
Do not use recursion in either part of this problem.
D. Define a function zip: 'a list * 'b list > ('a * 'b) list
that takes a pair of lists (of equal length) and returns the equivalent list of pairs. If the lengths don’t match, raise the exception Mismatch
, which you must define. Do not use any functions from the Standard Basis Library.
You are welcome to translate a solution from μScheme, but you must either use a clausal definition or write code containing at most one case
expression. Do not use if
.
E. Define a function
val pairfoldrEq : ('a * 'b * 'c > 'c) > 'c > 'a list * 'b list > 'c
that applies a threeargument function to a pair of lists of equal length, using the same order as foldr
. Do not use any functions from the Standard Basis Library. If pairfoldrEq
’s contract is violated (by calling it on lists of unequal lengths), it must raise an exception.
Define a function ziptoo : 'a list * 'b list > ('a * 'b) list
which does exactly the same things as zip
but which uses pairfoldrEq
for its implementation.
F. Define a function
val concat : 'a list list > 'a list
which takes a list of lists of 'a
and produces a single list of 'a
containing all the elements in the correct order. For example,
 concat [[1], [2, 3, 4], [], [5, 6]];
> val it = [1, 2, 3, 4, 5, 6] : int list
Do not use if
. You may use functions from the Standard Basis Library, except for List.concat
—code that uses List.concat
will earn No Credit.
To get full credit for this problem, your function should use no unnecessary cons cells. Keep in mind the cost of appending two lists.
Handling exceptions
Related Reading for problem G: In Learning Standard ML, read the section on Curried functions. Read the sections on Types (III) and Data (IV). Make sure you understand the difference between type abbreviations (type
) and algebraic data types (datatype
)—both are called “types.” Read the section on Exceptions, and make sure you know both how to raise
and how to handle
an exception.
G. Environments with exceptions.
(For this problem, it is OK to use imperative features.)
Define type 'a env
as an abbreviation for a function type, and define exception NotFound
, both as follows:
type 'a env = string > 'a
exception NotFound of string
Now define these functions:
val emptyEnv : 'a env = (* ... *)
val bindVar : string * 'a * 'a env > 'a env = (* ... *)
val lookup : string * 'a env > 'a = (* ... *)
such that you can use 'a env
for a type environment or a value environment. On an attempt to look up an identifier that doesn’t exist, raise the exception NotFound
. Don’t worry about efficiency.
Now define a function
val isBound : string * 'a env > bool
that tells if a name is bound in the environment. Calling isBound
must never raise an exception.
Hint: Review the “sets as characteristic functions” from the hofs
assignment.
Arithmetic by pattern matching on constructed data
Languages like C and C++ enable you to do arithmetic only on as many bits as are in a machine word. More civilized languages allow arithmetic on as many bits as will fit in memory.^{1} Every computer scientist should know how this feature is implemented. In this assignment, we’ll implement arithmetic on natural numbers only, where a natural number is represented as a list of digits.
Our representation is based on the Decimal proof system from the proofsystems handout. To make the rules readable on the web and not just in the printout, I recapitulate them in informal English:
A digit is an integer in the range 0 to 9 inclusive.
Zero is a natural number (rule DecimalZero).
If m is a natural number and d is a digit, then 10 × m + d is a natural number (rule DecimalNat).
I choose to represent a natural number as a data structure defined by an algebraic data type:
datatype nat = ZERO
 TIMES10PLUS of nat * int
(This data structure is equivalent to a list of digits with the leastsignificant digit first.) The meaning of the data structure is determined by an abstraction function and representation invariants:
The abstraction function says that
ZERO
stands for the natural number zero, and if m is a natural number and d is a digit, then TIMES10PLUS (m, d) stands for the natural number m × 10 + d.The representation invariants say that in any value of the form TIMES10PLUS (m, d),
 m and d are not both zero.
 d is a machine integer in the range 0 ≤ d < 10.
The first representation invariant is maintained by the following function, which is called a “smart constructor”:
fun times10plus (ZERO, 0) = ZERO
 times10plus (m, d) = TIMES10PLUS (m, d)
If you like, you can add code to the second case to enforce the invariant on d
.
Here are a couple of useful special cases:
(* times10 : nat > nat *)
fun times10 n = times10plus (n, 0)
(* natOfDigit : int > nat *)
fun natOfDigit d = times10plus (ZERO, d)
Put the definitions of nat, times10plus, and the specialcase functions into your code.
You will define functions for conversion, addition, subtraction. Multiplication is extra credit.
For testing, you will find it useful to convert a list of decimal digits to a natural number. You are welcome to use this code:
fun flip f (x, y) = f (y, x)
(* natOfDigits : int list > nat *)
fun natOfDigits ds = foldl (flip times10plus) ZERO ds
Related Reading:
If needed, review the reading on pattern matching you’ve already done.
In Learning Standard ML, read the section on datatypes—Data IV. Make sure you understand how to pattern match on constructed values.
For detailed descriptions of algorithms for addition and subtraction, with examples, read Programming Languages: Build, Prove, and Compare, section 9.10.2, which starts on page 729. Chapter 9 is not fully included in your abridged edition, but the pages containing this section are there.
To understand how
op
is used in the unittest examples, consult Expressions VII: Infix operators as functions in Learning Standard ML.
H. Naturalnumber conversions.
You will convert between natural numbers, machine integers, and strings.
Define a function
val intOfNat : nat > int
that converts a natural number into a machine integer, or if the natural number is too large, raises
Overflow
. (Use the builtin operators+
and*
, which do machine arithmetic and which automatically raiseOverflow
when needed.)Example:
 intOfNat (natOfDigits [1, 2, 3]); > val it = 123 : int
Write a unit test confirming what the example shows: that
intOfNat (natOfDigits [1, 2, 3])
is 123.Define a function
val natOfInt : int > nat
that converts a nonnegative machine integer into a natural number.
Example:
 natOfInt 2018; > val it = TIMES10PLUS(TIMES10PLUS(TIMES10PLUS(TIMES10PLUS(ZERO, 2), 0), 1), 8) : nat
Use pattern matching, not
if
.A nonnegative machine integer is either zero or it has the form n = 10 × m + d. In the second case, d is
(n mod 10)
and m is(n div 10)
.Write a unit test confirming the
natOfInt
example.Define function
natString
, which converts anat
to a string the way we normally write it (with the most significant digit first).val natString : nat > string
Examples:
 natString (natOfDigits [3, 2, 1]); > val it = "321" : string  natString (natOfDigits [2, 0, 1, 8]); > val it = "2018" : string
Function
natString
must never return an empty string.To earn a passing grade,
natString
must work on 30digit numbers. It is safe to useInt.toString
on a single digit, but if you try to use it on a natural number, the code will fail. INSTRUCTOR: Change the classifier.Hint: Go back to the handout on proof systems for natural numbers. The representation above is based on the Decimal system. But the
natString
function needs to be based on the DecNumeral system. Find a way to deal with the difference.Write a unit test confirming the
natString
example.
My solutions take 7 lines of code and 18 lines of (paranoid) unit tests.
I. Naturalnumber arithmetic.
You will add and subtract natural numbers.
Define function
carryIntoNat : nat * int > nat
. This function takes a natural number n and a carry bit c, and it returns n + c. A carry bit is a machine integer that is either 0 or 1.The function is defined by these algebraic laws:
carryIntoNat (n, 0) == n carryIntoNat (0, c) == c carryIntoNat (10 * m + d, 1) == 10 * carryIntoNat (m, (d + 1) div 10) + ((d + 1) mod 10)
To convert these laws into code, you will need to write the naturalnumber patterns for
0
and for10 * m + d
as constructeddata patternsZERO
andTIMES10PLUS (m, d)
. And you will need to write the naturalnumber arithmetic on the righthand side using the smart constructortimes10plus
.Define function
addWithCarry : nat * nat * int > nat
. This function takes two natural numbers n_{1} and n_{2}, and a carry bit c, and it returns n_{1} + n_{2} + c. To earn a passing grade, it must be capable of adding 30digit numbers, regardless of the number of bits available in a machine integer.The function is defined by these algebraic laws:
addWithCarry (n1, 0, c) = carryIntoNat (n1, c) addWithCarry (0, n2, c) = carryIntoNat (n2, c) addWithCarry (10 * m1 + d1, 10 * m2 + d2, c) = let val d = (d1 + d2 + c) mod 10 val c' = (d1 + d2 + c) div 10 (* the "carry out" *) in 10 * addWithCarry (m1, m2, c') + d end
To convert these laws into code, you will need to write the naturalnumber patterns as constructeddata patterns, and you will need to write the final operation between
in ... end
using the smartconstructor function.Define function
addNats : nat * nat > nat
, as follows:fun addNats (n1, n2) = addWithCarry (n1, n2, 0)
Define function
borrowFromNat : nat * int > nat
. This function takes a natural number n and a borrow bit b, and it returns n − b, provided that n − b is a natural number. If n − b is not a natural number,borrowFromNat
raises the exceptionNegative
, which you will need to define. The borrow bit, like a carry bit, is a machine integer that is either 0 or 1.The function is defined by these algebraic laws:
borrowFromNat (n, 0) == n borrowFromNat (10 * m + 0, 1) == 10 * borrowFromNat (m, 1) + 9 borrowFromNat (10 * m + d, 1) == 10 * m + (d  1), where d > 0
Notice there is no law for the lefthand side
borrowFromNat (0, 1)
. That’s because 0 − 1 is not a natural number—so if your code encounters this case, it should raise theNegative
exception.To convert these laws into code, you will need to write the naturalnumber patterns as constructeddata patterns, and you will need to write some of the arithmetic on the righthand side using the smartconstructor function.
Define function
subWithBorrow : nat * nat * int > nat
. This function takes two natural numbers n_{1} and n_{2}, and a borrow bit b, and if n_{1} − n_{2} − b is a natural number, it returns n_{1} − n_{2} − b. Otherwise it raises theNegative
exception.Like
addWithCarry
,subWithBorrow
must be capable of subtracting 30digit numbers.The function is defined by these algebraic laws:
subWithBorrow (n1, 0, b) = borrowFromNat (n1, b) subWithBorrow (10 * m1 + d1, 10 * m2 + d2, b) = let val d = (d1  d2  b) mod 10 val b' = if d1  d2  b < 0 then 1 else 0 (* the "borrow out" *) in 10 * subWithBorrow (m1, m2, b') + d end
Alert: These laws assume the Standard ML definition of
mod
, which is not what you get from the hardware. The result ofk mod 10
is always nonnegative.To convert these laws into code, you will need to write the naturalnumber patterns as constructeddata patterns, and you will need to write the final operation between
in ... end
using the smartconstructor function.Define function
subNats : nat * nat > nat
, as follows:fun subNats (n1, n2) = subWithBorrow (n1, n2, 0)
Here is a unit test to confirm that subtracting too large a number raises the proper exception: it should raise
Negative
and notMatch
:val () = Unit.checkExnSatisfiesWith natString "1  5" (fn () => subNats (natOfDigits [1], natOfDigits [5])) ("Negative", fn Negative => true  _ => false)
If you trust your conversion functions from the previous problem, you can write unit tests using higherorder functions. Here is an example:
fun opsAgree name intop natop n1 n2 =
Unit.checkExpectWith Int.toString name
(fn () => intOfNat (natop (natOfInt n1, natOfInt n2)))
(intop (n1, n2) handle Overflow => 0)
This function has type
val opsAgree :
string > (int * int > int) > (nat * nat > nat) >
int > int > unit
And it is used as follows
val () = opsAgree "123 + 2018" (op +) addNats 123 2018
val () = opsAgree "2018  123" (op ) subNats 2018 123
val () = opsAgree "2018 * 123" (op * ) mulNats 2018 123
val () = opsAgree "100  1 " (op ) subNats 100 1
(Multiplication is for extra credit.)
My addition functions total 14 lines of code, not counting unit tests. My subtraction functions also total 14 lines of code, not counting unit tests.
Hints:
Exploit the representation invariant. If a natural number matches the pattern
TIMES10PLUS (_, _)
, the representation invariant guarantees that the number is not zero.To maintain the representation invariant, use
TIMES10PLUS
only in pattern matching. On the righthand side of any algebraic law, build natural numbers using the smart constructortimes10plus
.
An immutable, persistent alternative to linked lists
Related Reading for problem J: In Learning Standard ML, read the section on datatypes—Data IV. Make sure you understand how to pattern match on constructed values.
J. For this problem I am asking you to define your own representation of a new abstraction: the list with finger. A list with finger is a nonempty sequence of values, together with a “finger” that points at one position in the sequence. The abstraction provides constanttime insertion and deletion at the finger.
This is a challenge problem. The other problems on the homework all involve old wine in new bottles. To solve this problem, you have to think of something new.
Define a representation for type
'a flist
. (Before you can define a representation, you will want to study the rest of the parts of this problem, plus the test cases.)Document your representation by saying, in a short comment, what sequence is meant by any value of type
'a flist
.Define function
val singletonOf : 'a > 'a flist
which returns a sequence containing a single value, whose finger points at that value.
Define function
val atFinger : 'a flist > 'a
which returns the value that the finger points at.
Define functions
val fingerLeft : 'a flist > 'a flist val fingerRight : 'a flist > 'a flist
Calling
fingerLeft xs
creates a new list that is likexs
, except the finger is moved one position to the left. If the finger belonging toxs
already points to the leftmost position, thenfingerLeft xs
should raise the same exception that the Basis Library raises for array access out of bounds. FunctionfingerRight
is similar. Both functions must run in constant time and space.Please think of these functions as “moving the finger”, but remember no mutation is involved. Instead of changing an existing list, each function creates a new list.
Define functions
val deleteLeft : 'a flist > 'a flist val deleteRight : 'a flist > 'a flist
Calling
deleteLeft xs
creates a new list that is likexs
, except the valuex
to the left of the finger has been removed. If the finger points to the leftmost position, thendeleteLeft
should raise the same exception that the Basis Library raises for array access out of bounds. FunctiondeleteRight
is similar. Both functions must run in constant time and space. As before, no mutation is involved.Define functions
val insertLeft : 'a * 'a flist > 'a flist val insertRight : 'a * 'a flist > 'a flist
Calling
insertLeft (x, xs)
creates a new list that is likexs, except the value
x is inserted to the left of the finger. FunctioninsertRight
is similar. Both functions must run in constant time and space. As before, no mutation is involved. (These functions are related to “cons”.)Define functions
val ffoldl : ('a * 'b > 'b) > 'b > 'a flist > 'b val ffoldr : ('a * 'b > 'b) > 'b > 'a flist > 'b
which do the same thing as
foldl
andfoldr
, but ignore the position of the finger.
Here is a simple test case, which should produce a list containing the numbers 1 through 5 in order. You can use ffoldr
to confirm.
val test = singletonOf 3
val test = insertLeft (1, test)
val test = insertLeft (2, test)
val test = insertRight (4, test)
val test = fingerRight test
val test = insertRight (5, test)
You’ll want to test the delete
functions as well.
Here are some properties you can use for testing:
deleteLeft (insertLeft (x, xs)) == xs
fingerLeft (insertLeft (x, xs)) ==
fingerRight (insertRight (x, fingerLeft xs))
Properties like these can be converted to functions:
val prop1 = fn (x, xs) => deleteLeft (insertLeft (x, xs)) = xs
val prop2 = fn (x, xs) => fingerLeft (insertLeft (x, xs)) =
fingerRight (insertRight (x, fingerLeft xs))
Property prop1
holds of any list with finger. Property prop2
holds only if it is possible to move the finger to the left—otherwise it raises an exception.
Hints: The key is to come up with a good representation for “list with finger.” Once you have a good representation, the code is easy: over half the functions can be implemented in one line each, and no function requires more than two lines of code. The biggest risk is that you copy/paste something incorrectly for left/right, and you can mitigate this risk through testing.
One problem you can do with a partner (15%)
As noted above, MLlike languages excel at analyzing and manipulating programs. In the coming month, you will write many functions that analyze expressions; this problem will get you off to a good start.
The problem is numbered 2 because that’s the problem number in the book. You won’t be doing exercise 1, so you’re not missing anything.
Related Reading for exercise 2: Build, Prove, and Compare, section 5.11, which starts on page 382. Focus on the proof system for judgment y ∈ fv(e); it is provable exactly when freeIn e y
, where freeIn
is the most important function in exercise 2. Also read function eval
in section 5.4. You will modify the case for evaluating LAMBDA
.
2. Improving closures.
When a compiler translates a lambda expression, it doesn’t store the entire environment in the closure; it stores only the free variables of the lambda expression. Interpreters for languages like Lua and Python work in the same way. You’ll implement this code improvement in an interpreter for μScheme.
This problem appears in Build, Prove, and Compare as exercise 2 on page 389, and you’ll solve it in a prelude and four parts:
The prelude is to go to your copy of the book code and copy the file
bare/uschememl/mlscheme.sml
to your working directory. (This code contains all of the interpreter from Chapter 5.) Then make another copy and name itmlschemeimproved.sml
. You will editmlschemeimproved.sml
.The first part is to implement the freevariable predicate
val freeIn : exp > name > bool
.This predicate tells when a variable appears free in an expression. It implements the proof rules in section 5.11 of the book, which starts on page 382.
During this part I recommend that you compile early and often using
/usr/sup/bin/mosmlc c toplevel I /comp/105/lib mlschemeimproved.sml
We also require unit tests for
freeIn
. At minimum, write two tests for each short example in the readingcomprehension questions: one for a variable that is free, and one for a variable that appears in the expression but is not free. Unit tests forLET
forms are recommended but not required.The second part is to write a function that takes a pair consisting of a
LAMBDA
body and an environment, and returns a better pair containing the sameLAMBDA
body paired with an environment that contains only the free variables of theLAMBDA
. (In the book, in exercise 1 starting on page 388, this environment is explained as the restriction of the environment to the free variables.) I recommend that you call this functionimprove
, and that you give it the typeval improve : (name list * exp) * 'a env > (name list * exp) * 'a env
The third part is to use
improve
in the evaluation case forLAMBDA
, which appears in the book on page 371c. You simply applyimprove
to the pair that is already there, so your improved interpreter looks like this:(* more alternatives for [[ev]] for \uscheme 371c *)  ev (LAMBDA (xs, e)) = ( errorIfDups ("formal parameter", xs, "lambda") ; CLOSURE (improve ((xs, e), rho)) )
The fourth and final part is to see if it makes a difference. You will compile both versions of the μScheme interpreter using MLton, which is an optimizing, nativecode compiler for Standard ML. The compiler requires some annoying bureaucracy, but it compensates by providing nativecode speeds.
The original file, which has no unit tests, can be compiled without bureaucracy:
mlton verbose 1 output mlscheme mlscheme.sml
(If plain
mlton
doesn’t work, try/usr/sup/bin/mlton
.)Compiling your improved version requires some bureaucracy to incorporate the
Unit
module.copy105mlfileshere mlton verbose 1 output mlschemeimproved mlschemewithunit.mlb
The file
mlschemewithunit.mlb
tells MLton to compile your code with ourUnit
module. If you wish to do this on your own computer, you will also need filesunit.mlb
andunitmlton.sml
from/comp/105/lib
, and you will have to editmlschemewithunit.mlb
to refer to your local copy ofunit.mlb
, not the one in/comp/105/lib
.
Once compiled, you will run both versions and see if the “improvement” is measurable. For measurement, I have provided a script you can use. I also recommend that you compare the performance of the ML code with the performance of the C code in the course directory.
To get a good measurement, you will need to turn off the “CPU throttling” feature that is built into our interpreters. Use the following arcane Unix commands:
env BPCOPTIONS=nothrottle time runexponentialargmax 22 ./mlscheme
env BPCOPTIONS=nothrottle time runexponentialargmax 22 ./mlschemeimproved
env BPCOPTIONS=nothrottle time runexponentialargmax 22 /comp/105/bin/uscheme
(If you get an error message along the lines of “CPU time exhausted,” something is wrong.)
Hints:
Focus on function
freeIn
. This is the only recursive function and the only function that requires case analysis on expressions. And it is the only function that requires you to understand the concept of free variables. All of these concepts are needed for future assignments.Understanding free variables is hard, but once you understand, the coding is easy.
In Standard ML, the μScheme function
exists?
is calledList.exists
. You’ll have lots of opportunities to use it. If you don’t use it, you’re making extra work for yourself.In addition to
List.exists
, you may find uses formap
,foldr
,foldl
, orList.filter
.You might also find a use for these functions, which are already defined for you:
fun fst (x, y) = x fun snd (x, y) = y fun member (y, []) = false  member (y, z::zs) = y = z orelse member (y, zs)
The case for
LETSTAR
is gnarly, and writing it adds little to the experience. Here are two algebraic laws which may help:freeIn (LETX (LETSTAR, [], e)) y = freeIn e y freeIn (LETX (LETSTAR, b::bs, e)) y = freeIn (LETX (LET, [b], LETX (LETSTAR, bs, e))) y
It’s easier to write
freeIn
if you use nested functions. Mostly the variable y doesn’t change, so you needn’t pass it everywhere. You’ll see the same technique used in theeval
andev
functions in the chapter, as well as the model solution foreval
on thecontinuations
homework.If you can apply what you have learned on the
scheme
andhofs
assignments, you should be able to writeimprove
on one line, without using any explicit recursion.Let the compiler help you: compile early and often.
Once you have the interpreter working, test it by running it on your solutions from the
scheme
,hofs
, orcontinuation
assignments. (It’s possible to getfreeIn
right but to break everything by making a bad mistake inimprove
. Testing on a full solution set will reveal the problem.)
My implementation of freeIn
is 21 lines of ML.
Extra credit
There are three extracredit problems: MULTIPLY, FIVES, and VARARGS.
MULTIPLY. Multiplication of natural numbers.
Define a function
val mulNats : nat * nat > nat
that multiplies two natural numbers. Multiplication obeys these algebraic laws:
0 * n == 0
n * 0 == 0
(10 * m1 + d1) * (10 * m2 + d2) ==
d1 * d2 +
10 * (m1 * d2 + m2 * d1) +
100 * (m1 * m2)
Each of the summands has to be represented as a natural number:
Number
d1 * d2
can be computed using machine multiplication andnatOfInt
.You can multiply
m1 * d2
andm2 * d1
usingnatOfInt
andmulNats
. The recursive call is guaranteed to terminate because at least one argument is getting smaller.You can multiply
m1 * m2
usingmulNats
. The recursive call is guaranteed to terminate because both arguments are getting smaller.
My implementation of mulNats
is seven lines of ML, plus a couple of oneline helper functions.
FIVES. Expressions with arithmetic operators and five 5’s
Consider the class of wellformed arithmetic computations using the numeral 5. These are expressions formed by taking the integer literal 5, the four arithmetic operators +, , *, and /, and properly placed parentheses. Such expressions correspond to binary trees in which the internal nodes are operators and every leaf is a 5. If you enumerate all such expressions, you can answer these questions:
What is the smallest positive integer that cannot be computed by an expression involving exactly five 5’s?
What is the largest prime number that can computed by an expression involving exactly five 5’s?
Exhibit an expression that evaluates to that prime number.
Write an ML function reachable
of type
('a * 'a > order) * ('a * 'a > 'a) list > 'a > int > 'a set
such that reachable (Int.compare, [op +, op , op *, op div]) 5 5
computes the set of all integers computable using the given operators and exactly five 5’s. (You don’t have to bother giving the answers to the questions above; the first two are easily gotten with setFold
.) My solution is under 20 lines of code. Such brevity is possible only because I rely heavily on the setFold
, nullset
, addelt
, and pairfoldrEq
functions defined earlier.
Hints:
 In order to be able to use
Int.compare
interactively, you will either have to runmosml P full
or else tell Moscow ML interactively toload "Int";
Begin your function definition this way:
fun reachable (cmp, operators) five n = (* produce set of expressions reachable with exactly n fives *)
 Use dynamic programming.
 Create a list of length k1 in which element i is a set containing all the integers that can be computed using exactly i elements. Now compute the kth element of the list by combining 1 with k1, 2 with k2, etcetera.
 Try doing the above by passing a list and its reverse, then use
pairfoldrEq
with a suitable function.  The initial list contains a set with exactly one element (in the example above, 5).
Make sure your solution has the completely general type given above, so you could use it with different operations and with different representations of numbers.
VARARGS. Variadic functions in Scheme.
Extend μScheme to support procedures with a variable number of arguments. This is Exercise 8 on page 391 of Build, Prove, and Compare.
Avoid common mistakes
It’s a common mistake to use any of the functions length
, hd
, and tl
. Instant No Credit.
If you redefine a type that is already in the initial basis, code will fail in baffling ways. (If you find yourself baffled, exit the interpreter and restart it.) If you redefine a function at the toplevel loop, this is fine, unless that function captures one of your own functions in its closure.
Example:
fun f x = ... stuff that is broken ...
fun g (y, z) = ... stuff that uses 'f' ...
fun f x = ... new, correct version of 'f' ...
You now have a situation where g is broken, and the resulting error is very hard to detect. Stay out of this situation; instead, load fresh definitions from a file using the use
function.
Never put a semicolon after a definition. I don’t care if Jeff Ullman does it—don’t you do it. It’s wrong! Write a semicolon only when you are deliberately using imperative features.
It’s a common mistake to become very confused by not knowing where you need to use op
. Ullman covers op
in Section 5.4.4, page 165.
It’s a common mistake to include redundant parentheses in your code. To avoid this mistake, use the checklist in the section Expressions VIII (Parentheses) in Learning Standard ML.
It’s a common mistake to do both your pair work and your solo work in the same directory. The submit scripts will balk.
It’s not a common mistake, but it can be devastating: when you’re writing a type variable, be sure to use an ASCII quote mark, as in 'a
, not with a Unicode right quote mark, as in ’a
. As you may remember from class, some text editors, web browsers, or Bluetooth keyboards may use or display Unicode without being asked. Thanks, Apple!
It’s not a common mistake, but do not copy Unit.sml
into your submission directory—you won’t be able to submit.
What to submit and how to submit it
There is no README file for this assignment.
Submitting your individual work
For your individual work, please submit the files cqs.ml.txt
and warmup.sml
. If you have implemented mulNats
, please include it in warmup.sml
. If you have done either of other the extracredit problems, please submit them as varargs.sml
or fives.sml
.
In comments at the top of your warmup.sml
file, please include your name and the names of any collaborators, and a note about any extracredit work you have done.
As soon as you have a warmup.sml
file, run submit105mlsolo
to submit a preliminary version of your work. As you edit your files, keep submitting; we grade only the last submission.
Submitting your improved μScheme interpreter
For your your improved μScheme interpreter, which you may have done with a partner, please submit the file mlschemeimproved.sml
, using the script submit105mlpair
.
How your work will be evaluated
The criteria are mostly the same as for the scheme
and hofs
assignments, but because the language is different, we’ll be looking for indentation and layout as described in the Style Guide for Standard ML Programmers.