Plaintext
Following the Grammar with Haskell
Gary T. Leavens
CS-TR-13-01
January 2013
Keywords: Recursion, programming recursive functions, recursion pattern, inductive definition, BNF
grammar, follow the grammar, functional programming, list recursion, programming languages, concrete
syntax, abstract syntax, helping functions, Haskell.
2001 CR Categories: D.1.1 [Programming Techniques] Applicative (Functional) Programming — de-
sign, theory; D.2.4 [Software Engineering] Coding Tools and Techniques — design, theory; D.3.1 [Program-
ming Languages] Formal Definitions and Theory — syntax; D.3.3 [Programming Languages] Language
Constructs and Features — recursion;
i BY:i
cc Thisdocument is distributed under the terms of the Creative Commons Attribution License, version
2.0, see http://creativecommons.org/licenses/by/2.0/.
Computer Science
University of Central Florida
4000 Central Florida Blvd.
Orlando, FL 32816-2362 USA
Following the Grammar with Haskell
Gary T. Leavens
437D Harris Center (Bldg. 116)
Computer Science, University of Central Florida
4000 Central Florida Blvd., Orlando, FL 32816-2362 USA
leavens@eecs.ucf.edu
January 23, 2013
Abstract
1
This document explains what it means to “follow the grammar” when writing recursive programs, for
several kinds of different grammars. It is intended to be used in classes that teach functional programming
using Haskell, especially those used for teaching principles of programming languages. In such courses
traversal over an abstract syntax tree defined by a grammar are fundamental, since they are the technique
used to write compilers and interpreters.
1 Introduction
An important skill in functional programming is being able to write a program whose structure mimics the
structure of a context-free grammar. This is important for working with programming languages [4, 9], as
they are described by such grammars. Therefore, the proficient programmer’s motto is “follow the grammar.”
This document attempts to explain what “following the grammar” means, by explaining a series of graduated
examples.
The “follow the grammar” idea was the key insight that allowed computer scientists to build compilers for
complex languages, such as Algol 60. It is fundamental to modern syntax-directed compilation [1]. Indeed
the idea of syntax-directed compilation is another expression of the idea “follow the grammar.” It finds
clear expression in the structure of interpreters used in the Friedman, Wand and Haynes book Essentials of
Programming Languages [4].
The idea of following the grammar is not new and not restricted to programming language work. An
early expression of the idea is Michael Jackson’s method [6], which advocated designing a computation
around the structure of the data in a program. Object-oriented design [2, 10] essentially embodies this idea,
as in object-oriented design the program is organized around the data.
Thus the “follow the grammar” idea has both a long history and wide applicability.
1.1 Grammar Background
To explain the concept of following a grammar precisely, we need a bit of background on grammars.
A context-free grammar consists of several nonterminals (names for sets), each of which is defined by
one or more alternative productions. In a context-free grammar, each production may contain recursive uses
of nonterminals. For example, the context-free grammar below has four non-terminals, hStmti, hVari, Exp,
and hStmtListi.
hStmti ::= Skip | Assign hVari hExpi | Compound hStmtListi
hStmtListi ::= [] | hStmti : hStmtListi
hVari ::= hStringi
hExpi ::= hIntegeri
1 This paper is adapted from an Oz verion, [8].
1
The nonterminal hStmti has three alternatives (separated by the vertical bars). The production for hStmti has
a recursive call to hStmtListi, which, in turn, recursively calls hStmti.
Context-free grammars correspond very closely to Haskell data declarations. For example, the data dec-
laration below corresponds directly to the grammar for hStmti above.
data Stmt = Skip | Assign Var Exp | Compound StmtList
type StmtList = [Stmt]
type Var = String
type Exp = Integer
Notice how the built-in Haskell types for lists, Strings, and Integers are used. More importantly, however,
notice that Skip, Assign, and Compound are constructors of the datatype Stmt.
1.2 Definition of Following the Grammar
Following the grammar means making a set of functions whose structure mimics that of the grammar in a
certain way. This is explained in the following definition.
Definition 1.1 Consider a context-free grammar, G and a set of functions Prog. We say that Prog follows
the grammar G if:
1. For every nonterminal, hXi, in G, there is a function, fX, in Prog that takes an argument from the set
described by the nonterminal hXi.
2. If a nonterminal hXi has alternatives, then the corresponding function fX decides between the alterna-
tives offered in hXi’s grammatical productions, and there is (at least) one case in the definition of fX
for each alternative production for hXi.
3. If the nonterminal hXi has an alternative whose production contains a nonterminal hYi that is defined
in the grammar G, then:
(a) the corresponding case in fX has a call to fY, where fY is the function in Prog that handles data
described by the nonterminal hYi, and
(b) each call from fX to fY passes to fY a part of the data fX received as an argument, including at
least that part described by hYi.
Since this definition does not prescribe the details of the set of functions, we often say that Prog has an
outline that follows a grammar G if Prog follows G.
We give many examples in the sections that follow.
Following the grammar thus organizes the program around the structure of the data, in much the same
way as one would organize the methods of an object-oriented program in classes, so that each method deals
with data of the class in which it resides. In a functional program, each kind of data (i.e., each nonterminal)
has its own function that only handles that kind of data. This division of responsibility makes it easy to
understand and modify the program.
A closer object-oriented analogy is to the structure of the methods in the Visitor pattern [5], for a particular
visitor. The story there is the same: each kind of data has a method (the visitor) that handles that particular
kind of data.
1.3 Overview
In the following we will consider several different examples of grammars and functions that follow them.
Our exploration is gradual, and based on increasing complexity in the structures of the grammars we treat.
In Section 2, we show how to follow a grammar that only has alternatives. In Section 3, we do the same for
grammars that only have recursion. In Section 4 we add the complication of multiple nonterminals. Finally,
in Section 5 we treat a series of grammars that combine these features.
2
2 Only Alternatives, No Recursion
The simplest kind of grammar has no recursion, but just has alternatives.
2.1 Temperature Grammar
For example, consider the following grammar for temperatures. In this grammar, all of the alternatives are
base cases.
hTemperaturei ::=
Cold
| Warm
| Hot
In Haskell, this corresponds to the following data definition.
data Temperature = Cold | Warm | Hot
Notice that the nonterminal hTemperaturei is translated into the type named Temperature in Haskell, and
each of the alternatives in this grammar is translated into a constructor (Cold, Warm, and Hot) of the Haskell
data type.
2.1.1 Example
A function that takes a Temperature as an argument will have the outline typified by the following example.
selectOuterwear :: Temperature -> String
selectOuterwear Cold = "down jacket"
selectOuterwear Warm = "wind breaker"
selectOuterwear Hot = "none"
Notice that there are three alternatives in the grammar, and so there are three cases in the function, each of
which corresponds to a condition tested in the body of the function. There is no recursion in the Temperature
grammar, so there is no recursion in the function.
2.1.2 isFreezing Exercise
Which of the following has a correct outline for a function
isFreezing :: Temperature -> Bool
that follows the grammar for Temperature?
1. isFreezing (Cold:_) = True
isFreezing [] = False
isFreezing (_:ts) = isFreezing ts
2. isFreezing [] = False
isFreezing (t:ts) = (t == Cold) || (isFreezing ts)
3. isFreezing Cold = True
isFreezing Hot = False
isFreezing _ = False
4. isFreezing temp =
case temp of
Cold -> isFreezing temp
Hot -> not (isFreezing temp)
_ -> False
Answer: 3. Note that 4 has recursion, which does not follow the grammar in this case, since the grammar is not recursive.
3
2.2 Color Grammar Exercises
Consider another example with simple alternatives and no recursion
hColori ::= Red | Yellow | Green | Blue
and which corresponds to the following Haskell data definition.
data Color = Red | Yellow | Green | Blue deriving (Show)
Write the function:
equalColor :: Color -> Color -> Bool
that takes two Colors and returns True just when they are the same. For example equalColor Red Red
would be True. (Note that you cannot use == in your code, since there is no instance of Eq for this type
available.)
3 Only Recursion, No Alternatives
Another kind of grammar is one that just has recursion, but no alternatives.
3.1 Infinite Sequence Grammar
The following is an example of a grammar with no alternatives, which is a grammar of infinite Integer
sequences.
hISeqi ::= hIntegeri :# hISeqi
The above corresponds to the following Haskell data definition, where :# is used as an infix constructor
data ISeq = Integer :# ISeq
In Haskell one can easily create such infinite sequences using lazy evaluation, as in the following two
examples
ones :: ISeq
ones = 1 :# ones
nats :: ISeq
nats = let from n = n :# (from (n+1))
in (from 0)
3.1.1 Example
A function
iSeqMap :: (Integer -> Integer) -> ISeq -> ISeq
that follows the above grammar is the following.
iSeqMap f (n:#ns) = (f n) :# (iSeqMap f ns)
Following the grammar in this example means that the function does something with the number n in the
head of the pattern and recurses on the tail of the pattern ns, which is an ISeq. In this example, there is no
base case or stopping condition, because the grammar has no alternatives to allow one to stop, which is why
it is so helpful that functions in Haskell are lazy by default.
Although this example does not have a stopping condition, other functions that work on this grammar
might allow stopping when some condition holds, as in the next example.
4
3.1.2 AnyNegative Exercise
Which of the following has a correct outline for a function
anyNegative :: ISeq -> Bool
that follows the grammar for ISeq?
1. anyNegative [] = False
anyNegative (n:ns) = (n<0) || (anyNegative ns)
2. anyNegative iseq =
case iseq of
(n :# ns) -> (n<0) || (anyNegative ns)
3. anyNegative iseq = (n<0) || (anyNegative ns)
4. anyNegative Cold = True
anyNegative _ = False
5. anyNegative (n:#ns) = (n<0) || (anyNegative ns)
Answer: 2 and 5. Note that 3 would be okay if n and ns were defined in its definition, but as it is, it does not type check.
3.1.3 Filter Infinite Sequence Exercise
Write a lazy function
filterISeq :: (Integer -> Bool) -> ISeq -> ISeq
that takes a predicate, pred, an infinite sequence, iseq, and returns an infinite sequence of all elements in
iseq for which pred returns True when applied to the element. Elements retained are left in their original
order. For example:
filterISeq (\n -> n `mod` 2 == 0) nats
would return an ISeq that is 0:#2:#4:#6:#8:# . . ., since that ISeq is only the even elements of nats.
4 Multiple Nonterminals
When the grammar has multiple nonterminals, there should be a function for each nonterminal in the gram-
mar, and the recursive calls between these functions should correspond to the recursive uses of nonterminals
in the grammar. That is, when a production for a nonterminal, hXi, uses another nonterminal, hYi, there
should be a call from the function for hXi to the function for hYi that passes an instance of hYi as an argu-
ment.
4.1 Rectangle Grammar
Consider the following Haskell data definitions for Rectangles and Points. (We dispense with the standard
form of the grammar from now on, since the Haskell data definitions correspond to them very closely.)
data Rectangle = Rect {ul :: Point, lr :: Point} deriving (Show)
data Point = Integer :@ Integer deriving (Show)
5
4.1.1 Example
To follow this grammar when writing a program like
moveUp :: Rectangle -> Integer -> Rectangle
one would structure the code into two functions, one for each of the two types (i.e., nonterminals), as shown
in Figure 1. Since the type (i.e., production) for Rectangle uses the type Point twice, the function
moveUp calls the moveUpPoint function twice, once on each of the points in the Rectangle. Note that the
arguments to these functions are parts of the Integer described by the corresponding nonterminals, and are
extracted from the number by the pattern match in the body of MoveUp
moveUp (Rect {ul = pul, lr = plr}) i =
(Rect {ul = moveUpPoint pul i, lr = moveUpPoint plr i})
moveUpPoint (x :@ y) i = x :@ (y+i)
Figure 1: The two functions that move Rectangles up.
4.1.2 DoubleRect Exercise
Which of the following is a correct outline of a function
doubleRect :: Rectangle -> Rectangle
that follows the grammar for Rectangles?
1. doubleRect (Rect {ul = (xul:@yul), lr = (xlr:@ylr)}) =
Rect {ul = 2*xul:@2*yul, lr = 2*xlr:@2*ylr}
2. doubleRect shape =
case shape of
(Rect {ul = pul, lr = plr}) -> Rect {ul = doubleRect pul,
lr = doubleRect plr}
(x:@y) -> (2*x) :@ (2*y)
3. doubleRect (ul:lr:[]) = doublePoint ul : doublePoint lr : []
doublePoint (x :@ y) = (2*x) :@ (2*y)
4. doubleRect (Rect {ul = pul, lr = plr}) =
(Rect {ul = doublePoint pul, lr = doublePoint plr})
doublePoint (x :@ y) = (2*x) :@ (2*y)
Answer: 4. Note that 2 does not type check in Haskell! Also 1 while it type checks, does not follow the rule about having one function per type, since there is no function for Points given.
4.1.3 ShrinkRect Exercise
Write a function,
shrinkRect :: Rectangle -> Integer -> Rectangle
that takes a Rectangle, rect, and a Number factor (which is greater than or equal to 1) and shrinks the
rectangle’s sides by factor, leaving its upper left point in place.
6
4.2 Multiple Nonterminal Exercise
Suppose you have a grammar with 10 nonterminals, how many functions would be contained in an outline
that followed that grammar?
5 Combination of Different Grammar Types
Most interesting examples of grammars involve a combination of the three types discussed above. That is,
there are alternatives, and recursions, and multiple nonterminals.
The discussion in this section starts with some examples that only involve two of these features. The
grammar for flat lists and the grammar for “window layouts” both involve only one nonterminal. The gram-
mar for boolean expressions that follows has multiple nonterminals, but does not have mutual recursion.
Following these simpler combination examples is a discussion of an example of a “statement and expres-
sion” grammar that involves all three features at once.
5.1 Flat Lists
The grammar for flat lists is simpler than other combinations, as it has alternatives and recursion, but only
one nonterminal. A grammar for flat lists of elements of type t is as follows.
h[t]i ::= [] | hti : h[t]i
where hti is the nonterminal that generates elements of type t. This corresponds to the following Haskell data
definition for its built-in flat lists.
data [] t = [] | t : [t]
The above grammar means we are not interested in the structure of the hti elements of the lists; that is the
sense in which the list is “flat.”
(For flat lists, “following the grammar” is essentially equivalent to following the first and fourth com-
mandments in Friedman and Felleisen’s book The Little Schemer [3].)
5.1.1 Example of Recursion over Flat Lists
The map' function in Figure 2, whose type is shown below,
map' :: (t -> s) -> [t] -> [s]
is a paradigmatic example of following the grammar for flat lists.
map' f [] = []
map' f (x:xs) = (f x) : (map' f xs)
Figure 2: The map’ function for flat lists.
7
5.1.2 ExtractNames Exercise
Which, if any, of the following is a correct outline for a function
extractNames :: [Person] -> [String]
where
data Person = Human {name :: String, address :: String}
deriving (Eq, Show)
that follows the grammar for flat lists? List all that have a correct outline for recursion over flat lists. (Note:
we are mainly asking whether these have the right outline, but having the wrong outline will cause them not
to work as they should.) Note that the function
name :: Person -> String
is defined due to the declaration of the field name in the definition of the type Person.
1. extractNames (p:ps) = (name p) : (extractNames ps)
2. extractNames [] = []
extractNames (p:ps) = (name p) : (extractNames ps)
3. extractNames lst = if lst == []
then []
else (name (head lst)) : extractNames (tail lst)
4. extractNames lst = case lst of
(p:ps) -> (name p) : ps
[] -> []
5. extractNames [] = []
extractNames (Human {name = n, address = _}) -> [n]
6. extractNames = map name
Answer: 2, 3, 6. In 6, the map function is following the grammar on behalf of extractNames.
5.1.3 DeleteListing Exercise
Which, if any, of the following is a correct outline for a function
deleteListing :: String -> [Person] -> [Person]
that follows the grammar for flat lists? List all that have a correct outline for recursion over flat lists.
(Recall that the function name is defined in the definition of the Person type.)
1. deleteListing _ [] = []
deleteListing n (p:ps) = if name p == n
then (deleteListing n ps)
else p:(deleteListing n ps)
2. deleteListing _ [] = []
deleteListing n (p:ps)
| name p == n = (deleteListing n ps)
| otherwise = p:(deleteListing n ps)
8
3. deleteListing n ps = filter ((/= n) . name) ps
4. deleteListing n ps =
case ps of
[] -> []
(per:pers) -> append if n = name per then [] else [per]
(deleteListing n pers)
5. deleteListing n ps =
case ps of
(per:pers) | n == (name per) -> (deleteListing n pers)
_ -> per:(deleteListing n pers)
6. deleteListing n (per:pers) = if name per == n
then (deleteListing n pers)
else per:(deleteListing n pers)
7. deleteListing n ps = [p | p <- ps, name p == n]
Answer: 1, 2, 3, 4, 7. Note that in 1 the function filter is following the grammar, and in 7 the language is following the grammar in the semantics of the list comprehension.
5.2 Window Layouts
For purposes of this paper, a “window layout” is an instance of the data type below. This type is designed to
describe a data structure for placing windows on a computer screen, similar in some ways to layout managers
in various graphical user interface libraries.
The window layout grammar has only one nonterminal, but more alternatives than the grammar for flat
lists. Thus functions that follow its grammar have more recursive calls than functions that follow the grammar
for flat lists.
module WindowLayout where
data WindowLayout = Window {wname :: String, width :: Int, height :: Int}
| Horizontal [WindowLayout]
| Vertical [WindowLayout]
deriving (Show, Eq)
5.2.1 Example
An example function that follows the above grammar is shown in Figure 3. This example is coded using
Haskell’s built-in map function (a version of which was shown in Figure 2 on page 7). Doing that avoids
having to write out a separate function for recursing over a list of WindowLayouts. Writing out a separate
function (or two different functions, in general) to recurse over the two lists in the grammar would still follow
the grammar. These helping function(s) would, of course, follow the grammar for flat lists.
Note, however, that it would not be following the grammar to write something like badDoubleSize
shown in Figure 4 on the next page. The problem with badDoubleSize is that the function works on
both lists and window layouts. Writing code like this leads to confusion, especially as grammars get more
complicated. Sometimes the function also plays a different role depending on what nonterminal it serves, and
so the same argument, if it occurs as part of two different nonterminals can make it very difficult to write the
function correctly. While these are not problems in this simple example, they will cause difficulties in larger
and more complex examples.
In the calls to map in Figure 3, the function argument is doubleSize itself. This is equivalent to
(but slightly more efficient than) passing the function (\ wl -> doubleSize wl). However, in other
problems using such an anonymous function (or a named helping function) may be needed, because the
9
module DoubleSize where
import WindowLayout
doubleSize :: WindowLayout -> WindowLayout
doubleSize (Window {wname = n, width = w, height = h}) =
Window {wname = n, width = 2*w, height = 2*h}
doubleSize (Horizontal wls) = Horizontal (map doubleSize wls)
doubleSize (Vertical wls) = Vertical (map doubleSize wls)
Figure 3: The function doubleSize, which follows the grammar for window layouts.
badDoubleSize (Window {wname = n, width = w, height = h}) =
Window {wname = n, width = 2*w, height = 2*h}
badDoubleSize (Horizontal wls) = Horizontal (badDoubleSize wls)
badDoubleSize (Vertical wls) = Vertical (badDoubleSize wls)
-- the following cases make no sense in this function!
badDoubleSize [] = []
badDoubleSize (w:ws) then (badDoubleSize w):(badDoubleSize ws)
Figure 4: The function badDoubleSize, which does not follow the grammar for window layouts, because
it incorrectly combines two recursions in one function. Indeed, this does not even type check in Haskell!
10
module AddToSize where
import WindowLayout
addToSize n wl =
let addWithFixedN wl = addToSize n wl
in case wl of
(Window {wname = nm, width = w, height = h}) ->
(Window {wname = nm, width = n+w, height = n+h})
(Horizontal wls) -> Horizontal (map addWithFixedN wls)
(Vertical wls) -> Vertical (map addWithFixedN wls)
Figure 5: The function AddToSize that illustrates the use of a helper function when calling map.
function argument to map must be a function that only needs one argument to do its work. For an example,
consider writing a function addToSize, such that {addToSize n wl} returns a window layout that is
exactly like wl, except that each window has n added to both its width and height. In solving such a problem it
is convenient to use a function that takes a window layout, wl and returns the value of {addToSize n wl}
for a fixed n. This is shown in Figure 5.
In Haskell, one can partially apply addToSize, so that one can simplify the second line in Figure 5 to:
let addWithFixedN wl = addToSize n wl
An even shorter version would be to let that line be just the following (by cancelling the wl from both sides
of the above):
let addWithFixedN = addToSize n
Also in Haskell, one can avoid giving a name to this function and just use the expression (addToSize n)
wherever addWithFixedN is used in Figure 5.
5.2.2 MultSize Exercise
Which, if any, of the following is a correct outline for a function
multSize :: Int -> WindowLayout -> WindowLayout
that follows the grammar for window layouts? List all that have a correct outline for recursion over window
layouts.
1. multSize m (Window {wname = nm, width = w, height = h}) =
Window {wname = nm, width = m*w, height = m*h}
multSize m (Horizontal wls) = Horizontal (map (multSize m) wls)
multSize m (Vertical wls) = Vertical (map (multSize m) wls)
2. multSize m wl =
case wl of
(Window {wname = nm, width = w, height = h}) ->
Window {wname = nm, width = m*w, height = m*h}
(Horizontal wls) -> Horizontal (multSize m wls)
(Vertical wls) -> Vertical (multSize m wls)
[] -> []
(w:ws) -> (multSize m w):(multSize m ws)
11
3. multSize m (Window {wname = nm, width = w, height = h}) =
Window {wname = nm, width = m*w, height = m*h}
multSize m (Horizontal wls) = Horizontal (multSizeList m wls)
multSize m (Vertical wls) = Vertical (multSizeList m wls)
multSizeList m [] = []
multSizeList m (w:ws) = (multSize m w):(multSizeList m ws)
4. multSize m wl =
case wl of
(Window {wname = nm, width = w, height = h}) ->
Window {wname = nm, width = m*w, height = m*h}
(Horizontal wls) -> Horizontal (multSizeList m wls)
(Vertical wls) -> Vertical (multSizeList m wls)
where
multSizeList m [] = []
multSizeList m (w:ws) = (multSize m w):(multSizeList m ws)
5. multSize m wl =
case wl of
(Window {wname = nm, width = w, height = h}) ->
Window {wname = nm, width = m*w, height = m*h}
(Horizontal wls) -> Horizontal wls'
(Vertical wls) -> Vertical wls'
where
wls' = (map (multSize m) wls)
6. multSize m wl =
case wl of
(Window {wname = nm, width = w, height = h}) ->
Window {wname = nm, width = m*w, height = m*h}
(Horizontal wls) -> Horizontal (multSize m wls)
(Vertical wls) -> Vertical (multSize m wls)
[] -> []
(w:[]) -> (multSize m w):[]
(w:ws) -> (multSize m w):(multSize m ws)
Answer: 1, 3, 4, 5. It’s important to note that the others do not follow the grammar. The problem is that they combine two recursions in one function, which leads to confusion.
5.2.3 TotalWidth Exercise
Write a function,
totalWidth :: WindowLayout -> Int
that takes a WindowLayout, wl, and returns the total width of the layout. The width is defined by cases. The
width of a WindowLayout of the form
Window { wname = nm, width = w, height = h}
is w. The width of a hWindowLayouti of the form
Horizontal [wl1 , . . ., wlm ]
is the sum of the widths of wl1 through wlm (inclusive). The width of a hWindowLayouti of the form
Vertical [wl1 , . . ., wlm ]
12
is the maximum of the widths of wl1 through wlm (inclusive). If the list is empty, the width should be taken
as 0.
Several example tests that show how totalWidth works are shown in Figure 6. (These use the
eqTest function from the course library’s Testing module. Note that the testing module assumes that
totalWidth is implemented in a module called TotalWidth, which is imported.)
-- $Id: TotalWidthTests.hs,v 1.2 2013/01/24 19:29:19 leavens Exp leavens $
module TotalWidthTests where
import Testing
import WindowLayout
import TotalWidth
main = dotests "TotalWidthTests $Revision: 1.2 $" tests
tests :: [TestCase Int]
tests =
[(eqTest (totalWidth (Window {wname = "olympics", width = 50, height = 33}))
"==" 50)
,(eqTest (totalWidth (Horizontal [])) "==" 0)
,(eqTest (totalWidth (Vertical [])) "==" 0)
,(eqTest (totalWidth
(Horizontal [(Window {wname = "olympics", width = 80, height = 33})
,(Window {wname = "news", width = 20, height = 10})]))
"==" 100)
,(eqTest (totalWidth
(Vertical [(Window {wname = "olympics", width = 80, height = 33})
,(Window {wname = "news", width = 20, height = 10})]))
"==" 80)
,(eqTest (totalWidth
(Vertical [(Window {wname = "Star Trek", width = 40, height = 100})
,(Window {wname = "olympics", width = 80, height = 33})
,(Window {wname = "news", width = 20, height = 10})]))
"==" 80)
,(eqTest (totalWidth
(Horizontal
[(Vertical [(Window {wname = "Tempest", width = 200, height = 100})
,(Window {wname = "Othello", width = 200, height = 77})
,(Window {wname = "Hamlet", width = 1000, height = 600})])
,(Horizontal [(Window {wname = "baseball", width = 50, height = 40})
,(Window {wname = "track", width = 100, height = 60})
,(Window {wname = "golf", width = 70, height = 30})])
,(Vertical [(Window {wname = "Star Trek", width = 40, height = 100})
,(Window {wname = "olympics", width = 80, height = 33})
,(Window {wname = "news", width = 20, height = 10})])
]))
"==" 1300)
]
Figure 6: Tests for the totalWidth exercise.
Feel free to use Haskell’s map and max functions (as well as foldl or foldr if you have seen those).
13
module SalesData where
data SalesData =
Store { address :: String, amounts :: [Int] }
| Group { gname :: String, members :: [SalesData] }
deriving (Show, Eq)
Figure 7: Data declaration for the type SalesData.
5.2.4 Design Your own WindowLayout Problem Exercise
Design another problem for the type WindowLayout. Give an English explanation, the function’s type, and
some examples. Then solve your problem, first on paper, then on the computer.
For example, you might do something like computing the total area of the window layout, or a list of all
the names of the windows.
5.3 Sales Data
The grammar for SalesData is shown in Figure 7. It has a single nonterminal and like the window layouts
grammar, also uses lists. The grammar for hStringi is the same as for Haskell, which is essentially [Char].
5.3.1 NormalizeSalesData Example
The grammar for sales data is interesting in that its fields contain several different kinds of lists. Since the
different lists play different roles in the sales data grammar, it is thus especially important to follow the
grammar in the sense that only data of type SalesData should be passed to functions that work on that
type, and no lists should be passed to such functions.
The reason this is important is illustrated by the following example. Suppose we want to write a function
normalizeSalesData :: SalesData -> SalesData
that takes a sales data argument, sd, and returns a result that is just like sd, except that in each store record,
each address string is put into ALL CAPITAL LETTERS and the amounts list is trimmed to be just the first
5 elements of the argument’s list, and in each group record, the name field is put into all capital letters, and
each of the members is also normalized.
Figure 8 on the following page gives some examples of how this program is supposed to work. These use
the Test procedure from the course library.
We suggest that you try to write out a solution for this problem before looking at our solution.
Correct code for NormalizeSalesData is shown in Figure 9 on page 16. This follows the grammar
in that it only passes sales data to the function NormalizeSalesData. Note that lists of characters and
numbers are handled by helper functions capitalize, map, and the built-in function take. Can you see
why this code works correctly?
Now consider the NormalizeSalesDataBad, shown in Figure 10 on page 16, which attempts to
solve the same problem. This does not follow the grammar, but resembles programs that some students try to
write because they don’t want to use helper functions. Although it still correctly uses take, all other lists are
handled by making all recursive calls to itself. These recursive calls do not follow the grammar, because they
pass lists to a function that is supposed to receive Sales Data records. Does the code in Figure 10 on page 16
even type check? No, it does not, and there are several type errors that Haskell will point out to try to steer
you back to following the grammar.
Indeed the code in Figure 10 on page 16 would not be correct, even if Haskell would execute it! Can
you see why? First, remember that in Haskell, Strings are lists of Chars. The last two cases in the function
definition are attempting to deal with lists of characters separately from lists of sales data records, but even if
14
module NormalizeSalesDataTests where
import Testing
import SalesData
import NormalizeSalesData
main = dotests "$Revision: 1.1 $" tests
tests =
[(eqTest (normalizeSalesData
(Store {address = "1005 Alafaya Trail Rd."
,amounts = [101, 102, 103, 104, 105, 106]}))
"==" (Store {address = "1005 ALAFAYA TRAIL RD."
,amounts = [101, 102, 103, 104, 105]}))
,(eqTest (normalizeSalesData
(Store {address = "227 International Dr."
,amounts = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]}))
"==" (Store {address = "227 INTERNATIONAL DR."
,amounts = [1, 2, 3, 4, 5]}))
,(eqTest (normalizeSalesData
(Group {gname = "Target"
,members = [(Store {address = "227 International Dr."
,amounts=[1,2,3,4,5,6,7,8,9,10]})
,(Store {address = "1005 Alafaya Trail Rd."
,amounts=[101,102,103,104,105,106]})]}))
"==" (Group {gname = "TARGET"
,members = [(Store {address = "227 INTERNATIONAL DR."
,amounts=[1,2,3,4,5]})
,(Store {address = "1005 ALAFAYA TRAIL RD."
,amounts=[101,102,103,104,105]})]}))
,(eqTest (normalizeSalesData
(Group {gname = "Target"
,members =
[(Group {gname = "Target Florida"
,members = [(Store {address="227 International Dr."
,amounts=[1,2,3,4,5,6,7,8,9,10]})
,(Store {address="1005 Alafaya Trail Rd."
,amounts=[101,102,103,104,105,106]})]})
,(Group {gname = "Target Iowa"
,members = [(Store {address="1024 Binary Ave."
,amounts=[1,2,4,8]})
,(Store {address="256 Logical Blvd."
,amounts=[7,25,15,44,36,8,3,7]})]})
,(Group {gname = "Target New York"
,members =
[(Group {gname = "Target NYC"
,members = [(Store {address = "Fifth Ave."
,amounts= [8,2,1,3,3,5]})
,(Store {address= "Albany"
,amounts = []})]})]})]})))
"==" (Group {gname = "TARGET"
,members=
[(Group {gname = "TARGET FLORIDA"
,members = [(Store {address = "227 INTERNATIONAL DR."
,amounts=[1,2,3,4,5]})
,(Store {address = "1005 ALAFAYA TRAIL RD."
,amounts=[101,102,103,104,105]})]})
,(Group {gname = "TARGET IOWA"
,members = [(Store {address = "1024 BINARY AVE."
,amounts=[1,2,4,8]})
,(Store {address = "256 LOGICAL BLVD."
,amounts=[7,25,15,44,36]})]})
,(Group {gname = "TARGET NEW YORK"
,members= [(Group {gname = "TARGET NYC"
15,members = [(Store {address= "FIFTH AVE."
,amounts=[8,2,1,3,3]})
,(Store {address="ALBANY"
,amounts=[]})]})]})]})]
module NormalizeSalesData (normalizeSalesData) where
import SalesData
import Data.Char (toUpper)
normalizeSalesData :: SalesData -> SalesData
normalizeSalesData (Store {address = addr, amounts = amts}) =
(Store {address = (capitalize addr), amounts = take 5 amts})
normalizeSalesData (Group {gname = name, members = membs}) =
(Group {gname = (capitalize name), members = map normalizeSalesData membs})
capitalize str = map toUpper str
Figure 9: The function normalizeSalesData, which follows the Sales Data grammar.
module NormalizeSalesData (normalizeSalesDataBad) where
import SalesData
import Data.Char (toUpper)
normalizeSalesDataBad :: SalesData -> SalesData
normalizeSalesDataBad (Store {address = addr, amounts = amts}) =
(Store {address = (normalizeSalesDataBad addr), amounts = take 5 amts})
normalizeSalesDataBad (Group {gname = name, members = membs}) =
(Group {gname = (normalizeSalesDataBad name)
, members = normalizeSalesDataBad membs})
-- The following code should not be in this function!
normalizeSalesDataBad [] = []
normalizeSalesDataBad (sd:sds) = (normalizeSalesDataBad sd)
: (normalizeSalesDataBad sds)
normalizeSalesDataBad "" = ""
normalizeSalesDataBad (c:cs) = (toUpper c) : (normalizeSalesDataBad cs)
Figure 10: The function NormalizeSalesDataBad, which does not follow the Sales Data grammar.
this type checked, the pattern matching would never get to these cases. Can you see how the pattern matching
is confused? Is the problem solved by reordering the last 4 cases? Why not?
5.4 Boolean Expressions
The grammar for Boolean expressions is shown in Figure 11 on the next page. It has multiple nonterminals,
but does not have mutual recursion among the nonterminals. The grammar type String is Haskell’s built-in
String type. These Strings represent variable identifiers in Boolean expressions.
5.4.1 Example
An example using this grammar is shown in Figure 12.
5.4.2 Beval Exercise
Write a function
bEval :: (Eq a) => Bexp -> (String -> a) -> Bool
16
module Bexp where
data Bexp = And Bexp Bexp | Or Bexp Bexp | Not Bexp | Comp Comparison
deriving (Eq, Show)
data Comparison = Equals String String | NotEquals String String
deriving (Eq, Show)
Figure 11: Grammar for Boolean expressions.
module NegateBexp where
import Bexp
negateBexp :: Bexp -> Bexp
negateBexp (And l r) = (Or (negateBexp l) (negateBexp r))
negateBexp (Or l r) = (And (negateBexp l) (negateBexp r))
negateBexp (Not e) = e
negateBexp (Comp c) = (Comp (negateComparison c))
negateComparison :: Comparison -> Comparison
negateComparison (Equals v1 v2) = (NotEquals v1 v2)
negateComparison (NotEquals v1 v2) = (Equals v1 v2)
Figure 12: The function negateBexp, which follows the Boolean expression grammar.
that takes 2 arguments: a Bexp, e, and a function f from atoms to values of some Eq type. Assume that f is
defined on each String literal that occurs in e. The function bEval should evaluate the expression e, using f
to determine the values of all Strings that occur within it. The bEval function should: conjoin the values that
result from the evaluation of the subexpressions in an And expression, disjoin the values of subexpressions
in an Or negate the value of the subexpression in a Not, and use the value of the comparison contained in
a Comp expression. For the Comparison expressions, the value of Equals v w should be true just when
f v == f w, and the value of NotEquals v w should be true just when f v /= f w. Examples are shown in
Figure 13 on the following page.
5.5 Statements and Expressions
The grammar for statements and expressions below involves mutual recursion. Statements can contain ex-
pressions and expressions can contain statements. This mutual recursion allows for arbitrary nesting. The
type definition is shown in Figure 14 on page 19.
5.5.1 Examples
An example that follows this grammar is shown in Figure 15. It is instructive to draw arrows from each use
of StmtAdd1 and ExpAdd1, in the figure, to where these names are defined, and to draw arrows from the
uses of the corresponding nonterminals in the grammar to where they are defined. That helps show how the
recursion patterns match.
Note that, in Figure 15 the recursive call from expAdd1 to StmtAdd1 occurs in the case for beginExp,
following the grammar. However, this recursive call is actually made by map, since the recursion in the gram-
mar is inside a list. It would be fine to use a separate helping function for this list recursion, and that will be
necessary in cases that map cannot handle.
17
-- $Id: BEvalTests.hs,v 1.1 2013/01/25 01:24:45 leavens Exp leavens $
module NegateBexpTests where
import Testing
import Bexp
import BEval
main = dotests "BEvalTests $Revision: 1.1 $" tests
-- This function is just for ease of testing
stdEnv v =
case v of
"p" -> 1
"q" -> 2
"r" -> 4020
"x" -> 76
"y" -> 0
_ -> undefined
tests :: [TestCase Bool]
tests =
[(assertTrue (bEval (Comp (Equals "q" "q")) stdEnv))
,(assertFalse (bEval (Comp (NotEquals "q" "q")) stdEnv))
,(assertFalse (bEval (Comp (Equals "q" "r")) stdEnv))
,(assertTrue (bEval (Comp (NotEquals "p" "q")) stdEnv))
,(assertTrue (bEval (Or (Comp (NotEquals "p" "q"))
(Comp (Equals "x" "x"))) stdEnv))
,(assertFalse (bEval (And (Comp (NotEquals "p" "q"))
(Comp (NotEquals "x" "x"))) stdEnv))
,(assertFalse (bEval (And (Not (Comp (Equals "p" "p")))
(Comp (Equals "x" "x"))) stdEnv))
,(assertTrue (bEval (Not (And (Not (Comp (Equals "p" "p")))
(Comp (Equals "x" "x")))) stdEnv))
,(assertTrue (bEval (Or (Not (And (Not (Comp (Equals "p" "p")))
(Comp (Equals "x" "x"))))
(Or (Comp (Equals "p" "q"))
(Comp (Equals "x" "x")))) stdEnv))
,(assertFalse (bEval (Or
(And (Not (Comp (Equals "p" "p")))
(Comp (Equals "x" "x")))
(Or (Comp (Equals "p" "q"))
(Comp (Equals "x" "y")))) stdEnv))
]
Figure 13: Testing for BEval.
18
-- $Id: StatementsExpressions.hs,v 1.1 2013/01/25 13:44:12 leavens Exp leavens $
module StatementsExpressions where
data Statement = ExpStmt Expression
| AssignStmt String Expression
| IfStmt Expression Statement
deriving (Eq, Show)
data Expression = VarExp String
| NumExp Integer
| EqualsExp Expression Expression
| BeginExp [Statement] Expression
deriving (Eq, Show)
Figure 14: The Statements and Expressions grammar.
-- $Id$
module StmtAdd1 where
import StatementsExpressions
stmtAdd1 :: Statement -> Statement
stmtAdd1 (ExpStmt e) = (ExpStmt (expAdd1 e))
stmtAdd1 (AssignStmt v e) = (AssignStmt v (expAdd1 e))
stmtAdd1 (IfStmt e s) = (IfStmt (expAdd1 e) (stmtAdd1 s))
expAdd1 :: Expression -> Expression
expAdd1 (VarExp v) = (VarExp v)
expAdd1 (NumExp i) = (NumExp (i+1))
expAdd1 (EqualsExp e1 e2) = (EqualsExp (expAdd1 e1) (expAdd1 e2))
expAdd1 (BeginExp stmts e) = (BeginExp (map stmtAdd1 stmts) (expAdd1 e))
Figure 15: The functions StmtAdd1 and ExpAdd1. These mutually-recursive functions work on the state-
ment and expression grammar.
19
5.5.2 SubstIdentifier Exercise
Write a function
substIdentifier :: Statement -> String -> String -> Statement
that takes a statement stmt and two Strings, new and old, and returns a statement that is just like stmt,
except that all occurrences of old in stmt are replaced by new. Examples are shown in Figure 16 on the
following page.
5.5.3 More Statement and Expression Exercises
Invent one or more other problem on the statement and expression grammar above. For example, try reversing
the list of statements in every BeginExp whose expression contains a hNumExpi with a number less than
3. For a more challenging exercise, write an evaluator for this grammar.
Acknowledgments
Thanks to Faraz Hussain for corrections to the Oz version of this paper [8], and to Brian Patterson and Daniel
Patanroi for comments on drafts of a Scheme version of this paper [7]. Earlier versions were supported in
part by NSF grants CCF-0428078, CCF-0429567, CNS 07-09217, and CCF-08-16350. This version was
supported in part by NSF grants CCF-0916715, 0916350, 1017262, 1017334, and 1228695.
References
[1] R. Sethi A. V. Aho and J. D. Ullman. Compilers. Principles, Techniques and Tools. Addison-Wesley,
1986.
[2] Graham M. Birtwistle, Ole-Johan Dahl, Bjorn Myhrhaug, and Kristen Nygaard. SIMULA Begin. Auer-
bach Publishers, Philadelphia, Penn., 1973.
[3] Daniel P. Friedman and Matthias Felleisen. The Little Schemer. MIT Press, fourth edition, 1996.
[4] Daniel P. Friedman, Mitchell Wand, and Christopher T. Haynes. Essentials of Programming Languages.
The MIT Press, New York, NY, second edition, 2001.
[5] Erich Gamma, Richard Helm, Ralph Johnson, and John Vlissides. Design Patterns: Elements of
Reusable Object-Oriented Software. Addison-Wesley, Reading, Mass., 1995.
[6] Michael A. Jackson. Principles of Program Design. Academic Press, London, 1975.
[7] Gary T. Leavens. Following the grammar. Technical Report 05-02a, Department of Computer Sci-
ence, Iowa State University, Ames, Iowa, 50011, January 2006. Available by anonymous ftp from
ftp.cs.iastate.edu.
[8] Gary T. Leavens. Following the grammar. Technical Report CS-TR-07-10b, School of EECS, University
of Central Florida, Orlando, FL, 32816-2362, November 2007.
[9] Peter Van Roy and Seif Haridi. Concepts, Techniques, and Models of Computer Programming. The
MIT Press, Cambridge, Mass., 2004.
[10] Rebecca Wirfs-Brock, Brian Wilkerson, and Lauren Wiener. Designing Object-Oriented Software.
Prentice-Hall, Englewood Cliffs, NJ 07632, 1990.
20
module SubstIdentifierTests where
import StatementsExpressions
import SubstIdentifier
import Testing
main = dotests "BEvalTests $Revision$" tests
tests :: [TestCase Statement]
tests =
[(eqTest (substIdentifier (ExpStmt (NumExp 7)) "x" "y")
"==" (ExpStmt (NumExp 7)))
,(eqTest (substIdentifier (ExpStmt (VarExp "y")) "x" "y")
"==" (ExpStmt (VarExp "x")))
,(eqTest (substIdentifier
(ExpStmt (BeginExp [(AssignStmt "x" (NumExp 7))]
(EqualsExp (VarExp "x") (VarExp "q"))))
"z" "x")
"==" (ExpStmt (BeginExp [(AssignStmt "z" (NumExp 7))]
(EqualsExp (VarExp "z") (VarExp "q")))))
,(eqTest (substIdentifier
(ExpStmt (BeginExp [] (EqualsExp (VarExp "z") (VarExp "z"))))
"a" "z")
"==" (ExpStmt (BeginExp [] (EqualsExp (VarExp "a") (VarExp "a")))))
,(eqTest (substIdentifier (ExpStmt (VarExp "a")) "a" "z")
"==" (ExpStmt (VarExp "a")))
,(eqTest (substIdentifier (AssignStmt "y" (VarExp "m")) "q" "y")
"==" (AssignStmt "q" (VarExp "m")))
,(eqTest (substIdentifier (IfStmt (EqualsExp (VarExp "y") (NumExp 10))
(AssignStmt "y" (NumExp 0)))
"w" "y")
"==" (IfStmt (EqualsExp (VarExp "w") (NumExp 10))
(AssignStmt "w" (NumExp 0))))
,(eqTest (substIdentifier (IfStmt
(BeginExp [(AssignStmt "x" (NumExp 3))
,(AssignStmt "y" (VarExp "q"))
,(AssignStmt "b"
(EqualsExp (VarExp "x") (NumExp 2)))]
(EqualsExp (VarExp "b") (NumExp 0)))
(AssignStmt "q" (NumExp 17)))
"z" "x")
"==" (IfStmt
(BeginExp [(AssignStmt "z" (NumExp 3))
,(AssignStmt "y" (VarExp "q"))
,(AssignStmt "b" (EqualsExp (VarExp "z") (NumExp 2)))]
(EqualsExp (VarExp "b") (NumExp 0)))
(AssignStmt "q" (NumExp 17))))
]
Figure 16: Tests for the function SubstIdentifier.
21