UNIT-II part 2

Expressions and Statements

Arithmetic Expressions
Their evaluation was one of the motivations for the development of the first
programming languages

- Arithmetic expressions consist of operators, operands, parentheses, and function

calls

Design issues for arithmetic expressions:

1. What are the operator precedence rules?

2. What are the operator associativity rules?

3. What is the order of operand evaluation?

4. Are there restrictions on operand evaluation

side effects?

5. Does the language allow user-defined operator overloading?

6. What mode mixing is allowed in expressions?

2
 A unary operator has one operand

A binary operator has two operands

A ternary operator has three operands

Def: The operator precedence rules for expression evaluation define the
order in which “adjacent” operators of different precedence levels are
evaluated (“adjacent” means they are separated by at most one operand)

- Typical precedence levels

1. parentheses
2. unary operators
3. ** (if the language supports it)
4. *, /
5. +, -
Def: The operator associativity rules for expression evaluation define the
order in which adjacent operators with the same precedence level are
evaluated

3
- Typical associativity rules:
- Left to right, except **, which is right to left
- Sometimes unary operators associate right to left (e.g., FORTRAN)

- APL is different; all operators have equal precedence and all operators
associate right to left

Precedence and associativity rules can be overriden with parentheses

Operand evaluation order

- The process:
1. Variables: just fetch the value
2. Constants: sometimes a fetch from memory; sometimes the constant is
in the machine language instruction
3. Parenthesized expressions: evaluate all operands and operators first
4. Function references: The case of most interest!
- Order of evaluation is crucial
Functional side effects - when a function changes a two-way parameter or a
nonlocal variable

4
The problem with functional side effects:
- When a function referenced in an expression alters another operand of the
expression
e.g., for a parameter change:
a = 10;
b = a + fun(&a);
/* Assume that fun changes its parameter */

Two Possible Solutions to the Problem:

1. Write the language definition to disallow functional side effects
- No two-way parameters in functions
- No nonlocal references in functions
- Advantage: it works!
- Disadvantage: Programmers want the flexibility of two-way parameters
(what about C?) and nonlocal references

2. Write the language definition to demand that operand evaluation order

be fixed
- Disadvantage: limits some compiler
optimizations
5
Conditional Expressions

- C, C++, and Java (?:)

e.g. average = (count == 0)? 0 : sum / count;

Operator Overloading
- Some is common (e.g., + for int and float)
- Some is potential trouble (e.g., * in C and C++)
- Loss of compiler error detection (omission of an operand should be a
detectable error)
- Can be avoided by introduction of new symbols (e.g., Pascal’s div)

- C++ and Ada allow user-defined overloaded operators

Potential problems:

- Users can define nonsense operations

- Readability may suffer

6
Implicit Type Conversions

Def: A narrowing conversion is one that converts an object to a type that

cannot include all of the values of the original type

Def: A widening conversion is one in which an object is converted to a type

that can include at least approximations to all of the values of the original
type

Def: A mixed-mode expression is one that has operands of different types

Def: A coercion is an implicit type conversion

- The disadvantage of coercions:

- They decrease in the type error detection ability of the compiler

- In most languages, all numeric types are coerced in expressions, using
widening conversions
- In Modula-2 and Ada, there are virtually no coercions in expressions

7
Explicit Type Conversions

- Often called casts

e.g.
Ada:
FLOAT(INDEX) -- INDEX is INTEGER type
C:
(int)speed /* speed is float type */

Errors in Expressions
- Caused by:
- Inherent limitations of arithmetic e.g. division by zero
- Limitations of computer arithmetic e.g. overflow
- Such errors are often ignored by the run-time system

8
 Relational Expressions

- Use relational operators and operands of various types

- Evaluate to some boolean representation
- Operator symbols used vary somewhat among languages (!=, /=, .NE., <>,
#)

Boolean Expressions
- Operands are boolean and the result is boolean
- Operators:
FORTRAN 77 FORTRAN 90 C Ada
.AND. and && and
.OR. or || or
.NOT. not ! not
xor
- C has no boolean type--it uses int type with 0 for false and nonzero for
true
- One odd characteristic of C’s expressions: a < b < c is a legal
expression, but the result is not what you might expect

9
Precedence of All Operators:

Pascal: not, unary -, *, /, div, mod, and +, -, or relops

Ada: **, *, /, mod, rem, unary -, not, +, -, &, relops, and, or, xor

C, C++, and Java have over 50 operators and 17 different levels of

precedence

Short Circuit Evaluation

Pascal: does not use short-circuit evaluation
Problem: table look-up
index := 1;
while (index <= length) and
(LIST[index] <> value) do
index := index + 1

10
 C, C++, and Java: use short-circuit evaluation for the usual Boolean
operators (&& and ||), but also provide bitwise Boolean operators
that are not short circuit (& and |)

Ada: programmer can specify either (short-circuit is specified with

and then and or else)

FORTRAN 77: short circuit, but any side-affected place must be set to
undefined

Short-circuit evaluation exposes the potential problem of side effects

in expressions e.g. (a > b) || (b++ / 3)

Assignment Statements
The operator symbol:
1. = FORTRAN, BASIC, PL/I, C, C++, Java
2. := ALGOLs, Pascal, Modula-2, Ada
= can be bad if it is overloaded for the
relational operator for equality
e.g. (PL/I) A = B = C;
Note difference from C
11
More complicated assignments:
1. Multiple targets (PL/I)
A, B = 10

2. Conditional targets (C, C++, and Java)

(first = true) ? total : subtotal = 0

3. Compound assignment operators (C, C++, and Java)

sum += next;

4. Unary assignment operators (C, C++, and Java)

a++;

C, C++, and Java treat = as an arithmetic binary operator

e.g.
a = b * (c = d * 2 + 1) + 1

This is inherited from ALGOL 68

12
Assignment as an Expression

- In C, C++, and Java, the assignment statement produces a result

- So, they can be used as operands in expressions

e.g. while ((ch = getchar() != EOF) { ... }

Disadvantage
- Another kind of expression side effect

Mixed-Mode Assignment
- In FORTRAN, C, and C++, any numeric value can be assigned to any
numeric scalar variable; whatever conversion is necessary is done
- In Pascal, integers can be assigned to reals, but reals cannot be assigned to
integers (the programmer must specify whether the conversion from real
to integer is truncated or rounded)
- In Java, only widening assignment coercions are done
- In Ada, there is no assignment coercion

13
Levels of Control Flow:

1. Within expressions
2. Among program units
3. Among program statements

Evolution:

- FORTRAN I control statements were based directly on IBM 704 hardware

- Much research and argument in the1960s about the issue

- One important result: It was proven that all
flowcharts can be coded with only two-way
selection and pretest logical loops

Def: A control structure is a control statement and the statements whose

execution it controls

Overall Design Question:

What control statements should a language have, beyond selection and pretest
logical loops?
14
Compound statements - introduced by ALGOL 60 in the form of begin...end

A block is a compound statement that can define a new scope (with local
variables)

Selection Statements
Design Issues:
1. What is the form and type of the control expression?
2. What is the selectable segment form (single statement, statement
sequence, compound statement)?
3. How should the meaning of nested selectors be specified?

Single-Way Examples

FORTRAN IF: IF (boolean_expr) statement

Problem: can select only a single statement; to select more, a goto must be
used, as in the following example

15
FORTRAN example:

IF (.NOT. condition) GOTO 20

...
...
20 CONTINUE

ALGOL 60 if:
if (boolean_expr) then
begin
...
end

Two-way Selector Examples

ALGOL 60 if:
if (boolean_expr)
then statement (the then clause)
else statement (the else clause)
- The statements could be single or compound
16
Nested Selectors
e.g. (Pascal) if ... then
if ... then
...
else ...
Which then gets the else?
Pascal's rule: else goes with the nearest then

ALGOL 60's solution - disallow direct nesting

if ... then if ... then
begin begin
if ... if ... then ...
then ... end
else ... else ...
end

17
FORTRAN 77, Ada, Modula-2 solution - closing
special words

Advantage: flexibility and readability

Modula-2 uses the same closing special word for

for all control structures (END)

- This results in poor readability

18
Multiple Selection Constructs

Design Issues:

1. What is the form and type of the control expression?

2. What segments are selectable (single, compound, sequential)?
3. Is the entire construct encapsulated?
4. Is execution flow through the structure restricted to include just a single
selectable segment?
5. What is done about unrepresented expression values?

Early Multiple Selectors:

1. FORTRAN arithmetic IF (a three-way selector) IF (arithmetic expression)

N1, N2, N3
Bad aspects:
- Not encapsulated (selectable segments could be anywhere)
- Segments require GOTOs
2. FORTRAN computed GOTO and assigned GOTO
19
Modern Multiple Selectors

1. Pascal case (from Hoare's contribution to ALGOL W)

case expression of
constant_list_1 : statement_1;
...
constant_list_n : statement_n
end

Design choices:
1. Expression is any ordinal type (int, boolean, char, enum)
2. Segments can be single or compound
3. Construct is encapsulated
4. Only one segment can be executed per execution of the construct
5. In Wirth's Pascal, result of an unrepresented control expression value is
undefined (In 1984 ISO Standard, it is a runtime error)
- Many dialects now have otherwise or else clause

20
2. The C and C++ switch
switch (expression) {
constant_expression_1 : statement_1;
...
constant_expression_n : statement_n;
[default: statement_n+1]
}
Design Choices: (for switch)
1. Control expression can be only an integer type
2. Selectable segments can be statement sequences, blocks, or compound
statements
3. Construct is encapsulated
4. Any number of segments can be executed in one execution of the
construct (there is no implicit branch at the end of selectable segments)
5. default clause is for unrepresented values (if there is no default, the
whole statement does nothing)
- Design choice 4 is a trade-off between reliability and flexibility
(convenience)
- To avoid it, the programmer must supply a break statement for each
segment
21
3. Ada's case is similar to Pascal's case, except:
1. Constant lists can include:
- Subranges e.g., 10..15
- Boolean OR operators
e.g., 1..5 | 7 | 15..20
2. Lists of constants must be exhaustive
- Often accomplished with others clause
- This makes it more reliable
Multiple Selectors can appear as direct extensions to two-way selectors, using else-if
clauses (ALGOL 68, FORTRAN 77, Modula-2, Ada)
Ada:
if ...
then ...
elsif ...
then ...
elsif ...
then ...
else ...
end if

- Far more readable than deeply nested if's

- Allows a boolean gate on every selectable group

22
Iterative Statements
- The repeated execution of a statement or compound statement is
accomplished either by iteration or recursion; here we look at iteration,
because recursion is unit-level control

General design Issues for iteration control statements:

1. How is iteration controlled?
2. Where is the control mechanism in the loop?

Counter-Controlled Loops
Design Issues:
1. What is the type and scope of the loop var?
2. What is the value of the loop var at loop termination?
3. Should it be legal for the loop var or loop parameters to be changed in
the loop body, and if so, does the change affect loop control?
4. Should the loop parameters be evaluated only once, or once for every
iteration?

23
1. FORTRAN 77 and 90

- Syntax: DO label var = start, finish [, stepsize]

- Stepsize can be any value but zero
- Parameters can be expressions

- Design choices:
1. Loop var can be INTEGER, REAL, or DOUBLE
2. Loop var always has its last value
3. The loop var cannot be changed in the loop, but the parameters can;
because they are evaluated only once, it does not affect loop control
4. Loop parameters are evaluated only once

FORTRAN 90’s Other DO

- Syntax:
[name:] DO variable = initial, terminal [, stepsize]
…
END DO [name]
- Loop var must be an INTEGER

24
2. ALGOL 60
- Syntax: for var := <list_of_stuff> do statement where <list_of_stuff> can
have:
- list of expressions
- expression step expression until expression
- expression while boolean_expression

for index := 1 step 2 until 50,

60, 70, 80,
index + 1 until 100 do
(index = 1, 3, 5, 7, ..., 49, 60, 70, 80, 81, 82, ..., 100)

- ALGOL 60 Design choices:

1. Control expression can be int or real; its scope is whatever it is
declared to be
2. Control var has its last assigned value after loop termination
3. The loop var cannot be changed in the loop, but the parameters can,
and when they are, it affects loop control
4. Parameters are evaluated with every iteration, making it very complex
and difficult to read

25
3. Pascal
- Syntax:
for variable := initial (to | downto) final do
statement
- Design Choices:
1. Loop var must be an ordinal type of usual scope
2. After normal termination, loop var is undefined
3. The loop var cannot be changed in the loop; the loop parameters can be
changed, but they are evaluated just once, so it does not affect loop
control
4. Just once

4. Ada
- Syntax:
for var in [reverse] discrete_range loop
...
end loop

26
Ada Design choices:

1. Type of the loop var is that of the discrete range; its scope is the
loop body (it is implicitly declared)
2. The loop var does not exist outside the loop
3. The loop var cannot be changed in the loop, but the discrete range
can; it does not affect loop control
4. The discrete range is evaluated just once

5. C
- Syntax:
for ([expr_1] ; [expr_2] ; [expr_3]) statement
- The expressions can be whole statements, or even statement sequences,
with the statements separated by commas
- The value of a multiple-statement expression is the value of the last
statement in the expression
e.g.,
for (i = 0, j = 10; j == i; i++) ...

27
- If the second expression is absent, it is an infinite
loop
C Design Choices:
1. There is no explicit loop var
2. Irrelevant
3. Everything can be changed in the loop
4. Pretest
5. The first expression is evaluated once, but the other two are evaluated
with each iteration
- This loop statement is the most flexible
6. C++
- Differs from C in two ways:
1. The control expression can also be Boolean
2. The initial expression can include variable definitions (scope is from
the definition to the end of the function in which it is defined)
7. Java
- Differs from C++ in two ways:
1. Control expression must be Boolean
2. Scope of variables defined in the initial expression is only the loop body
28
Logically-Controlled Loops

- Design Issues:
1. Pretest or postest?
2. Should this be a special case of the counting loop statement (or a
separate statement)?

- Language Examples:
1. Pascal has separate pretest and posttest logical loop statements (while-do
and repeat-until)
2. C and C++ also have both, but the control expression for the posttest
version is treated just like in the pretest case (while - do and do - while)
3 Java is like C, except the control expression must be Boolean (and the
body can only be entered at the beginning--Java has no goto)
4. Ada has a pretest version, but no posttest
5. FORTRAN 77 and 90 have neither

29
User-Located Loop Control Mechanisms
- Design issues:
1. Should the conditional be part of the exit?
2. Should the mechanism be allowed in an already controlled loop?
3. Should control be transferable out of more than one loop?

Examples:
1. Ada - conditional or unconditional; for any loop; any number of levels
for ... loop LOOP1:
... while ... loop
exit when ... ...
... LOOP2:
end loop for ... loop
...
exit LOOP1 when ..
...
end loop LOOP2;
...
end loop LOOP1;
30
2. C , C++, and Java - break
Unconditional; for any loop or switch; one level only (Java’s can have a
label)
There is also has a continue statement for loops; it skips the remainder of this
iteration, but does not exit the loop

3. FORTRAN 90 - EXIT
Unconditional; for any loop, any number of levels

FORTRAN 90 also has CYCLE, which has the same semantics as C's
continue

Iteration Based on Data Structures

- Concept: use order and number of elements of some data structure to
control iteration
- Control mechanism is a call to a function that returns the next element in
some chosen order, if there is one; else exit loop

C's for can be used to build a user-defined iterator

e.g. for (p=hdr; p; p=next(p)) { ... }

31
- Perl has a built-in iterator for arrays and hashes
- e.g., foreach $name (@names) { print $name }

Unconditional Branching
Problem: readability
- Some languages do not have them:e.g., Modula-2 and Java

Label forms:
1. Unsigned int constants: Pascal (with colon)
FORTRAN (no colon)
2. Identifiers with colons: ALGOL 60, C
3. Identifiers in << ... >>: Ada
4. Variables as labels: PL/I
- Can be assigned values and passed as parameters
- Highly flexible, but make programs impossible to read and difficult to
implement

32
Restrictions on Pascal's gotos:

 A statement group is either a compound statement or the body of a

repeat-until

 The target of a goto cannot be a statement in a statement group

that is not active

 Means the target can never be in a statement group that is at the

same level or is nested more deeply than the one with the goto

 An important remaining problem: the target can be in any

enclosing subprogram scope, as long as the statement is not in a
statement group

 This means that a goto can terminate any number of subprograms

33
Guarded Commands (Dijkstra, 1975)

Purpose: to support a new programming methodology (verification during

program development)

1. Selection: if <boolean> -> <statement>

[] <boolean> -> <statement>
...
[] <boolean> -> <statement>
fi

-Semantics: when this construct is reached,

- Evaluate all boolean expressions
- If more than one are true, choose one nondeterministically
- If none are true, it is a runtime error

Idea: if the order of evaluation is not important, the program should not
specify one

34
2. Loops do <boolean> -> <statement>
[] <boolean> -> <statement>
...
[] <boolean> -> <statement>
od

Semantics: For each iteration:

- Evaluate all boolean expressions
- If more than one are true, choose one nondeterministically; then start
loop again
- If none are true, exit loop

Connection between control statements and program verification is

intimate
- Verification is impossible with gotos
- Verification is possible with only selection and logical pretest loops
- Verification is relatively simple with only guarded commands

35