Exceptions
SWE 619: Software Construction
Last Modified Spring 2010
Saket Kaushik, Paul Ammann
Inputs to Procedures

Sqrt procedure
public float sqrt (float x)




Square root can only be calculated for non negative
floats. sqrt() does not makes sense for negative floating
point numbers.
Procedure definition allows any float value to be
passed (positive or negative)
How should a procedure cope with negative
floats?
Use Partial Procedures?

Usually a bad idea
2
Partial Procedures

Announces its contract for use




Set of preconditions that must be ensured at the
time of invocation
Theoretically sound…
Guarantees post conditions subject to
preconditions
BUT clients can always call method without
satisfying preconditions



How does code behave in such a case
Not possible to be sure what client gets back
What happens in next release?

Do clients rely on undocumented behavior?
3
Handling Undefined Inputs



“Normal” processing is not possible
Only option: Define behavior to inform client
How to inform client?
One option: Return special values
float sqrt (float x)
//Effects: if x>=0.0 returns approx square root;
// if x <0.0 returns 0.0

4
Problems with special values
float sqrt (float x)
//Effects: if x>=0.0 returns approx square root;
// if x <0.0 returns 0.0
Consider following calls:
float root1 = sqrt(-45.4); float root2 = sqrt(0.0);
if(root1-root2) {bla bla…}
 if special value is a possible return value of the
procedure, it results in ambiguities

no leftover values to use
5
Problems with special values

public int search(int [ ] a, int x)
//Effects: If a not null and not empty,
//returns index of x in a if present; else
//returns -1
not found? empty? null?
6
Alternate approach:
Exceptions


Provide an alternate return mechanism
No overlap with ‘real’ values


ArrayIndexOutofBoundsException never overlaps
with another real value, but -1 or null does
Exceptions force client to treat special values
‘specially’

usually in a separate code block; not in the normal
execution sequence.
7
Specifications

Signature:
public int foobar(int x, int y) throws <List of exceptions>
e.g: float sqrt (float x) throws NegativeNumberException



Liskov: List all Exceptions thrown
Bloch: List only checked exceptions
We will generally follow Liskov, but the
issue is mostly about style, not
substance
8
Requires  Effects
Effects clause should explain what causes each
exception to be thrown, define behavior for all
inputs in require clause
float sqrt (float x)
//Requires: x>=0.0
//Effects: …
float sqrt (float x)
//Requires:
//Effects: If x < 0.0 throw
//NegativeNumberException; …
float sqrt (float x) throws NNE
//Effects: If x< 0.0 throw NNE; …
9
Java Exception Mechanism
Throwable
Error
RuntimeException
Exception
…
Checked Exceptions
…
Unchecked Exceptions
10
Exceptions




Unchecked: client does not need to
explicitly write special code
Checked: client has to pay attention to
these. Must write special code blocks
Many different types of exceptions are
provided.
You can provide your own.
11
Custom Exceptions
public MyFavoriteException extends Exception {
public MyFavoriteException() {
super();
}
public MyFavoriteException(String s) {super(s);}


can add more
like state at the time of throwing the exception
methods that let the invalid state be printed and other useful
information
12
Throwing Exceptions



Explicit throws
throw new NullPointerException(“”);
throw new NPE(“class, method”);
System throws
String [ ] a = {“1”,”2”,”3”,”4”,”5”};
print(a[-1]);
System throws IndexOOBE
Passing through (goes up the call chain)

keeps going till it finds a handler
13
Catching Exceptions

In a special code block called try-block
try{
}catch(ExceptionType instance){
}
14
Catching Exceptions


Can have multiple catch blocks, with
different execution logic for different
exceptions raised
Class hierarchy matters in catching


If you catch an exception of type
Exception, all subtypes are also caught
catching Exception is dangerous. WHY?
15
Remember!


try-catch block should be as short as
possible.
catch the right type of exception: to be
sure of what’s the right thing to do
16
Reflection Vs. Masking

Masking: handle the exception in the
procedure.



client does not expect the exception
client may not understand the exception
Reflection: pass it over to the calling
code

client expects it
17
Reflection Example
public static int min (int[] a) throws NPE,EE {
// Effects if a is null throw NPE else if a is
empty throw EE else return min value in a
int m;
try { m = a[0];} catch (IOOBE e) {
throw new EE (“Arrays.min”); }
for (int i=0; i < a.length; i++)
if (a[i] < m) m = a[i];
return m;}
18
Masking Example
public static boolean sorted (int[] a) throws NPE {
// Effects if a is null throw NPE else if a is sorted
in ascending order return true else return false
int p;
try { p = a[0];} catch (IOOBE e) {
return true; }
for (int i=0; i < a.length; i++)
if (p < a[i]) {p = a[i];} else {return false;}
return true;}
19
Meyer vs Liskov




Meyer’s approach is clean and easy to
understand
software by contract! if preconditions
not met, it’s the client’s fault. Client
should suffer!
Recognizes 1 exception: Failure
exception
Industry practice is to follow Liskov
20