This article needs additional citations for verification. Please help improve this articlebyadding citations to reliable sources. Unsourced material may be challenged and removed.
Find sources: "Random testing" – news · newspapers · books · scholar · JSTOR (September 2014) (Learn how and when to remove this message) |
Random testing is a black-box software testing technique where programs are tested by generating random, independent inputs. Results of the output are compared against software specifications to verify that the test output is pass or fail.[1] In case of absence of specifications the exceptions of the language are used which means if an exception arises during test execution then it means there is a fault in the program, it is also used as a way to avoid biased testing.
Random testing for hardware was first examined by Melvin Breuer in 1971 and initial effort to evaluate its effectiveness was done by Pratima and Vishwani Agrawal in 1975.[2]
In software, Duran and Ntafos had examined random testing in 1984.[3]
The use of hypothesis testing as a theoretical basis for random testing was described by Howden in Functional Testing and Analysis. The book also contained the development of a simple formula for estimating the number of tests n that are needed to have confidence at least 1-1/n in a failure rate of no larger than 1/n. The formula is the lower bound nlogn, which indicates the large number of failure-free tests needed to have even modest confidence in a modest failure rate bound.[4]
Consider the following C++ function:
int myAbs(int x) {
if (x > 0) {
return x;
}
else {
return x; // bug: should be '-x'
}
}
Now the random tests for this function could be {123, 36, -35, 48, 0}. Only the value '-35' triggers the bug. If there is no reference implementation to check the result, the bug still could go unnoticed. However, an assertion could be added to check the results, like:
void testAbs(int n) {
for (int i=0; i<n; i++) {
int x = getRandomInput();
int result = myAbs(x);
assert(result >= 0);
}
}
The reference implementation is sometimes available, e.g. when implementing a simple algorithm in a much more complex way for better performance. For example, to test an implementation of the Schönhage–Strassen algorithm, the standard "*" operation on integers can be used:
int getRandomInput() {
// …
}
void testFastMultiplication(int n) {
for (int i=0; i<n; i++) {
long x = getRandomInput();
long y = getRandomInput();
long result = fastMultiplication(x, y);
assert(x * y == result);
}
}
While this example is limited to simple types (for which a simple random generator can be used), tools targeting object-oriented languages typically explore the program to test and find generators (constructors or methods returning objects of that type) and call them using random inputs (either themselves generated the same way or generated using a pseudo-random generator if possible). Such approaches then maintain a pool of randomly generated objects and use a probability for either reusing a generated object or creating a new one.[5]
According to the seminal paper on random testing by D. Hamlet
[..] the technical, mathematical meaning of "random testing" refers to an explicit lack of "system" in the choice of test data, so that there is no correlation among different tests.[1]
This section does not cite any sources. Please help improve this sectionbyadding citations to reliable sources. Unsourced material may be challenged and removed. (August 2014) (Learn how and when to remove this message)
|
Random testing is praised for the following strengths:
The following weaknesses have been described :
Some tools implementing random testing:
Random testing has only a specialized niche in practice, mostly because an effective oracle is seldom available, but also because of difficulties with the operational profile and with generation of pseudorandom input values.[1]
Atest oracle is an instrument for verifying whether the outcomes match the program specification or not. An operation profile is knowledge about usage patterns of the program and thus which parts are more important.
For programming languages and platforms which have contracts (e.g. Eiffel. .NET or various extensions of Java like JML, CoFoJa...) contracts act as natural oracles and the approach has been applied successfully.[5] In particular, random testing finds more bugs than manual inspections or user reports (albeit different ones).[9]