Equivalence partitioning (5.2) gives you the partitions. But it raises a new question: which sample should you pick from each partition? The sample carries a big responsibility, because it speaks for the whole group. The answer is boundary value analysis (BVA): most bugs live at the edges of a partition, not in its middle.
Think about a cinema. Children aged 12 and under get a cheap ticket. The programmer writes:
if (age < 12) {
price = childPrice; // bug! should be age <= 12
} else {
price = adultPrice;
}Ages 5, 8 and 10 all get the right price. A random test from the middle of the child partition passes every time. Only a child who is exactly 12 gets the wrong price. A test on the edge finds the bug immediately.
That is the whole reason for BVA. Programmers write the borders of partitions as comparisons (<, <=, >, >=), and the classic slip is choosing the wrong one, or being off by one. Nobody ever looks at the edges, except BVA testers.
For every boundary, test three values: exactly on the boundary, just below it, and just above it. If the input is a range from to , that means and plus the values just around them. If the input is a number of values (a count), test the minimum and maximum too.
The valid range is . It has two boundaries, so BVA gives six tests: −1, 0, 1 around the lower edge and 74, 75, 76 around the upper edge.
Notice that each edge produces one value from each side: −1 is in the invalid partition below, 0 and 1 are valid; 74 and 75 are valid, 76 is invalid. So the BVA values also exercise the invalid partitions from EP.
| Step | Technique | What you do |
|---|---|---|
| 1 | EP first | Find all the partitions: valid and invalid, inputs and outputs. |
| 2 | BVA next | For each partition, pick the samples on and around its boundaries. |
| 3 | Test | If a sample passes, the partition passes. If it fails, the partition fails. |
Exam radar: EP and BVA have different purposes but work hand in hand. EP finds the partitions; BVA picks effective samples from them. BVA answers the question: how do we know our samples are good?
Every test design should include both kinds of test. You met them in Lecture 2 with the calculator.
Think of a shop. A positive test is a polite customer: they ask for something on the menu, pay the right amount and wait their turn. A negative test is a clumsy or naughty customer: they order something that does not exist, pay with a button, or leave the form half-filled. Real software meets both kinds of customer, so it must be tested with both.
| Positive test | Negative test | |
|---|---|---|
| Goal | Check that the software works under normal conditions. | Check that the software handles bad or unexpected input safely. |
| Input | Valid data. Normal use. | Invalid, out of range, wrong type, missing data. |
| Expected result | The correct output. | An error message or a clean rejection. No crash. |
| Grade example (5.4) | exam 44, coursework 15 → B | exam 80 → fault message; exam 'Q' → fault message |
A negative test still has an expected result: “the program shows a fault message and keeps running” is a precise expectation. “Something goes wrong” is not.
Common mistake: testing only the polite customer. A test set made only of valid inputs never checks whether the program survives the clumsy one, and in production the clumsy customer always arrives.
Which values does BVA test for a range from 1 to 100?
The range has two boundaries, 1 and 100. For each, test on, just below and just above:
Six values: 0 and 101 are invalid (negative tests), 1, 2, 99 and 100 are valid.
Why do we need BVA if we already have EP?
EP tells us the partitions, but not which sample to pick from each one. Since every member of a partition is assumed to behave the same, a sample from the middle looks as good as any other. But most bugs sit at the boundaries (a < that should have been <=, an off-by-one), and a middle value never touches them.
BVA picks samples on and around the edges, so the one sample that speaks for each partition is also the one most likely to reveal a bug. EP finds the groups; BVA makes the samples effective.
A cinema gives a child price to ages 12 and under. The code says if (age < 12). Which of the test ages 5, 10, 12 and 13 reveals the bug? Which BVA tests would you choose for this boundary?
Compare the specification (child if age ) with the code (child if age ):
| Age | Spec says | Code gives | Bug revealed? |
|---|---|---|---|
| 5 | child | child | no |
| 10 | child | child | no |
| 12 | child | adult | yes |
| 13 | adult | adult | no |
Only age 12 reveals the bug. BVA chooses the values on and around the boundary, 11, 12 and 13, so it is guaranteed to include 12. Random middle values like 5 and 10 never find it.
An age field accepts 18 to 60. Give the EP partitions, then the BVA test values, and label each value as a positive or negative test.
EP: a range, so 1 valid + 2 invalid partitions: below 18 (invalid), 18 to 60 (valid), above 60 (invalid).
BVA: two boundaries, three values each.
| Value | Partition | Kind of test | Expected result |
|---|---|---|---|
| 17 | below 18 | negative | rejected with an error message |
| 18 | valid | positive | accepted |
| 19 | valid | positive | accepted |
| 59 | valid | positive | accepted |
| 60 | valid | positive | accepted |
| 61 | above 60 | negative | rejected with an error message |
Six tests: 17, 18, 19, 59, 60, 61, of which 17 and 61 are negative tests.
In the grade example (exam mark out of 75, coursework out of 25), give one positive and one negative test.
A colleague says: “A negative test has no expected output, because the input is wrong anyway.” Is that right?
No. Every test case needs an expected result, or you cannot decide whether it passed. For a negative test the expected result is the safe handling the specification requires: a specific error or fault message, or a clean rejection, and no crash. For example, exam mark 'Q' → expected “fault message”, program keeps running. If the program crashed, or quietly computed a grade, the negative test fails and has found a bug.