long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
5
@Test
factorialWith5Test() {
long obs = fact(5);
assertTrue(5 < obs);
}
Coverage
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
6
@Test
factorialWith5Test() {
long obs = fact(5);
assertTrue(5 < obs);
}
Coverage
7
@Test
factorialWith5Test() {
long obs = fact(5);
assertTrue(5 < obs);
}
@Test
factorialWith5Test() {
assertEqual(1, fact(0));}
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
8
@Test
factorialWith5Test() {
long obs = fact(5);
assertTrue(5 < obs);
}
@Test
factorialWith5Test() {
assertEqual(1, fact(0));}
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
Are these test cases good at
detecting bugs?
9
@Test
factorialWith5Test() {
long obs = fact(5);
assertTrue(5 < obs);
}
@Test
factorialWith5Test() {
assertEqual(1, fact(0));}
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
Are these test cases good at
detecting bugs?
Let’s mutate our code to see.
Mutation analysis
•Tests are good if they can detect bugs
•Principle: generate bugs and test the tests
•Mutation operators = types of bugs
•Mutant = Program with one seeded bug
11
Mutation analysis
Input : P, TS, Ops
Output : score, coverage
M <- generateMutants (P, OPs)
forAll (m in M)
run (TS,m)
if (one-test-fail)
then killed <- m
else alive <- m
score = killed / size(M)
12
long fact(int n) {
if(n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
14
long fact(int n) {
if(n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
15
n != 0 return 1+1
< --!(i<=n)
result/i
result+1
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
16
n != 0 return 1+1
< --!(i<=n)
result/i
result+1
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
17
@Test
factorialWith5Test() {
long obs = fact(5);
assertTrue(5 < obs); }
n != 0 return 1+1
< --!(i<=n)
result/i
result+1
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
18
@Test
factorialWith5Test() {
long obs = fact(5);
assertTrue(5 < obs); }
@Test
factorialWith5Test() {
assertEqual(1, fact(0));}
n != 0 return 1+1
< --!(i<=n)
result/i
result+1
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
Mutation reveals bugs in the test suite
19
@Test
factorialWith5Test() {
long obs = fact(5);
assertTrue(5 < obs); }
@Test
factorialWith5Test() {
assertEqual(1, fact(0));}
Bugs in the test suite:
- Weak oracle
- Missing input
n != 0 return 1+1
< --!(i<=n)
result/i
result+1
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
20
@Test
factorialWith5Test() {
long obs = fact(5);
assertEqual(120, obs); }
@Test
factorialWith5Test() {
assertEqual(1, fact(0));}
@Test
factorialWith1Test() {
assertEqual(1, fact(1));}
n != 0 return 1+1
< --!(i<=n)
result/i
result+1
Descartes
•Mutation operators: extreme mutation
•Active, open source development
•Pitest plugin
•Compared to PIT
• Less mutants
• Different type of feedback
• Same framework
22
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
Descartes - example
23
long fact(int n) {
if (n == 0)return 1;
long result = 1;
for(int i=2; i<=n; i++)
result = result * i;
return result;}
Descartes - example
24
long fact(int n){
return 0;}
long fact(int n){
return 1;}
PIT : 7 mutants
Descartes : 2 mutants
43
public boolean enableMetaApi() {return true;}
public boolean enableMetaApi() {return false;}
package com.spotify.apollo.environment;
public class ApolloConfig {
public boolean enableMetaApi() {
return optionalBoolean(apolloNode, "metaApi").orElse(true);
}
Covered by 8 test cases which all expect
enableMetaApi()to return true
44
package com.spotify.apollo.environment;
public class ApolloConfig {
public boolean enableMetaApi() {
return optionalBoolean(apolloNode, "metaApi").orElse(true);
}
public boolean enableMetaApi() {return true;}
public boolean enableMetaApi() {return false;}
Descartes
•Current
• Compatible with JUnit5 and latest Pitest
• Integrates with Maven and Gradle
•Future: Descartes and CI
• Incremental Descartes on a commit
• Descartes for pull requests
45
Conclusion
•Mutation analysis
• Automatic generation of mutants
• Evaluate the test suite
•Bugs in test suites
• Oracle
• Input space coverage
• Testability
• Indirectly tested code
46