diagnosis and semi-automatic correction of detected design inconsistencies in source code

Diagnosis and semi-automatic correction of

detected design inconsistencies in source code

Sergio CastroRELEASeD lab

Université catholique de [email protected]

Advisor: Kim Mens

1

1Monday 31 August 2009

mailto:[email protected]




An inconsistency example

AbstractAction

- (abstract) perform- isUndoable

^false.

- perform- isUndoable

^true.

- perform- undoAction

- perform- undoAction- isUndoable

^aBlock value.


^true.

2

Dummy

Action

Experimental

Action

DrawAction QueryAction



AbstractAction


^false.


^true.



^aBlock value.


^true.

2

Dummy

Action

Experimental

Action


The command design pattern



AbstractAction


^false.


^true.



^aBlock value.


^true.

2

Consistency property:! Action, an undoAction method should be provided

⇔ the isUndoable method returns true

Dummy

Action

Experimental

Action





AbstractAction


^false.


^true.



^aBlock value.


^true.

2



Dummy

Action

Experimental

Action


- isUndoable^true.




AbstractAction


^false.


^true.



^aBlock value.


^true.

2



Dummy

Action

Experimental

Action


- isUndoable^true.

- undoAction




AbstractAction


^false.


^true.



^aBlock value.?- perform- undoAction- isUndoable

^true.

2



Dummy

Action

Experimental

Action


- isUndoable^true.

- undoAction





User-defined consistency

An inconsistency exampleThe command design pattern

3


Inconsistency management activities

Predefined consistencydefinitions

Inconsistency detection

Diagnosing

Correcting

Many tools offering support for management of predefined

consistencies.

Such as Lint, CheckStyle, etc

4



User-definedconsistency definitions


Diagnosing

Correcting

We propose a general approach for the management of user-defined consistencies.

5




Diagnosing

Correcting

} IntensiVE

} Corrective framework

Explaining abnormal observations and

suggesting corrective actions

Defining and verifying consistency rules

6

Already used for inconsistency management in other areas of SE (e.g.,

requirements engineering, UML modelling, data bases)





Diagnosing

Correcting

} IntensiVE





6



( http://www.intensive.be )



http://www.intensive.be


7

IntensiVE







Defining and verifying constraints using IntensiVE

• Many kind of checks.

• Defined as relationships among sets of source code elements.

• Sets are intensionally defined and referred as Intensional Views.

8

with queries over source code


View with classes returning

true in the isUndoable

method

View with classes

implementing the undoAction

method

Verification of the Undoable Actions constraint

9






method

View with classes


method

Verification of the Undoable Actions constraintQUERY:

classChainReallyUnderstandsMethodWithName(?class,?method,isUndoable),

methodReturnsBoolean(?method,[true])

9






method

View with classes


method




QUERY:classChainReallyUnderstandsMethodWith

Name(?class,?,undoAction)

9






method

View with classes


method

If consistent, any class in the first view should be also in the

second, and vice versa






9




Verification of alternative views

AbstractAddAction

AddClassificationAction

ExperimentalAction

View with classes returning true in the isUndoable method

View with classes implementing the

undoAction method

AbstractAddAction



Name(?class,?method,isUndoable), methodReturnsBoolean(?method,[true])



10




Verification of alternative views

AbstractAddAction


ExperimentalAction

View with classes returning true in the isUndoable method

View with classes implementing the

undoAction method

AbstractAddAction


?


Name(?class,?method,isUndoable), methodReturnsBoolean(?method,[true])



10




IntensiVE(consistency checking)

11



{

11

Classes that return true in the isUndoable method



{ {

11

Classes that return true in the isUndoable method

Classes that implement the

undoAction method



12

This class does not return true at the

isUndoable method

But implements an undoAction method



12

This class does not return true at the

isUndoable method

But implements an undoAction method

ExperimentalAction


IntensiVE(diagnosing inconsistencies)

13

Why ExperimentalAction is

not a result of the query defining this set?

OR: Why ExperimentalAction is a result of the query defining

this set?

ExperimentalAction




Diagnosing

Correcting

} IntensiVE





14








Corrective framework

15

Diagnosing

Correcting


What is diagnosis?

16

Generating hypotheses explaining abnormal

observations

Abductive reasoning


What is diagnosis?

16

Generating hypotheses explaining abnormal

observations

Abductive reasoning

one of the fundamental forms of

human reasoning according to Pierce


Abduction is suitable for

• Generating hypotheses that would explain an evidence.

• Explanations expressed in terms of certain predicates, declared before hand as abducibles.

17





17

False predicates

that could be true





17

True predicates

that could be false

False predicates

that could be true


Abducibles




17

True predicates

that could be false

False predicates

that could be true

We can think in “Abducibles” as “Correctables”


Abducibles

bird(?x).

broken-wing(?x).

An example of abduction

Theory:

flies(?x) if bird(?x), not (ab(?x))

ab(?x) if broken-wing(?x)

18


Abducibles

bird(?x).

broken-wing(?x).


Theory:



18

Observation:

flies(opus)


Abducibles

bird(?x).

broken-wing(?x).


Theory:



18

Observation:

flies(opus)

failing


Abducibles

bird(?x).

broken-wing(?x).


Theory:



18

Observation:

flies(opus)

bird(opus)added to

our theory

failing


Abducibles

bird(?x).

broken-wing(?x).


Theory:



18

Observation:

flies(opus)

bird(opus)added to

our theory

succeeding


Abducibles

bird(?x).

broken-wing(?x).


Theory:



bird(tweety)

broken-wing(tweety)

19


Abducibles

bird(?x).

broken-wing(?x).


Theory:



bird(tweety)

broken-wing(tweety)

19

Observation:

flies(tweety)


Abducibles

bird(?x).

broken-wing(?x).


Theory:



bird(tweety)

broken-wing(tweety)

19

Observation:

flies(tweety)

failing


Abducibles

bird(?x).

broken-wing(?x).


Theory:



bird(tweety)

broken-wing(tweety)

19

Observation:

flies(tweety)

retracted from our theory

failing


Abducibles

bird(?x).

broken-wing(?x).


Theory:



bird(tweety)

broken-wing(tweety)

19

Observation:

flies(tweety)

retracted from our theory

succeeding


How to choose abducible predicates

?-classChainReallyUnderstandMethodWithName(ExperimentalAction, ?m, undoAction),

methodWithBooleanReturnStatement(?m, true)

:-superclassOf(?s, ExperimentalAction),

classChainReallyUnderstandMethodWithName(?s, ?m, undoAction),


:-methodWithNameInClass(?m, undoAction, ExperimentalAction),


20

- predicates that denote basic structural relationships- we know for sure that they will be evaluated as part of the resolution process of the query










20

Primitive predicates that can be corrected

- predicates that denote basic structural relationships- we know for sure that they will be evaluated as part of the resolution process of the query









21

Primitive predicates that can be corrected



Our framework allows ...

• The definition of abducible predicates.

• Declaration of multiple corrective actions.

• Generation of hypotheses explaining inconsistencies.

• Semi-automatic execution of corrective actions.

22


Defining positive explanations(new facts will be added to the theory)

23



23

Abducible predicate:in(methodWithBooleanReturnStatement(+?

method,+?boolean))



24



24

Corrective action:[[?class compile: (?selector asString, ‘^’,?boolean

asString)]]


Correcting inconsistencies

25


Correcting inconsistencies

25

Correct


The corrective browser

26



26

Corrective action name



26


Textual description



26


Textual description

Visualization in the proof three of where the

corrective action was found


Visualisation of the nodes with corrective actions

27

Color code:•Red filling means failure.

•Thick green border means availability of corrective

actions.


Future work

• How to choose among different solutions.

• Detect solutions that could cause new inconsistencies.

28


Conclusions

• We can infer solutions from the rules that define consistencies.

• A library of “primitive” solutions to small problems should be provided.

• These solutions can be composed and reused across distinct problems.

29


Many Thanks !

Questions and feedback are welcomed

30


diagnosis and semi-automatic correction of detected design inconsistencies in source code

Technology