fundamentals of program analysis dr. manuel oriol

Chair of Software Engineering

Fundamentals of Program Analysis

Dr. Manuel Oriol

2

Topics

Abstract Syntax Tree

Control Flow Graph

Fundamental of Dataflow Analysis Available Expressions Analysis Live Variables Analysis Very Busy Expressions Analysis Reaching Definition Analysis

3

Program Analysis

A set of techniques that consist in analyzing programs in order to prove properties on them.

Several techniques can be distinguished: Data flow analysis Constraint-based analysis Abstract Interpretation Types and effects analysis …

4

Structures

Source Code

Abstract Syntax Trees (AST)

Control-Flow Graphs (CFG)

Object code

5

Abstract Syntax Tree?

a := 1b := a + 2if (b > 3) then

Result := belse

Result := aend

Program

:=a 1 :=

b +

a 2

If… Then…Else…End

>

b 3:=

Result b :=Result a

6

How ASTs are computed? Compilers…

Use a lexical analyzer that gives you the tokens (e.g.: lex, flex, gelex…)

Couple it with a parser that groups the tokens and gives the structure (e.g.: yacc, bison, geyacc…)… generally the structure is an AST

7

Advantages and Disadvantages of ASTsAdvantages:

No more noise (comments, keywords etc…) Not ambiguous Exact

Disadvantages: Many similar forms: if/switch, for/foreach… Expressions may become big and going through

them may become costly.

So???

8

Control-Flow Graph

a := 1b := a + 2if (b > 3) then

Result := belse

Result := aendb := a

a:=1

b:=a+2

b>3

Result:=b Result:=a

b:=a

yes no

9

How CFGs are computed? Compilers…

Take the AST

Group the statements

Remove the control-flow instructions and keep only jumps and tests

Transform jumps into transitions

10

Control-Flow Graph with blocks…

a := 1b := a + 2if (b > 3) then

Result := belse

Result := ab := a

endb := a

a:=1b:=a+2

b>3

Result:=b Result:=ab:=a

b:=a

yes no

11

Advantages and DisadvantagesAdvantages: no more similar forms expressions are easy to understand: only

statements

Disadvantages: introduce temporaries less understandable, regarding to the original code

and the general structure

12

So what can we do with that?

Example:Prove that there is no call on a Void reference in this program:

create b.make (10)from a := 1 until a > 10 loop

b.testa := a + 1

end

13

And what about this program?

create b.make (10)from a := 1 until a > 10 loop

if (a = 10) then b := Voidelse b.testenda := a + 1

end

14

Dataflow AnalysisAs we could see in previous examples:

There is a need for an analysis of data manipulations

Traditionally, it leads to optimizations (e.g. statements that have no influence may be removed), loops may have statements extracted from them etc…

Based on all paths through programs

It is the analysis of labels by excellence

15

Technical Framework

The set of labels for a node are what really matters as it is what we want to prove.

Labels as the first thing to build and describe

The rule to build the set of labels is by itself the analysis

We can use a lot of ways to describe that but people usually use set theory in order to do that and describe it with a transitional set system on statements

In(S), Out(S), succ(S), pred(S)

16

Notations

S0In(S0)={…}Out(S0)={…}

S’1S1

S-1 S’-1pred(S0)={S-1,S’-1}

succ(S0)={S1,S’1}

17

Example: Attachment Analysis

Labels ::= varname=attached | varname=detached |varname=error

In (S) = ∪S0∈ pred(S) Out(S0)Out(S= create varname.creationFeature) =

(In(S) - {varname=detached})∪ {varname=attached}Out(S= varname:=Void) =

(In(S) - {varname=attached})∪ {varname=detached}Out(S= varname0:=varname1.FeatureName(…)) =

if {varname1=detached} ∈ In(S) then In(S) ∪ {varname1=error}

In(S) ∪ {varname0=attached} otherwise

18

Applications of the algorithm

create b.make (10)from a := 1 until a > 10 loop

b.testa := a + 1

end

create b.make (10)from a := 1 until a > 10 loop

if (a = 10) then b := Void

else b.testenda := a + 1

end

19

Available Expressions AnalysisAn expression is available at a point p if there is a value calculated for sure and the last value that was calculated for it is still the correct value.Labels ::= {expression}In(S) = ∩S0∈ pred(S) Out(S0)Out(S= varname:= expression0) =

(In(S) ∪ {expression0} ) - {e | varname ∈ e }

Kill(S)Gen(S)

20

Application of the algorithm

a := a - bb := a + bif (b > 3) then

Result := a - belse

Result := a + bend

21

Live Variables AnalysisA variable v is live at a point p if it may be used in the future before it is overwritten.Labels ::= {varname}Out(S) = ∪S0∈ succ(S) In(S0)In(S= varname:= expression) =

(Out(S)-{varname}) ∪ {vname | vname ∈ expression}

In(S | expression ∈ S) = Out(S) ∪ {vname | vname ∈ expression}

Kill(S) Gen(S)

22

Application of the algorithm

a := 1b := a + 2if (b > 3) then

Result := belse

Result := ab := a

endb := a

23

Very Busy Expressions Analysis

An expression e is very busy at a point p if all future execution path will need its evaluation before the value of e is changed

Labels ::= {expressions}Out(S) =∩S0∈ succ(S) In(S0)In(S= varname:=…) =

Out(S) - {e0 ∈ Out(S) | varname ∈ e0} (apply first)In(S ∋ expression0) = Out(S)∪ {expression0} (apply

last)

24

Application of the algorithm

a := 2if (b > 3) then

Result := a + belse

Result := a + bend

25

Reaching Definition Analysis (also def-use)

A definition (an assignment) of v reaches a point in a program if there is no other definition of v in-between.

Labels ::= {varname:statementNumber}In(S) = ∪S0∈ pred(S) Out(S0)Out(S= varname:=…) =

In(S)- {varname:statementNumber0 ∈ In(S)} ∪ {varname:statementNumber(S)}

Out(S) = In(S) otherwise

26

Application of the algorithm

a := 1b := a + 2if (b > 3) then

Result := belse

Result := a

b := aendb := a

27

Types of analysis…

Forward Analysis: the analysis goes in the same direction as the edgesBackward Analysis: the analysis goes in reverse direction comparing to the edges

Must analysis: all branches are accountedMay analysis: one branch is enough

28

Terminations of the Dataflow Analyses???

Label values generally form lattices so they are finite

Most functions are monotonic on these lattices so they are terminating.

It is however needed to verify that they are!!!

29

Exercise 1Find applications for each of the analyses we previously described:

Available Expressions Analysis

Live Variables Analysis

Very Busy Expressions Analysis

Reaching Definition Analysis

30

Exercise 2: Password leak

How to prove statically that a password is not going to be passed from its original definition to any output?

Setting: PASSWORD is a class Specify the interface Specify the analysis and its limitations

31

Exercise 3: Access rights for files

How to verify that a file can only be accessed by a thread that acquired the right to do it?

Setting: FILE is a class that has a limited number of

features SECURITY_MANAGER grants accesses Specify the analysis and its limitations

32

Conclusions

Abstract Syntax Tree

Control Flow Graph

Fundamental of Dataflow Analysis Available Expressions Analysis Live Variables Analysis Very Busy Expressions Analysis Reaching Definition Analysis