Metrics and Problem Detection
Tudor Gîrba
www.tudorgirba.com

Software is complex.
29% Succeeded
18% Failed
53% Challenged
The Standish Group, 2004

How large is your project?

How large is your project?
1’000’000 lines of code

How large is your project?
1’000’000 lines of code
* 2 = 2’000’000 seconds

How large is your project?
1’000’000 lines of code
* 2 = 2’000’000 seconds
/ 3600 = 560 hours

How large is your project?
1’000’000 lines of code
* 2 = 2’000’000 seconds
/ 3600 = 560 hours
/ 8 = 70 days

How large is your project?
1’000’000 lines of code
* 2 = 2’000’000 seconds
/ 3600 = 560 hours
/ 8 = 70 days
/ 20 = 3 months

}
}
{
{
}
}
{
{
g
rin
ee
gin
en
d
ar
rw
fo

fo
rw
ar
d
en
gin
ee
rin
g
{ {
{ { { {
} { {
}
} actual development } }
} { } } }

What is the current state?
fo
rw
ar
What should we do?
d
en
Where to start?
gin
ee
How to proceed?
rin
g
{ {
{ { { {
} { {
}
} actual development } }
} { } } }

fo
rw
g
rin
ar
ee
d
gin
en
gin
en
ee
se
rin
erv
g
re
{ {
{ { { {
} { {
}
} actual development } }
} { } } }

Reverse engineering is analyzing a subject system to:
identify components and their relationships, and
create more abstract representations.
Chikofky & Cross, 90

{ {
{ {
}
}
}
} { }
A large system contains lots of details.

ity?
its qual
ju dge
How to { {
{ {
}
}
}
} { }
A large system contains lots of details.

http://moose.unibe.ch
http://loose.upt.ro/incode

1
Metrics
2Design
Problems
3
Code Duplication

Metrics
1

Youcannot control
what you cannot measure.
Tom de Marco

Metrics are functions that assign numbers to
products, processes and resources.

Software metrics are measurements which
relate to software systems, processes or
related documents.

Metrics compress system traits into numbers.

Let’s see some examples...

Examples of size metrics
NOM - number of methods
NOA - number of attributes
LOC - number of lines of code
NOS - number of statements
NOC - number of children
Lorenz, Kidd, 1994
Chidamber, Kemerer, 1994

McCabe cyclomatic complexity (CYCLO) counts
the number of independent paths through the code of a
function.
McCabe, 1977
 it reveals the minimum number of tests to write
 interpretation can’t directly lead to improvement action

Weighted Method Count (WMC) sums up the
complexity of class’ methods (measured by the metric
of your choice; usually CYCLO).
Chidamber, Kemerer, 1994
 it is configurable, thus adaptable to our precise needs
 interpretation can’t directly lead to improvement action

Depth of Inheritance Tree (DIT) is the (maximum)
depth level of a class in a class hierarchy.
Chidamber, Kemerer, 1994
 inheritance is measured
 only the potential and not the real impact is quantified

Coupling between objects (CBO) shows the number
of classes from which methods or attributes are used.
Chidamber, Kemerer, 1994
 it takes into account real dependencies not just declared ones
 no differentiation of types and/or intensity of coupling

Tight Class Cohesion (TCC) counts the relative
number of method-pairs that access attributes of the
class in common.
Bieman, Kang, 1995
TCC = 2 / 10 = 0.2
 interpretation can lead to improvement action
 ratio values allow comparison between systems

Access To Foreign Data (ATFD) counts how many
attributes from other classes are accessed directly from
a measured class.
Marinescu 2006

...

Design Problems
2

McCall, 1977

Metrics Assess and Improve Quality!

Metrics Assess and Improve Quality!
a lly ?
Re

McCall, 1977

Problem 1: metrics granularity
?
capture symptoms, not causes of problems
in isolation,
they don’t lead to improvement solutions

Problem 1: metrics granularity
?
capture symptoms, not causes of problems
in isolation,
they don’t lead to improvement solutions
Problem 2: implicit mapping
we don’t reason in terms of metrics,
but in terms of design principles

2 big obstacles in using metrics:
Thresholds make metrics hard to interpret
Granularity make metrics hard to use in isolation

Can metrics help me
in what I really care for? :)

fo
rw
ar
d
en
gin
ee
rin
g
{ {
{ { { {
} { {
}
} actual development } }
} { } } }

fo
rw
ar
How do I understand code?
d
en
gin
ee
rin
g
{ {
{ { { {
} { {
}
} actual development } }
} { } } }

fo
rw
ar
How do I understand code?
d
en
How do I improve code?
gin
ee
rin
g
{ {
{ { { {
} { {
}
} actual development } }
} { } } }

etr ics!
ith m
de al w
an t to
on ot w
Id
fo
rw
ar
How do I understand code?
d
en
How do I improve code?
gin
ee
rin
g
{ {
{ { { {
} { {
}
} actual development } }
} { } } }

How to get an initial understanding of a system?

Metric Value
LOC 35175
NOM 3618
NOC 384
CYCLO 5579
NOP 19
CALLS 15128
FANOUT 8590
AHH 0.12
ANDC 0.31

Metric Value
LOC 35175
NOM 3618
NOC 384
CYCLO 5579
NOP 19
CALLS 15128
FANOUT 8590
AHH 0.12
ANDC 0.31

Metric Value
LOC 35175
NOM 3618
NOC 384
CYCLO 5579
NOP 19
CALLS 15128
FANOUT 8590
ha t?
ww
AHH
An d no 0.12
ANDC 0.31

We need means to compare.

hierarchies?
coupling?

The Overview Pyramid provides a metrics
overview. Lanza, Marinescu 2006
Inheritance
ANDC 0.31
AHH 0.12
20.21 NOP 19
9.42 NOC 384
9.72 NOM 3618 NOM 418
0.15 LOC 35175 15128 CALLS 0.56
CYCLO 5579 8590 FANOUT
Size Communication

The Overview Pyramid provides a metrics
overview. Lanza, Marinescu 2006
ANDC 0.31
AHH 0.12
20.21 NOP 19
9.42 NOC 384
9.72 NOM 3618 NOM 418
0.15 LOC 35175 15128 CALLS 0.56
CYCLO 5579 8590 FANOUT
Size

The Overview Pyramid provides a metrics
overview. Lanza, Marinescu 2006
ANDC 0.31
AHH 0.12
20.21 NOP 19
9.42 NOC 384
9.72 NOM 3618 NOM 418
0.15 LOC 35175 15128 CALLS 0.56
CYCLO 5579 8590 FANOUT
Communication

The Overview Pyramid provides a metrics
overview. Lanza, Marinescu 2006
Inheritance
ANDC 0.31
AHH 0.12
20.21 NOP 19
9.42 NOC 384
9.72 NOM 3618 NOM 418
0.15 LOC 35175 15128 CALLS 0.56
CYCLO 5579 8590 FANOUT

The Overview Pyramid provides a metrics
overview. Lanza, Marinescu 2006
ANDC 0.31
AHH 0.12
20.21 NOP 19
9.42 NOC 384
9.72 NOM 3618 NOM 418
0.15 LOC 35175 15128 CALLS 0.56
CYCLO 5579 8590 FANOUT

Java C++
LOW AVG HIGH LOW AVG HIGH
CYCLO/LOC 0.16 0.20 0.24 0.20 0.25 0.30
LOC/NOM 7 10 13 5 10 16
NOM/NOC 4 7 10 4 9 15
...

The Overview Pyramid provides a metrics
overview. Lanza, Marinescu 2006
ANDC 0.31
AHH 0.12
20.21 NOP 19
9.42 NOC 384
9.72 NOM 3618 NOM 418
0.15 LOC 35175 15128 CALLS 0.56
CYCLO 5579 8590 FANOUT
close to high close to average close to low

The Overview Pyramid provides a metrics
overview. Lanza, Marinescu 2006
close to high close to average close to low

etr ics!
ith m
de al w
an t to
on ot w
Id
fo
rw
ar
How do I understand code?
d
en
How do I improve code?
gin
ee
rin
g
{ {
{ { { {
} { {
}
} actual development } }
} { } } }

How do I improve code?

Quality is more than 0 bugs.
Breaking design principles, rules and best practices
deteriorates the code;
it leads to design problems.

Imagine changing just a small design fragment

Imagine changing just a small design fragment

Imagine changing just a small design fragment
and33%
of all classes
would require changes

expensive
Design problems
are frequent
unavoidable

expensive
Design problems
are frequent
unavoidable
th em?
ate
limin
nd e
ete ct a
to d
How

God Classes tend to centralize the intelligence of the
system, to do everything and to use data from small
data-classes.
Riel, 1996

God Classes tend
to centralize the intelligence of the system,
to do everything and
to use data from small data-classes.

God Classes
centralize the intelligence of the system,
do everything and
use data from small data-classes.

God Classes
are complex,
are not cohesive,
access external data.

God Classes
are complex,
WMC is high
are not cohesive,
TCC is low
access external data.
ATFD more than few

God Classes
are complex,
WMC is high
are not cohesive,
TCC is low
access external data.
ATFD more than few
sing
uer ies u
to q s
s in ator
etric per
se m ical o
Co mpo log

Detection Strategies are metric-based queries to
detect design flaws. Lanza, Marinescu 2006
Rule 1
METRIC 1 > Threshold 1
AND Quality problem
Rule 2
METRIC 2 < Threshold 2

A God Class centralizes too much intelligence in
the system. Lanza, Marinescu 2006
Class uses directly more than a
few attributes of other classes
ATFD > FEW
Functional complexity of the
class is very high
AND GodClass
WMC ! VERY HIGH
Class cohesion is low
TCC < ONE THIRD

An Envious Method is more interested in data
from a handful of classes. Lanza, Marinescu 2006
Method uses directly more than
a few attributes of other classes
ATFD > FEW
Method uses far more attributes
of other classes than its own
AND Feature Envy
LAA < ONE THIRD
The used "foreign" attributes
belong to very few other classes
FDP ! FEW

Data Classes are dumb data holders.
Lanza, Marinescu 2006
Interface of class reveals data
rather than offering services
WOC < ONE THIRD
AND Data Class
Class reveals many attributes and is
not complex

Data Classes are dumb data holders.
Lanza, Marinescu 2006
More than a few public
data
NOAP + NOAM > FEW
AND
Complexity of class is not
high
WMC < HIGH
Class reveals many
OR attributes and is not
Class has many public complex
data
NOAP + NOAM > MANY
AND
Complexity of class is not
very high
WMC < VERY HIGH

Shotgun Surgery depicts that a change in an
operation triggers many (small) in a lot of different
operation and classes.
Lanza, Marinescu 2006

Code Duplication
3

What is Code Duplication?

What is Code Duplication?
obl em?
a pr
hy is it
A nd w

Code Duplication Detection
Lexical Equivalence
Syntactical Equivalence
Semantic Equivalence

Visualization of Copied Code Sequences
File A File B
File A
File B

Transformation Comparison
Source Code Transformed Code Duplication Data
Author Level Transformed Code Comparison Technique
Johnson 94 Lexical Substrings String-Matching
Ducasse 99 Lexical Normalized Strings String-Matching
Baker 95 Syntactical Parameterized Strings String-Matching
Mayrand 96 Syntactical Metrics Tuples Discrete comparison
Kontogiannis 97 Syntactical Metrics Tuples Euclidean distance
Baxter 98 Syntactical AST Tree-Matching

Noise Elimination
…
//assign same fastid as container fastid=NULL;
fastid = NULL; constchar*fidptr=get_fastid();
const char* fidptr = get_fastid(); if(fidptr!=NULL)
if(fidptr != NULL) { intl=strlen(fidptr)
int l = strlen(fidptr); fastid = newchar[l+]
fastid = newchar[ l + 1 ];

Enhanced Simple Detection Approach

lines from source
a b c d a b c d
lines
from
source

lines from source
a b c d a x y d
lines
from
source

lines from source
a b c a b x y c
lines
from
source

lines from source
a x b x c x d x
lines
from
source

lines from source 2
lines
from
source 1

lines from source 2
lines
from
source 1

lines from source 2
lines
from
source 1
exact
chunk

lines from source 2
lines
from
source 1
exact line
chunk bias

lines from source 2
lines
from
source 1
exact line exact
chunk bias chunk

Significant Duplication is large and should be
looked at Lanza, Marinescu 2006

1
Metrics
2Design
Problems
3
Code Duplication

Shotgun
Surgery has
uses is
has (partial) Feature Data
Envy uses Class
is partially
God has
Intensive Class
Coupling Brain has
has
Method
Extensive Brain has Significant
Coupling Class Duplication
has
is
is
has
Refused
is Tradition
Parent
Breaker
Bequest
has (subclass)
Futile
Hierarchy Lanza, Marinescu 2006
Identity Collaboration Classification
Disharmonies Disharmonies Disharmonies

Follow a clear and repeatable process

Follow a clear and repeatable process

Follow a clear and repeatable process

Follow a clear and repeatable process
mb ers!
so f nu
in term
qu ality
ab out
re ason
D on’t

QA is part of the the Development Process
http://loose.upt.ro/incode

Tudor Gîrba
www.tudorgirba.com
creativecommons.org/licenses/by/3.0/