See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/221610483

UML Analysis Using State Diagrams.

Conference Paper · January 2006

Source: DBLP

CITATIONS READS
3 3,832

3 authors, including:

Mohammad Alanazi D.A. Gustafson

King Saud University Kansas State University
77 PUBLICATIONS 940 CITATIONS 67 PUBLICATIONS 757 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by D.A. Gustafson on 25 October 2014.

The user has requested enhancement of the downloaded file.

 UML ANALYSIS USING STATE DIAGRAMS
Mohammad N. Alanazi, Jason A. Belt, Dr. David A. Gustafson
Department of Computing and Information Sciences
College of Engineering
Kansas State University
Manhattan, KS, U.S.A
{alanazi, belt, dag}@ksu.edu

Abstract — This article demonstrates a new approach to in and the transitions which cause that object to change
analyzing UML designs using state and sequence state. A state diagram provides the component view of an
diagrams. From multiple state diagrams, a super-state object. To understand how the states of one object
diagram which includes the cross-product of the selected interact with the states of another object, a different
states is built along with a transition matrix of possible approach has to be used.
transitions. The closure of the transition matrix is used to
identify unreachable states and impossible transitions. The UML specification does not enforce any consistency
Additionally, the closure is also used to evaluate requirements between the information contained in the
consistency between the state diagrams and the sequence sequence and state diagrams. While this does allow for
greater flexibility in how UML can be used, it can lead to
diagrams. Missing and impossible sequences can be
inconsistent views of the system being modeled. The
identified. A prototype tool has been built to calculate the
problem of relating state diagrams with sequence
closure of the transition matrix and to compare the
diagrams has recently been the focus of research in the
results with the sequence diagrams. software engineering community [e.g. 1]. However, the
work has usually involved just one state diagram and one
1. Introduction or more sequence diagrams.

Finding errors in software designs before they are Our approach to consistency analysis combines the state
implemented is very important. Most researchers in information of multiple state diagrams into a composite
software engineering have found that the earlier an error super-state diagram. This super-state diagram details all
is found, the easier it is to correct. If we can improve the of the possible composite states the objects can be in as
analysis capabilities for software designs, we may have a well as the transition pairs which lead from one
significant impact on removing faults earlier in the composite state to another. In this way the super-state
software development process. diagram provides the complete collaborative view of a set
of objects in the model. A given sequence diagram then
The typical software design is specified using UML should be a valid subsequence of the set of sequences that
diagrams [7] including class diagrams, use case and are possible in a super- state diagram.
sequence diagrams, and state diagrams. Although class
diagrams and use case/sequence diagrams are the most 2. Transition Matrix
common diagrams in a software design, the state diagram
provides an excellent notation for specifying the behavior The basis of our analysis techniques is a transition matrix
of objects and how methods affect the objects. that details the possible global states of the system based
on a vector of states of individual instances of classes and
Sequence diagrams are interclass (interobject or the possible transitions between the states in the global
interagent) and they detail how objects in the model state vector. Consider a program that has class X and
interact via method calls. A sequence diagram can be class Y. Let class X have an initial state A and two other
viewed as a partial collaborative view of a set of objects. states, B and C, while class Y has an initial state D and a
Transitions in a sequence diagram occur as the result of second state E. The state diagrams will depict how
method calls between objects. Methods of different instances of X and Y can transition between those states.
objects are often paired together as the result of one Let class X call class Y whenever class X transitions
object’s method calling the method of another object. between state A and B. Table 1 shows possible
transitions in the super-state diagram that is the cross-
A state diagram is intraclass (intraobject or intraagent) product of all states with one instance of X and one
and it describes the states one object in the model can be instance of Y.
 Table 1 - Super-state transition matrix T1.
T1 AD BD CD AE BE CE
AD 0 0 0 0 1 0
BD 1 0 1 0 0 0
CD 0 1 0 0 0 0
AE 0 1 0 0 0 0
BE 0 0 0 1 0 1
CE 0 0 0 0 1 0

An entry in a cell in T1 (Table 1) shows that in one step,

the system can transition from the state of the row to the
state of the column. Taking the product of T1 by itself
gives a matrix that contains the transitions possible with
two steps. The closure of T1 is the sum of products, T1 +
T1*T1 + T1*T1*T1 +….The closure shows all possible
transitions in any number of steps. Although the closure
is represented as an infinite sum, it can be calculated in at
most the number of products equal to the rank of the
initial matrix. In most cases, it is even smaller than that Figure 1 - Class Diagram for the library example
number.

3. Error Discovery
The closure of the transition matrix for a cross-product of
a number of instances will be the basis for our analysis.
The closure of the transition matrix by itself can show
four kinds of errors:
1. existence of a bad combination of states
2. unreachability of a good combination of states
3. existence of a bad transition
4. unreachability of an important transition

The output from the closure of the transition matrix can

also be compared with the class diagram and the sequence
diagrams to look for consistency. Since all possible
transitions are generated by the closure, every sequence Figure 2 – State Diagram for Book
that appears in a sequence diagram should be in the
closure. Also, for completeness, every sequence in the
closure should be represented by a sequence diagram.
Although this is not absolute, it would be a good check
for completeness of the sequence diagrams to check those
closure matrix sequences to see if they warrant inclusion
in a sequence diagram.

4. Library Example
This example describes the interaction between a patron
of a library and the copies of books the library holds. In
order to simplify the model the library holds only one
copy of each book. Figure 1 is the class diagram for this
model and Figure 2 and Figure 3 are the state diagrams
for the patron and book objects. Note that the transitions
in the state diagrams are numbered for ease of reference.
This example along with the sequences diagrams was
Figure 3 – State Diagram for Patron
created by a team of students trying to create a correct
model of a simple library system.
The patron object can be in one of three states; Good Table 2 - Portion of A1
Standing, Too Many Books, and Overdue Fines. We will GOO GOC GOD GOR GCO
call these states G, T, and F respectively for the rest of GOO 1,21 1,11
this paper. A patron starts in G until the number of books GOC 26 23 2,22
the patron has checked out is equal to MAX or the patron GOD
returns an overdue book. In the former, the patron will GOR 25
transition to state T where they will remain until they GCO 16
return a book. In the latter, the patron will transition to F
where they will not be able to do anything until they pay The row headings are the initial states and the column
the fine that is owed. Note that this does not allow the headings are the final states. The numbers in the table
patron to return additional books if they currently owe a arise from Figure 2 and Figure 3. For the purpose of
fine (a subtle and unintended difference from most actual clarification we have assigned unique numeric identifiers
library systems). to the transitions for each instance of an object in our
system. The book object has two numeric identifiers for
A book object has four states; On Shelf, Checked Out, each transition since we have two instances of that object.
Overdue, and Returned. We will call these states O, C, D, For example, GOO → GOC represents a patron in good
and R respectively for rest of this paper. When a book is standing checking out the second book. The 1 indicates
checked out it transitions from O to C. the patron took the transition labeled checkout [n < MAX]
and the 21 indicates the second book took the transition
The two transitions from C labeled check represent the labeled checkout. If there is an entry for a cell in the
library determining if the book is overdue. If the book is matrix then the transition is valid.
overdue it will transition to D. Otherwise it will transition
to R where it will remain until the library places it back A2 is defined as A1 · A1 which identifies all the states we
on the shelf. can reach in two steps. Table 3 shows a portion of A2.

For our analysis we will assume the library has only one Table 3 - Portion of A2
patron and two books. We now pair the transitions from GOO GOC GOD
the patron and book objects that can occur together. An GOO (1,21)(26) (1,21)(23)
‘X’ indicates we are not concerned about the state of the GOC (2,22)(25) (26)(26) (26)(23)
object.
GOD
GOX → GCX GOR (25)(1,21)
patron checks out a book GCO (2,12)(15)
GXO → GXC
GCX → GRX
patron returns a book on time From Table 3 we can observe that it is possible to go from
GXC → GXR
GDX → FRX GOC to GOO by first returning the second book and then
patron returns an overdue book shelving it.
GXD → FXR
TCX → GRX patron with MAX books returns a
TXC → GXR book The closure of A (i.e. A*) for this model occurs after 5
multiplications of A1. Since A* will give all of the
TDX → FRX patron with MAX books returns an
possible sequences through the combined state diagrams,
TXD → FXR overdue book
we would expect the cells for the unreachable states to be
GCC → TCC patron attempts to check out MAX +
empty.
1 books
For this model the unreachable states include two sets.
The following transitions can occur independent of the The first set includes the states where the patron is in T
states of the other objects and one of the two books is in O or R. Clearly the patron
can not have MAX books checked out if one of the books
F→G patron pays fine is not checked out. The other set of unreachable states
C→D book becomes overdue occurs when the patron is in F and one book is in C and
C→C book remains checked out the other is in D or both books are in C or D. In order for
R→O book is re-shelved the patron to be in F one of the two books would have
had to have been returned. An analysis of A* for this
The initial transition matrix A1 has column and row model shows that the columns for these unreachable
headings with triples representing the states of the three states are empty.
objects. For this model there are 3 · 4 · 4 = 48
combinations of the three objects. Table 2 shows a Some of the faults in the design of the library example
portion of the initial transition matrix A1. can be discovered by simply analyzing the transition
matrix. One such fault was a missing transition. From If a user has no preference of the state of a particular
FRC and FCR there is not valid single-step transition to object in a composite state, then an x can be used to
FRR. This means that if one book is returned late, the denote “don’t care.” For example, to show the effect of
patron goes to F status and can not return the other book calling putOnShelf on the first book, which has previously
until they pay the fine. been returned, the user input line would read as:

The remainder of the analysis of the library model will be x, Returned, x > putOnShelf > x, On Shelf, x
done using the tool we have developed which implements
our consistency checking approach. Here, the states of the patron and the second book will
remain the same, but the first book will go from R to O.
5. UML Design Analysis Tool
6. Results From the Library Example
The UML Design Analysis Tool is a tool to check
consistency between diagram sets within a single Rational We now present the output we obtained after running the
Rose Project and a Transition Set that is user defined. tool on several sequence diagrams developed for the
These XMI files are created from Rational Rose via XMI library example.
export. The file contains an XMI representation of all
models created within the project set. The Transition Set
is a text based file created by the user and is documented
in a later section. These files are parsed into an
appropriate data structure for validation based
comparisons. A set of results is generated in the form of
error messages and are displayed in text upon program
completion.

The Transition Matrix is constructed using matrix

arithmetic on the string values of transitions to populate
the transitive closure. Multiplication is a logical AND on
the string contents and addition is a logical OR.

Although many of the individual transitions are obvious,

we have not yet automated the identification of these
transitions. Instead, the user supplies the Transition Set
file which is used to complete transition matrix. Figure 4 Figure 5 - Sequence for returning overdue book
shows the Transition Set file for the library example:

Figure 4 – Transition Set file

Figure 6 - Tool output for Figure 5 sequence diagram
For the library example, an entry in the Transition Set file
has the format: The first sequence diagram (Figure 5) is checking out two
books and returning an overdue book. The tool’s output
Initial state, Initial state, Initial state > Transition(s) > (Figure 6) shows that comparing the class diagram with
Final state, Final state, Final state the sequence diagram found that there was no association
between patron and book. Additionally, it found that the
Each state triplet represents a composite state of the three sequence in the figure 5 was not legal since a check
objects. By taking the transitions listed, the model will action on the book must occur before a book becomes
transition from the initial composite state to the final overdue.
composite state.
 8. Related Work
There are several different approaches that have been
proposed to perform consistency checking between UML
diagrams.

Boris Litvak et al. [6] presents an algorithmic approach to

consistency checking between UML Sequence and State
diagrams. They created the BVUML (Behavioral
Validator of UML) tool which automates the behavioral
validation process. Their approach associates states with
the Sequence diagrams object lifeline so a single run of
the tool validates consistency for only one object.
Therefore the tool must be run multiple times in order to
check the consistency of an entire sequence diagram.

Orest Pilskalns et al. [8] presents an approach that

combines structural and behavioral UML representations
in order to derive and execute test cases to validate a
UML model. They develop a method for encapsulating
Figure 7 - A corrected sequence for overdue book the behavioral aspects (i.e. message paths between
objects) that exists in sequence diagrams into a directed
acyclic graph. The objects in the graph are then
associated with class attribute/parameter values which are
used to generate and execute test cases. Their approach
would require OCL object constraints to be written.

Krishnan [5] describes a framework in which UML

diagrams can be formalized to perform consistency
checking. UML diagrams are translated into
specifications of the theorem proving tool PVS (Prototype
Verification System). The PVS is a language that allows
for the introduction of abstract data types, functions etc.
Figure 8 - Tool output for Figure 7 sequence diagram To check for consistency between sequence and class
diagrams, the class diagrams must first be annotated with
Figure 8 shows that the tool correctly identified the OCL constraints. The PVS will check if the sequence of
sequences in Figure 7 as a correct set of sequences. states described in the sequence diagram can be obtained
from the class diagrams. Custom traces (i.e. sequence of
7. Screen Snapshots states) can also be supplied by the user to check if other
properties hold.

Soon-Kyeong Kim and David Carrington [4] describes

how consistency checking between different UML
models can be accomplished by using a formal object-
oriented metamodeling approach. They formally define
the abstract syntax and semantics of the UML model
using Object-Z as a metalanguage. They then define
consistency constraints that logically exist between
semantically equivalent elements in the metamodel but
are not defined in the current UML metamodel structure.
Once the consistency constraints have been defined for
each of the UML model elements, consistency checking
between different model elements can be achieved by
verifying that the combined models preserve all of the
consistency constraints for the individual model elements.
 Alexander Egyed [3] presents a transformation-based 10. References
approach to consistency checking. They define a set of
model transformation rules to enable the conversion of [1] Bontemps, Y.; Heymans, P.; Schobbens, P.-Y. “From
one UML diagram into another. They also define a set of Live Sequence Charts to State Machines and Back: A
comparison rules to compare the transformed diagram Guided Tour”. IEEE Transactions on Software
with an existing one of the same type. For example, to Engineering, 31(12): 999--1014, Dec. 2005.
check for inconsistencies between a sequence diagram
and a class diagram, they first transform the sequence [2] Yves Dumond, Didier Girardet , Flavio Oquendo. “A
diagram into an interpreted class diagram. The interpreted relationship between sequence and statechart diagrams”.
class diagram is then compared with the existing class Dynamic Behaviour in UML Models: Semantic
diagram. Questions, UML 2000 Workshop.
Yves Dumond et al. [2] shows that it is possible to [3] Alexander Egyed. “Scalable Consistency Checking
integrate semi-formal and formal methods for the between Diagrams – The ViewIntegra Approach”.
dynamic behavior of the UML models. The objective is to Proceedings of the 16th Annual International Conference
favour the integration of formal techniques in the actual on Automated Software Engineering, pages 387--390,
practice of software engineering. They introduce an 2001.
approach to formalize sequence diagrams and verify
coherence with the statechart diagrams. The approach [4] Soon-Kyeong Kim and David Carrington. “A Formal
translates the UML sequence diagrams into the pi- Object-Oriented Approach to defining Consistency
calculus, by preserving the object paradigms. The Constraints for UML Models”. Proceedings of the 2004
consistency between sequence diagrams and statechart Australian Software Engineering Conference, pages 87--
diagrams can be checked by verifying that the messages 94, 2004.
in the sequence diagrams trigger states in statechart
diagrams. [5] Padmanabhan Krishnan. “Consistency Checks for
UML”. Proceedings of the Seventh Asia-Pacific
9. Conclusions and Future Work Software Engineering Conference, pages 162--169, 2000.

In this paper we have shown how consistency checking [6] Boris Litvak, Shmuel Tyszberowics, and
between state and sequence diagrams can be AmiramYehudai. “Behavioral Consistency Validation
accomplished through the use of a transition matrix for a of UML Diagrams”. Proceedings of the First
vector of states representing the state of more than just International Conference on Software Engineering and
one instance object. The closure of this transition matrix Formal Methods, pages 118--125, 2003.
can be used to identify reachable and unreachable states,
identify which transitions are valid, to validate class [7] OMG Unified Modeling Language Specification,
diagrams, and to validate sequence diagrams. Version 1.5, Object Management Group, 2003,
http://www.uml.org
We also introduced the UML Design Analysis Tool
which implements our consistency checking [8] Orest Pilskalns, Anneliese Andrews, Sudipto Ghosh,
methodology. We have shown how our methodology and Robert France. “Rigorous Testing by Merging
accurately identified the sequence diagrams which were Structural and Behavioral UML Representations”. UML
inconsistent with the state diagrams. 2003 - The Unified Modeling Language: Modeling
Languages and Applications, 2863: 234--248, 2003.
Further work needs to be done to show that these
techniques can be used on more complex State Diagrams
such as those that include concurrency and nested states.
We are also working on automating the process of
identifying the transition pairs.

View publication stats