SEKE 2012 Proceedings - Knowledge Systems Institute

nique estimate the impact set based on transitive closure on
the call graph [3]. This paper also used the call graph to
represent the system and then computed the impact set on
this representation. But different from traditional call graph
based CIA technique, this paper attempts to compute a more
precise impact set in respect to two factors: setting a stable
state for the impact set computation on the call graph and
considering interference among multiple changed entities
in the change set. The case study on the real-world program
has shown the effectiveness of our approach.
This paper is organized as follows: in the next section,
we introduce our CIA approach. Section 3 gives the experimental
evaluation to show the effectiveness of our CIA technique.
In Section 4, some related work of CIA techniques
are presented. Finally, we conclude our CIA technique and
show some future work in Section 5.
2 Our Approach
Given the change set, CIA is employed to estimate the
impact set of the changes made to the software. In this paper,
CIA is performed based on mining the call graph.
2.1 Basics for CIA
As proposed above, our approach uses call graph to represent
the system and CIA is performed relying on mining
this call graph. In this section, some definitions and some
operations in call graph are given. First, we introduce the
definition of the call graph [8].
Definition 1 (Call Graph) A call graph is a directed graph
G =(V,E). V is a set of vertices representing methods in
the system, and E ⊆ V × V is a set of edges representing
the call relationship between the caller and callee.
In addition, there are two useful definitions to help mining
the call graph: inner-degree and inter-degree of a subgraph
G ′ . They are defined as follows:
Definition 2 (Inner-degree) Given a subgraph G ′
of the
call graph, inner-degree of G ′
is the total number of the
edges for each vertex in G ′ .
Definition 3 (Inter-degree) Given a subgraph G ′
of the
call graph G, inter-degree of G ′ on G is the total number of
edges adjacent to (connected to) the vertices in G − G ′ .
As discussed in the above phenomenon ”water wave ripple”,
there is an important spreading center to propagate
the ripples of the water wave. Similar to this, there is also
a center to propagate the change ripples during the process
of change impact analysis. On the call graph, we call this
center as core, defined as follows:
Definition 4 (Core) Core is a subgraph G ′ of the call
graph G,, which satisfies that the ratio of their innerdegree
over the inter-degree is greater than the threshold
ε.
The threshold ε indicates the cohesion of the relationship
among the methods within the subgraph G ′ . The bigger
the ε is, the higher the cohesion of the relationship among
these methods is. In our CIA approach, it is used to take the
interference among the changed methods in the change set
into account. The ε is at least bigger than 1, which shows
the relationship among the methods within the subgraph is
stronger than connecting to its outer methods.
In graph theory, breadth-first-search (BFS) is widely
used in graph search algorithm, defined as follows:
Definition 5 (BFS) BFS begins from the origin node and
explores all the neighboring nodes. Then for each of those
nearest nodes, it searches their unexplored neighbor nodes,
and so on, until it finds the goal.
This search algorithm is also used for change impact
analysis for searching the ripples of the changes. In the next
section, our CIA approach is presented based on these definitions.
2.2 Change Impact Analysis
In our approach, change set is proposed to be composed
of a set of changed methods, CIA is then performed relying
on mining the call graph, and generates a set of potentially
impacted methods. It includes two steps. First, we identify
the ”center” of multiple changes ripples, which we call
initial impact set. It is expected that the methods in the initial
impact set are less likely to be false-positives. Then, we
compute the changes effects of this center, and obtain the
final impact set. And this set is expected to cover all the
ripples affected by the changes.
As discussed above, most of current CIA techniques
compute the impact set from the individual entities in the
change set, without considering the relationship existed a-
mong them. In practice, multiple changes are performed in
combination to finish a change proposal, hence there exists
the relationship among the changed entities in the source
code. Such a relationship is called the interference in this
paper. We take the interference among the entities in the
change set into account to estimate the impact set. First
we attempt to identify the center of these changes, which
includes the most probably impacted entities potentially affected
by these multiple changes. These entities are collected
into the initial impact set, which is defined as follows:
Definition 6 (Initial Impact Set (IIS)) Initial impact set
is computed using the BFS algorithm from the methods
8