COHERENT DEPENDENCE CLUSTERS 1 Syed Islam AGENDA Coherent

COHERENT DEPENDENCE CLUSTERS 1 Syed Islam

AGENDA Coherent Clusters (Specialized Dependence Cluster) Visualization Tool for Dependence Clusters Mapping Source Code Constructs to Clusters Overlapping Clusters Program Comprehension / Maintenance Fault Prediction 2

DEPENDENCE CLUSTER A dependence cluster is a maximal set of program statements where each statement depends on all other. Minimal Slice 3

DEPENDENCE CLUSTER (SLICE-BASED) Two statements that mutually depend upon each other must be in each others slice. i j A Slice-Based Dependence Clusters is a maximal set of program statements all of which are in each others slice. 4

DEPENDENCE CLUSTER. . Remembering all the slices is too expensive. Approximation techniques were used Same Slice Cluster Same Slice-Size Cluster. 5

MSG -MONOTONE SLICE-SIZE GRAPH Calculate the Slice Sizes for each SDG vertex of a program. Plot all slice sizes of the program in one graph in monotonically increasing order of the sizes. X axis: percentage of slices represented Y axis: normalised slice size 6

COHERENT CLUSTERS Same-Slice Dependence cluster have internal and external requirements. Statements of Clusters constructed from Backward Slices are all influenced by the same set of statements. (Backward-Slice Cluster) Statements of Clusters constructed from Forward Slices all influence the same set of statements. (Forward-Slice Cluster). A Coherent Cluster is a set of statements where each statement has the same backward slice and the 7 same forward slice.

Dependence Cluster Forward-Slice Backward-Slice Coherent Cluster WHAT FORMS A CLUSTER? x 1 2 5 3 4 y 8

SCG – SLICE/CLUSTER-SIZE GRAPH 9

SCG – SLICE/CLUSTER-SIZE GRAPH B-MSG F-MSG 10

SCG – SLICE/CLUSTER-SIZE GRAPH B-MCG F-MCG 11

SCG – SLICE/CLUSTER-SIZE GRAPH Coherent Clusters 12

RESULTS COHERENT CLUSTERS 13

14

FUNCTION MAPPING TO CLUSTER Tool Maps: source lines and files to clusters. functions represent particular computation mapping functions to clusters. Initial Data: Multiple functions form a cluster. Multiple clusters within the same function. 15

APPROXIMATION CONSTRAINTS Calculate percentage of pair of nodes whose slices are the same; to that of where the nodes in the pair are in each others slice. x and y, where nodes x and y are nodes of a pdg | { x, y : S(x) = S(y) }| | { x, y : x S(y) y S(x) }| S() – Backward Slice The algorithm runs in T(n) = O(n 3) hence 30 days BC – The results is 60%. 16

COHERENT CLUSTER DEPENDENCE GRAPH 1 2 5 3 4 What does this mean? 17 17

COHERENT CLUSTER DEPENDENCE GRAPH. . 1 2 5 3 4 Does this mean we are looking at a dependence cluster (or. . MDS)? 18

70. 43% 19

DAGSTUHL SEMINAR BEYOND PROGRAM SLICING Hypothesis 1: short program slices have fewer faults Hypothesis 2: code common to many program slices has fewer faults Hypothesis 3: cliff faces in dependence clusters indicate faults 20

21

ONGOING WORK Complete implementation of the tool Mapping Source Code Constructs to Clusters Combining Cluster – Larger Clusters Longitudinal study Faults Metrics 22