Fig. 1. Implementation variations of a 1-n association from Queue to Slot.

Fig. 2. A summary of our approach.

Fig. 3. Class Queue and ChatServer.

Fig. 4. The inter-variable relationships for the code example in Fig. 3.

Fig. 5. The construction of the abstract heap structure using alias analysis for the code example in Fig. 3.

Fig. 6. Abstract syntax of Java bytecode constructs.

Fig. 7. Part of the 1-n associations reverse-engineered from the benchmarks.

Extraction rules for Java bytecode constructs

The “load-pop” column shows load or pop commands in a sequence from the top to the bottom. The “store-push” column gives store or push commands in a sequence from the top to the bottom. The “Inter-variable relationships” column shows inter-variable relationships extracted from each bytecode construct assuming that the program-point is p. “none” means that there are no corresponding commands or relationships and “…” means that the corresponding commands vary and so are omitted.

General statistics for the benchmarks

#classfiles, “#bytecode instrs”, #methods, and #fields are the numbers of classfiles, bytecode instructions, methods, and fields in the benchmarks. “#container fields” is the number of fields that are declared with container types such as Vector or Hashtable.

Accuracy and performance of our reverse-engineering mechanism

“#container fields” is the number of fields that are declared with container types. “#all 1-n” is the number of the relevant 1-n associations that are implemented via containers. “#found 1-n” is the number of 1-n associations inferred by our approach. 1-n associations implemented by one-dimensional containers and two-dimensional containers are added to the number. %recall is the recall ratio and %precision is the precision ratio. The last column gives the time taken for the inference in seconds.

Comparison (recall and precision)

“>”, “<” indicate “more than” and “less than”. “?” indicates that there is no way to determine the exact number because the tool described by Jackson’s method could only produce the result for part of AWT.

Comparison (analysis time)

“#classes analyzed” is the sum of the number of the classes analyzed by each method. “all analysis time” is the time taken for analyzing all the classes. “average analysis time” is the average analysis time for each class analyzed.

Statistics for the shared usage of containers

#containers gives the number of rep objects that refer to other rep objects. “#containers w/field” is the number of containers that are referred to by fields. “#shared containers” represents the number of containers that are shared by different fields.

The effect of iterators

#containers gives the number of rep objects that refer to other rep objects. “#containers w/iterators” is the number of containers whose elements were extracted via iterators. “#found 1-n w/iterators” gives the number of 1-n associations inferred by iterators as well as containers and %ratio gives its percentage.

Results for nested containers

“#nested containers” gives the number of containers that have other container as their elements. “element containers” is the number of such elements. #elements is the number of rep objects that are referred to by element containers. “#found 1-n” gives the number of the 1-n associations implemented by nested containers.

Comparison (characteristics)

“o” means that a method handles the corresponding issue while “x” means that a method does not handle it.