Path-Aware Static Program Analyses for Specification Mining* - Mining Software Specifications: Methodologies and Applications - page 329

TABLE 9.2:

Transactions Associated with Calls to readbuf Shown in

Figure 9.8

Variables Transactions

c1 c2

packets f(>, 0)g

f(>, 0)g

packflag f(:=, true )g
size f(:=, MAXSIZE ), f(:=, MINSIZE ),
(arg(1), allocbuf ), (arg(1), allocbuf ),
(arg(2), readbuf )g (arg(2), readbuf )g
buf f(:=,res( allocbuf )), f(:=,res( allocbuf )), (6=, 0),
(arg(1), readbuf )g (arg(1), lock ),
(arg(1), readbuf )g

f(=, 0), (:=,res( lock ))g

l

i
Variables Transactions
c3 c4
packets f(>, 0)g f(>, 0)g
packflag f(:=,true)g
size f(:=, MINSIZE ), f(:=, MAXSIZE ),
(arg(1), allocbuf ), (arg(1), allocbuf ),
(arg(2), readbuf )g (arg(2), readbuf )g
buf f(:=,res( allocbuf )), (6=, 0), f(:=,res( allocbuf )), (6=, 0),
(arg(1), lock ), (arg(1), lock ),

(arg(1), readbuf )g (arg(1), readbuf )g

l f(=, 0), (:=,res( lock ))g

f(=, 0), (:=,res(lock))g

if f(:= 0)g

f(:= 0)g

size : f(arg(1), allocbuf ), (arg(2), readbuf )g
l : f(=,0), (:=,res( lock ))g
buf : f(arg(1), lock ), (arg(1), readbuf ),
(6=,0), (:=,res( allocbuf ))g
Depending upon on the level of precision required by the user, the above
mining technique can be easily translated into the more restrictive intersection
technique by simply fixing n to be the total number of call-sites (confidence
= 100%).
Sequence Mining. For control-flow predicates, frequent itemset mining
does not suce since the order of elements in the transaction is not con-
sidered. For deriving precedence relations [37], we use sequence mining [2]. A
sequence mining algorithm takes as input a set of sequences (I), a user-defined
confidence threshold, and outputs a set (S) of sequences that occur as subse-
quences in a minimum fraction (as specified by the confidence threshold) of
input sequences. Observe that if a subsequence s is frequently occurring, all
subsequences of s also occur at least as frequently as s. Therefore, we consider