[#STR-528] dynamic rule intersection inside a fix point operator sometimes goes wrong

Description

Running constant propagation on this Java code:

public class Foo
{
public void foo()
{
int x = 10;

while (something())
{
x = 5;
if (cond())
{
x = 10;
}
else
{}

call(x);
}
}
}

should not propagate x = 5 in the call(x) inside the while, but it does. However, changing the assignment x = 10 in the if (cond()) to x = 1 (or anything other than 10 for that matter) makes constant propagation behave correctly again.

Description

Running constant propagation on this Java code: public class Foo { public void foo() { int x = 10; while (something()) { x = 5; if (cond()) { x = 10; } else {} call(x); } } } should not propagate x = 5 in the call(x) inside the while, but it does. However, changing the assignment x = 10 in the if (cond()) to x = 1 (or anything other than 10 for that matter) makes constant propagation behave correctly again.

Show »

[ Permlink | « Hide ]

Bogdan Dumitriu [2006-03-07 14:17]

This has the same root cause as ~~STR-511~~. Both set- and add-in-rule-scope do not work properly in all cases when multiple change sets are created.

[ Permlink | « Hide ]

Bogdan Dumitriu [2006-04-27 13:09]

Easier code to generate the same problem:

int x = 0;

if (cond()) {
  x = 1;
  if (cond()) {
    x = 0;
    f(x);
  }
}

x = 1 will be propagated in f(x).