ImpossibleConsistency.lf

/**
 * In this pattern, there are no finite STA offsets that enable each Update reactor to advance its tag.
 * A positive value that is the same for both federates won't work because it will delay the outputs by
 * the same amount, so waiting that amount offers no information about the remote tag. Asymmetric values
 * won't work either because the federate with the smaller STA will potentially prematurely advance its tag.
 * This situation is described in the [CAL theorem](https://dl.acm.org/doi/10.1145/3609119) paper, where
 * the processing offsets go to infinity.
 *
 * Note that the centralized coordinator works fine for this case. Each federate periodically notifies
 * the RTI of the advancement of its _physical_ time without advancing its _logical_ time (something that
 * cannot be accomplished with null messages sent based on a timer).
 *
 * @author Edward A. Lee
 */
target C {
  coordination: decentralized
}
reactor Update {
  input remote: int
  output out: int
  state balance: int = 0
  physical action p
  reaction(startup) -> p {= =}
  reaction(p) -> out {=
  =}
  reaction(remote) {=
    self->balance += remote->value;
  =}
}
federated reactor {
  u1 = new Update()
  u2 = new Update()
  u1.out -> u2.remote
  u2.out -> u1.remote
}