Exactly, they are hints to tell the matching engine not to bother backtracking. Hypothetically the optimiser should be able to determine them all automatically and you shouldnt need them, but its a lot easier to let people do it themselves in terms of implementation.

Auto-possessiveification is something that sure does need doing. If you consider the general case XqY, where X and Y are literals and q is a quantifer, you should be able to do Xq+Y whenever X cannot overlap Y. Ive not got round to it yet tho.