0'... would give a
syntax error for non-Latin 1 characters. 
cumulative/2 could crash on special cases. 
Since 4.3.0, geost/[2,3,4] could be too conservative in greedy assignment mode
labeling([minimize(X), time_out(1000, F)], [X]) and similar goals
would leave F unbound upon success. 
debug
CLPFD flag.  Now, all constraints are debuggable, and constraint
expansion does not depend on any flags. 
process:process_wait/3 with the timeout/1 option would
raise an exception if called more than once even if the first call
reported timeout.