🚀 go-pugleaf

The ending point of Seif's message "without changing the semantics of your
program" should be read "without changing the semantics of your algorithm"
to make sense, I guess.

I (and also Seif, implicitly) mentioned the possibility to "implement"
setarg/3 using var/1 and !/0. Perhaps it should be clarified that what
is meant is that you can represent a variable that you wish to modify
during the execution as a partially instantiated structure (an open
list containing all values assigned to the variable). This is not exactly
an "implementation" since you cannot unify two "variables" represented in
this way if they do not share the same instantiation history, but it is a
practical programming technique, that is appropriate in many situations,
if you do not wish to use setarg/3.
The efficiency problem mentioned by Seif should be obvious from this.
The semantic problems seem to be similar to those introduced by viewing 
i.e. [1.2.3|X]-X as the same d-list regardless of what X is bound to. 
In both cases you READ your program in a particular way, and JUSTIFY your 
statement about equality of "the semantics" by some more or less 
well-specified METHODOLOGY for implementing a given algorithm.