You seem to be reformulating your assertion. In the last comment you say that "default Prolog SLD resolution is depth-first". That reformulation also doesn't make sense. The search used by Prolog is depth-first search. But SLD resolution is not "depth-first". There is no context in which "depth-first" makes sense outside of depth-first search and there is no context in which SLD resolution can be said to be "depth-first". That's regardless of implementation.
>> General resolution (SLD or otherwise) works for some theorem-proving, but not for running general Prolog programs.
I'm sorry, I don't understand this. If I understand correctly, "general" resolution refers to resolution for arbitrary clauses? SLD resolution operates on definite clauses and is sound and refutation complete for definite logic programs. SLDNF resolution is also sound for normal programs. So, yes, resolution "works" for running general Prolog programs.
Depth-first search is used to find matching literals, but SLD resolution is not depth-first search; or it would be called "depth-first search".
SLD resolution is an algorithm that eliminates literals from a clause until the empty clause is derived, at which point a refutation of the original goal succeeds.
For example, let P = {q(a,b) ∧ (p(X,Y) ∨ ¬q(X,Y))} be a definite (logic) program and G = ¬p(X,Y) be a Horn goal. A refutation proof of ¬p(X,Y) by resolution might go like this:
a) p(X,Y) unifies with ¬p(X,Y) with substitution θ = ∅
b) ¬q(X,Y) ∧ q(a,b) is derived from (a) (by elimination of the two unifying literals with opposing signs)
c) ¬q(X,Y) unifies with q(a,b) with substitution θ = {X/a,Y/b}
d) □ [the empty clause] is derived from (c)
Thus, ¬p(X,Y) is refuted, with substitution {X/a,Y/b}, i.e. p(X,Y) is true with those variable bindings.
Or, as a tree diagram:
Note that in Prolog, P = {q(a,b), p(X,Y):-q(X,Y)} and G = :-p(X,Y). I've written it in clausal form above to help with the elimination of literals.So, depth-first search is used in Prolog (but not mandated by any description of resolution) to find unifying literals for a "query" (i.e. a goal) but depth-first search is not resolution and resolution is not depth-first search. Resolution allows us to derive a set of literals from another set of literals, by elimination of literals that unify but have opposite signs. Indeed, resolution can be notated as:
Or, from p OR q and NOT q we can infer p.[Note: "elimination" is my own terminology, not often used in logic programming books. The correct description is that from p and ¬p we can derive empty clause.]