Defining eos//0 portably.

[bakaq]

I was revisiting #2260 to try to adapt it into the Scryer libraries and I wondered how I could implement eos//0 (end of string) in a portable way. The usual way you would do it is like this:

eos --> call(eos_raw).
eos_raw([], []).

But from what I remember of the discussion there, call//1 doesn't actually guarantee anything about the meaning of the two arguments, so they may not represent a difference list and this definition breaks. There are now the new phrase//[2,3] predicates that may help with this, but I'm getting confused all over again and I'm not sure how to implement eos//0 with them. Is this a predicate that can't be defined portably using call//1 and phrase//[2,3]?

[bakaq]

Oh, I think I got it:

eos_new --> call(eos_new_raw).
eos_new_helper(S0, S, S0, S).
eos_new_raw(S0, S) :-
    phrase(eos_new_helper(S0, S), [], []).

Because this uses phrase/3 to get the difference list, it is portable. From some quick tests this seems to work the same as the naive eos//0 implementation. Maybe this could be cleaned up a bit:

% Gets the difference list at the given point in the grammar rule body
diff_list(Ls0, Ls) --> call(call(diff_list_raw, Ls0, Ls)).
diff_list_helper(S0, S, S0, S).
diff_list_raw(Ls0, Ls, S0, S) :-
    phrase(diff_list_helper(S0, S), Ls0, Ls).

eos_new --> diff_list([], []).

[bakaq]

Even though this is not necessary because apparently call//1 does indeed give the list difference, I think this is a cool way to implement call//1 if a system doesn't have support for it for some reason.

[triska]

The original definition you provided seems to be perfectly valid, using the non-terminal call//1 to invoke a predicate from within a DCG in such a way that the predicate can reason about the list difference.

The discussion in #2260 was about a different issue, where the predicate call/3 was originally used to invoke a non-terminal, and that is not portable. Use the predicates phrase/N, and the non-terminals phrase//N, to invoke DCG non-terminals.

[bakaq]

Oh, I knew I was confusing things again. So, to be completely clear, call//1 is guaranteed to provide a list difference in the order Complete-Tail to its argument? The only guarantee the DCG working draft seems to give is that "phrase(call(G_2), S0,S) is true iff phrase(G_2, S0,S)", which seems to imply that the arguments are the same as the underlying implementation of non-terminals.

[triska]

This is a mistake in the draft, it should read: "... is true iff call(G_2, S0,S)"

[bakaq]

Ok, I'm going to trust you here. I also agree that this behavior makes much more sense, because with the one I described call(G_2) and G_2 do literally the same thing and this makes call//1 pretty redundant.

[UWN]

Also recall that many implementations that do not explicitly handle call//1 will translate it into call/3 and thus, say, ?- phrase(call([]),[]). will in many implementations result in an existence_error(procedure,[]/2). except for those (currently YAP) that provide a predicate for each grammar control construct. Added as #32.