To create, manage and compare specifications.
- §1. Kinds vs specifications
- §7. Pretty-printing specifications
- §8. Sorting
- §10. Nothingness
- §11. The Unknown
- §12. The evaluation internal test
§1. Kinds vs specifications. Specifications are portions of the parse tree representing data, or conditions on data, or ways to manipulate it. This chapter contains utility routines for creating and using them.
Kinds can be faithfully represented as specifications, but not vice versa:
define VALUE_TO_KIND_FUNCTION Specifications::to_kind define RVALUE_TO_KIND_FUNCTION Specifications::rvalue_to_kind
parse_node *Specifications::from_kind(kind *K) { return Descriptions::from_kind(K, FALSE); } kind *Specifications::to_kind(parse_node *spec) { if (Node::is(spec, AMBIGUITY_NT)) spec = spec->down; if (Specifications::is_description(spec)) return Descriptions::to_kind(spec); if (Lvalues::is_lvalue(spec)) return Lvalues::to_kind(spec); else if (Rvalues::is_rvalue(spec)) return Rvalues::to_kind(spec); return NULL; } kind *Specifications::rvalue_to_kind(parse_node *spec) { if (Rvalues::is_rvalue(spec)) return Rvalues::to_kind(spec); return NULL; } kind *Specifications::to_true_kind(parse_node *spec) { if (Lvalues::is_lvalue(spec)) return Lvalues::to_kind(spec); else if (Rvalues::is_rvalue(spec)) return Rvalues::to_kind(spec); return NULL; } kind *Specifications::to_true_kind_disambiguated(parse_node *spec) { if (Node::is(spec, AMBIGUITY_NT)) spec = spec->down; if (Node::is(spec, TEST_VALUE_NT)) spec = spec->down; if (Lvalues::is_lvalue(spec)) return Lvalues::to_kind(spec); else if (Rvalues::is_rvalue(spec)) return Rvalues::to_kind(spec); return NULL; } int Specifications::is_value(parse_node *spec) { node_type_metadata *metadata = NodeType::get_metadata(Node::get_type(spec)); if ((metadata) && ((metadata->category == LVALUE_NCAT) || (metadata->category == RVALUE_NCAT))) return TRUE; return FALSE; } int Specifications::is_condition(parse_node *spec) { node_type_metadata *metadata = NodeType::get_metadata(Node::get_type(spec)); if ((metadata) && (metadata->category == COND_NCAT)) return TRUE; return FALSE; } int Specifications::is_phrasal(parse_node *spec) { if (NodeType::has_flag(Node::get_type(spec), PHRASAL_NFLAG)) return TRUE; return FALSE; }
§2. We say that a specification is "kind-like" if it's one of those which could result from Specifications::from_kind: for example, the description "numbers" is kind-like, but "even numbers" is not.
int Specifications::is_kind_like(parse_node *spec) { if (Descriptions::is_kind_like(spec)) return TRUE; return FALSE; }
§3. Another useful characterisation is "description-like". Most nouns are not description-like, but names of objects are. Thus "Mrs Jones" is description-like, but "12" is not.
define NP_IS_DESCRIPTIVE Specifications::is_description_like
int Specifications::is_description_like(parse_node *p) { int g = FALSE; if (Specifications::is_description(p)) g = TRUE; if (Rvalues::is_object(p)) g = TRUE; if (Rvalues::is_nothing_object_constant(p)) g = FALSE; return g; } int Specifications::is_description(parse_node *p) { if ((Node::is(p, TEST_VALUE_NT)) && (Rvalues::is_CONSTANT_construction(p->down, CON_description))) return TRUE; return FALSE; }
define NP_TO_PROPOSITION Specifications::to_proposition
pcalc_prop *Specifications::to_proposition(parse_node *p) { if (p == NULL) return NULL; if (Specifications::is_description(p)) return Descriptions::to_proposition(p); return Node::get_proposition(p); } inference_subject *Specifications::to_subject(parse_node *spec) { inference_subject *infs = NULL; pcalc_prop *prop = Specifications::to_proposition(spec); if (prop) { parse_node *val = Propositions::describes_value(prop); if (val) { infs = InferenceSubjects::from_specification(val); } else { kind *K = Binding::kind_of_variable_0(prop); if (Kinds::Behaviour::is_subkind_of_object(K) == FALSE) K = K_object; infs = KindSubjects::from_kind(K); } } else infs = InferenceSubjects::from_specification(spec); return infs; }
§5. Specifications which talk about objects lie in two different families: "Mrs Jones" is a CONSTANT with kind "object", but "the open Bronze Gateway" is a DESCRIPTION. The following extracts the object, if any, from either case:
instance *Specifications::object_exactly_described_if_any(parse_node *spec) { if (spec == NULL) return NULL; if (Specifications::is_description(spec)) return Descriptions::to_instance(spec); if (Rvalues::is_object(spec)) { if (Rvalues::is_nothing_object_constant(spec)) return NULL; return Rvalues::to_instance(spec); } return NULL; }
§6. It's convenient to use specifications to represent the requirement on a new variable declaration, such as that in "The tally is a number that varies." We do this with the kind-like specification for "K that varies". Such a node is called "new-variable-like".
parse_node *Specifications::new_new_variable_like(kind *K) { K = Kinds::unary_con(CON_variable, K); parse_node *spec = Specifications::from_kind(K); return spec; } kind *Specifications::kind_of_new_variable_like(parse_node *S) { kind *K = Specifications::to_kind(S); return Kinds::unary_construction_material(K); } int Specifications::is_new_variable_like(parse_node *spec) { if ((Specifications::is_kind_like(spec)) && (Kinds::get_construct(Specifications::to_kind(spec)) == CON_variable)) return TRUE; return FALSE; }
§7. Pretty-printing specifications. We need to be able to print legible forms of translations in order to produce good error messages, and also in order to describe phrases in the Index; those have to be English language forms.
void Specifications::write_out_in_English(OUTPUT_STREAM, parse_node *spec) { if (spec == NULL) WRITE("something unknown"); else if (Lvalues::is_lvalue(spec)) Lvalues::write_out_in_English(OUT, spec); else if (Rvalues::is_rvalue(spec)) Rvalues::write_out_in_English(OUT, spec); else if (Specifications::is_condition(spec)) Conditions::write_out_in_English(OUT, spec); else if (Node::is(spec, AMBIGUITY_NT)) Specifications::write_out_in_English(OUT, spec->down); else WRITE("something unrecognised"); }
§8. Sorting. Some specifications are used to describe the applicability of rules and phrases, and since those must be sorted in order of how specific they are, we will need a way of telling when one specification is more specific than another. For instance, "Will Parker in the vineyard" beats "Will Parker" beats "a man" beats "a person" beats "an object" beats "a value".
A sequence of "laws" is used to decide this — or at least, that's how we describe the outcome to the Index of a project. We simply remember the most recent law applied, because then when a decision finishes, it must have been the decisive one:
int cco = 0; comparison count: used to make the debugging log vaguely searchable text_stream *c_s_stage_law = NULL; name of the law being applied, which caused this to be called void Specifications::law(text_stream *L) { c_s_stage_law = L; } text_stream *Specifications::law_applied(void) { return c_s_stage_law; }
§9. The following is one of Inform's standardised comparison routines, which takes a pair of objects A, B and returns 1 if A makes a more specific description than B, 0 if they seem equally specific, or \(-1\) if B makes a more specific description than A. This is transitive, and intended to be used in sorting algorithms.
int Specifications::compare_specificity(parse_node *spec1, parse_node *spec2, int *wont_mix) { LOGIF(SPECIFICITIES, "Law %S (test %d): comparing $P with $P\n", Specifications::law_applied(), cco++, spec1, spec2); Existence is itself something specific9.1; int a = Specifications::is_description(spec1), b = Specifications::is_description(spec2); instance *I1 = Specifications::object_exactly_described_if_any(spec1); instance *I2 = Specifications::object_exactly_described_if_any(spec2); An actual specification is more specific than a generic one9.2; An exact object is more specific than a vague one9.3; if (I1) Enclosing regions beat enclosed ones, and regions beat rooms9.4; if (wont_mix) If one matches the other, but not vice versa, it must be more specific9.5; if ((a == TRUE) && (b == TRUE)) { case 1: both are descriptions return Descriptions::compare_specificity(spec1, spec2); } else if ((a == FALSE) && (b == FALSE)) { case 2: neither is a description Table entries are more specific than other non-descriptions9.6; } else { case 3: one is a description, the other isn't When is a description more specific than a non-description?9.7; } return 0; }
§9.1. Whether or not, as Bertrand Russell thought in 1894, existence is itself a good ("Great God in Boots! — the ontological argument is sound!"), a specification which exists is certainly more significant than one which does not; and there is nothing to choose between two specifications, neither of which exists.
A God who wears boots is an incongruous thought. Sandals, possibly. But maybe Russell meant the local Cambridge branch of Boots, the chemist's shop. At any rate he changed his mind in 1896.
Existence is itself something specific9.1 =
if ((spec1 == NULL) && (spec2 != NULL)) return -1; if ((spec1 != NULL) && (spec2 == NULL)) return 1; if ((spec1 == NULL) && (spec2 == NULL)) return 0;
- This code is used in §9.
§9.2. Hard to argue with this one: "34", for instance, is more specific than "a number". (We might quibble about whether or not "a number which equals 34" is really less specific than "34" — Inform says it is; but in fact it doesn't much matter either way.)
An actual specification is more specific than a generic one9.2 =
int aa = TRUE; if ((a) && (I1 == NULL)) aa = FALSE; int ba = TRUE; if ((b) && (I2 == NULL)) ba = FALSE; if ((aa) && (!ba)) return 1; if ((!aa) && (ba)) return -1;
- This code is used in §9.
§9.3. For instance, "the open Marble Door" is more specific than "an open door" or even just "a door", even though the latter is linguistically simpler.
An exact object is more specific than a vague one9.3 =
if ((I1) && (I2 == NULL)) return 1; if ((I1 == NULL) && (I2)) return -1;
- This code is used in §9.
§9.4. Suppose both specifications exactly describe objects, I1 and I2, and all other considerations are equal. It's not quite true that one object is as good as another, because a region can be thought of as a set of rooms. (Without the following criterion, rules such as "After waiting in Russia" and "After waiting in Vladivostok Railway Station Waiting Room" would have equal status.)
Enclosing regions beat enclosed ones, and regions beat rooms9.4 =
LOGIF(SPECIFICITIES, "Test %d: Comparing specificity of instances $O and $O\n", cco, I1, I2); int pref = PluginCalls::more_specific(I1, I2); if (pref != 0) return pref;
- This code is used in §9.
§9.5. If one matches the other, but not vice versa, it must be more specific9.5 =
int ev1 = ((a) || (Specifications::is_value(spec1))); int ev2 = ((b) || (Specifications::is_value(spec2))); if (Lvalues::get_storage_form(spec1) == PROPERTY_VALUE_NT) ev1 = FALSE; if (Lvalues::get_storage_form(spec2) == PROPERTY_VALUE_NT) ev2 = FALSE; if ((ev1) && (ev2)) { int x = Dash::compatible_with_description(spec1, spec2); int y = Dash::compatible_with_description(spec2, spec1); if (x == ALWAYS_MATCH) { if (y == ALWAYS_MATCH) return 0; else return 1; } if (y == ALWAYS_MATCH) { if (x == ALWAYS_MATCH) return 0; else return -1; } if (wont_mix) *wont_mix = TRUE; }
- This code is used in §9.
§9.6. Here neither specification is a description or an actual object. For the most part, then, we're left with two specifications of about equal merit, and we don't choose between them. But in one case we do intervene: a table entry reference beats anything else left, so that "Instead of taking the magic key corresponding to Merlin in the Table of Arcana" is more specific than "Instead of taking the brass key". It's questionable whether this is a good convention, but users reported the previous absence of such a convention as a bug, which is usually telling.
Table entries are more specific than other non-descriptions9.6 =
int t1 = Node::is(spec1, TABLE_ENTRY_NT); int t2 = Node::is(spec2, TABLE_ENTRY_NT); if ((t1 == TRUE) && (t2 == FALSE)) return 1; if ((t1 == FALSE) && (t2 == TRUE)) return -1;
- This code is used in §9.
§9.7. To explicate the following: a description of an exact object beats any non-description — thus "the open Marble Door" (a description) beats "the Marble Door" (a constant instance). But any non-description beats a description which is vague about the object — thus "the Marble Door" beats "an open door", which is not news since rules above would enact that anyway, but also "the tallest door in the Castle" (a phrase) beats "an open door".
When is a description more specific than a non-description?9.7 =
if (I1) { if (a == TRUE) return 1; else return -1; } else { if (a == TRUE) return -1; else return 1; }
- This code is used in §9.
define DETECT_NOTHING_CALCULUS_CALLBACK Rvalues::is_nothing_object_constant define PRODUCE_NOTHING_VALUE_CALCULUS_CALLBACK Specifications::nothing
parse_node *Specifications::nothing(void) { return Lvalues::new_actual_NONLOCAL_VARIABLE( NonlocalVariables::nothing_pseudo_variable()); }
§11. The Unknown. We begin with s-nodes used to represent text not yet parsed, or for which no meaning could be found.
define UNKNOWN_PREFORM_RESULT_SYNTAX_CALLBACK Specifications::new_UNKNOWN
parse_node *Specifications::new_UNKNOWN(wording W) { return Node::new_with_words(UNKNOWN_NT, W); }
§12. The evaluation internal test.
void Specifications::perform_evaluation_internal_test(OUTPUT_STREAM, struct internal_test_case *itc) { parse_node *spec = NULL; if (<s-value>(itc->text_supplying_the_case)) spec = <<rp>>; else spec = Specifications::new_UNKNOWN(itc->text_supplying_the_case); Dash::check_value(spec, NULL); kind *K = Specifications::to_kind(spec); WRITE("Kind of value: %u", K); if (Kinds::Behaviour::is_quasinumerical(K)) WRITE(" scaled at k=%d", Kinds::Behaviour::scale_factor(K)); WRITE("\n"); }