Inferences that a relation holds between two subjects or values.

§1. Relation inferences are made about a relation, and say that two other subjects or values are related by it. Thus, if Charles "knows" Sebastian, this fact is an inference about the knowledge relation, not about either Charles or Sebastian, who are only the terms listed in its data.

inference_family *relation_inf = NULL;

§2.

void RelationInferences::start?Usage of RelationInferences::start:
Knowledge Module - §3(void) {
    relation_inf = Inferences::new_family(I"relation_inf");
    METHOD_ADD(relation_inf, LOG_DETAILS_INF_MTID, RelationInferences::log_details);
    METHOD_ADD(relation_inf, COMPARE_INF_MTID, RelationInferences::cmp);
}

§3. Terms can be given either as subjects or as arbitrary values. This was a late change to Inform, which came in with dynamic relations between (say) numbers, and therefore the need to set up an initial state for those relations.

The terms will either both be subjects, or both be values, so at all times exactly one of these pairs of pointers is NULL.

typedef struct relation_inference_data {
    struct inference_subject *terms_as_subjects[2];
    struct parse_node *terms_as_values[2];
    CLASS_DEFINITION
} relation_inference_data;

inference *RelationInferences::new?Usage of RelationInferences::new:
§4(inference_subject *subj0,
    inference_subject *subj1, parse_node *val0, parse_node *val1) {
    PROTECTED_MODEL_PROCEDURE;
    relation_inference_data *data = CREATE(relation_inference_data);
    data->terms_as_subjects[0] = InferenceSubjects::divert(subj0);
    data->terms_as_subjects[1] = InferenceSubjects::divert(subj1);
    data->terms_as_values[0] = val0;
    data->terms_as_values[1] = val1;
    return Inferences::create_inference(relation_inf,
        STORE_POINTER_relation_inference_data(data), prevailing_mood);
}

• The structure relation_inference_data is private to this section.

§4. As promised, these are drawn using either subjects or values, but not both:

void RelationInferences::draw(binary_predicate *bp,
    inference_subject *subj0, inference_subject *subj1) {
    inference *i = RelationInferences::new(subj0, subj1, NULL, NULL);
    Inferences::join_inference(i, RelationSubjects::from_bp(bp));
}

void RelationInferences::draw_spec(binary_predicate *bp,
    parse_node *val0, parse_node *val1) {
    if ((val0 == NULL) || (val1 == NULL)) internal_error("malformed value relation");
    inference *i = RelationInferences::new(NULL, NULL, val0, val1);
    Inferences::join_inference(i, RelationSubjects::from_bp(bp));
}

§5. And here are the method calls:

void RelationInferences::log_details?Usage of RelationInferences::log_details:
§2(inference_family *f, inference *inf) {
    relation_inference_data *data = RETRIEVE_POINTER_relation_inference_data(inf->data);
    if (data->terms_as_subjects[0]) LOG("-1:$j", data->terms_as_subjects[0]);
    if (data->terms_as_subjects[1]) LOG("-2:$j", data->terms_as_subjects[1]);
    if (data->terms_as_values[0]) LOG("-s1:$P", data->terms_as_values[0]);
    if (data->terms_as_values[1]) LOG("-s2:$P", data->terms_as_values[1]);
}

§6. Note that the topic of a relation inference depends on both terms. If Charles knows Sebastian and Charles also knows Julia, then these two inferences both belong to the knowledge relation, but "differ in topic", so that there is no contradiction. If in fact the knowledge relation wants to make this a one-to-one relationship, it will have to detect the contradiction of Charles knowing both of them elsewhere: the inference machinery can't do so for it.

int RelationInferences::cmp?Usage of RelationInferences::cmp:
§2(inference_family *f, inference *i1, inference *i2) {
    relation_inference_data *data1 = RETRIEVE_POINTER_relation_inference_data(i1->data);
    relation_inference_data *data2 = RETRIEVE_POINTER_relation_inference_data(i2->data);

    int c = Inferences::measure_infs(data1->terms_as_subjects[1]) -
            Inferences::measure_infs(data2->terms_as_subjects[1]);
    if (c > 0) return CI_DIFFER_IN_TOPIC; if (c < 0) return -CI_DIFFER_IN_TOPIC;
    c = Inferences::measure_infs(data1->terms_as_subjects[0]) -
        Inferences::measure_infs(data2->terms_as_subjects[0]);
    if (c > 0) return CI_DIFFER_IN_TOPIC; if (c < 0) return -CI_DIFFER_IN_TOPIC;
    c = Inferences::measure_pn(data1->terms_as_values[1]) -
        Inferences::measure_pn(data2->terms_as_values[1]);
    if (c > 0) return CI_DIFFER_IN_TOPIC; if (c < 0) return -CI_DIFFER_IN_TOPIC;
    c = Inferences::measure_pn(data1->terms_as_values[0]) -
        Inferences::measure_pn(data2->terms_as_values[0]);
    if (c > 0) return CI_DIFFER_IN_TOPIC; if (c < 0) return -CI_DIFFER_IN_TOPIC;

    c = Inferences::measure_inf(i1) - Inferences::measure_inf(i2);
    if (c > 0) return CI_DIFFER_IN_COPY_ONLY; if (c < 0) return -CI_DIFFER_IN_COPY_ONLY;
    return CI_IDENTICAL;
}

§7. And finally two access functions:

void RelationInferences::get_term_subjects?Usage of RelationInferences::get_term_subjects:
Relation Subjects - §3
The Model World - §4.3(inference *i,
    inference_subject **subj1, inference_subject **infs2) {
    if ((i == NULL) || (i->family != relation_inf)) internal_error("not a relation inf");
    relation_inference_data *data = RETRIEVE_POINTER_relation_inference_data(i->data);
    if (subj1) *subj1 = data->terms_as_subjects[0];
    if (infs2) *infs2 = data->terms_as_subjects[1];
}

void RelationInferences::get_term_specs?Usage of RelationInferences::get_term_specs:
Relation Subjects - §4(inference *i,
    parse_node **val1, parse_node **spec2) {
    if ((i == NULL) || (i->family != relation_inf)) internal_error("not a relation inf");
    relation_inference_data *data = RETRIEVE_POINTER_relation_inference_data(i->data);
    if (val1) *val1 = data->terms_as_values[0];
    if (spec2) *spec2 = data->terms_as_values[1];
}