Action Name Lists

Home
Inform7
if
Chapter 4: Actions
Action Name Lists

Action name lists are used in parsing action patterns, and identify which action names seem to be possible within them.

§1. Data structure. An ANL is a head object, an action_name_list, which points to a linked list of anl_entry objects.

typedef struct action_name_list {
    struct anl_entry *entries;
    int negation_state;
    int test_this_in_ap_match;
} action_name_list;

action_name_list *ActionNameLists::new_list(anl_entry *first, int state) {
    action_name_list *list = CREATE(action_name_list);
    list->entries = first;
    list->negation_state = state;
    list->test_this_in_ap_match = TRUE;
    return list;
}

The structure action_name_list is private to this section.

§2. The "negation state" of a list is one of three possibilities, not two. The difference is that "doing something other than examining the box", for example, is ANL_NEGATED_LISTWISE, whereas "doing something other than examining to the box" is ANL_NEGATED_ITEMWISE. Note that the state is set irrevocably when the list is created.

define ANL_POSITIVE 1
define ANL_NEGATED_LISTWISE 2
define ANL_NEGATED_ITEMWISE 3

int ActionNameLists::listwise_negated(action_name_list *list) {
    if ((list) && (list->negation_state == ANL_NEGATED_LISTWISE)) return TRUE;
    return FALSE;
}

int ActionNameLists::itemwise_negated(action_name_list *list) {
    if ((list) && (list->negation_state == ANL_NEGATED_ITEMWISE)) return TRUE;
    return FALSE;
}

int ActionNameLists::positive(action_name_list *list) {
    if ((list == NULL) || (list->negation_state == ANL_POSITIVE)) return TRUE;
    return FALSE;
}

§3. It is sometimes possible when matching action patterns to prove that it is unnecessary to test that the action appears in this list; and then the list can be marked accordingly:

void ActionNameLists::suppress_action_testing(action_name_list *list) {
    list->test_this_in_ap_match = FALSE;
}

int ActionNameLists::testing(action_name_list *list) {
    if (list == NULL) return FALSE;
    return list->test_this_in_ap_match;
}

§4. An entry has some book-keeping fields, and is otherwise divided into the item itself — either an action name or a named action pattern — and some parsing data needed by the complicated algorithms for turning text into an action list.

typedef struct anl_entry {
    struct anl_item item;
    struct anl_parsing_data parsing_data;
    int marked_for_deletion;
    struct anl_entry *next_entry;  next in this ANL list
} anl_entry;

anl_entry *ActionNameLists::new_entry_at(wording W) {
    anl_entry *entry = CREATE(anl_entry);
    entry->item = ActionNameLists::new_item();
    int at = -1;
    if (Wordings::nonempty(W)) at = Wordings::first_wn(W);
    entry->parsing_data = ActionNameLists::new_parsing_data(at);
    entry->marked_for_deletion = FALSE;
    return entry;
}

The structure anl_entry is accessed in 4/pc and here.

§5. The model here is that the list can be reduced in size by marking entries for deletion and then, subsequently, having all such entries removed. Note that the head action_name_list object remains valid even if every entry is removed.

void ActionNameLists::mark_for_deletion(anl_entry *X) {
    if (X) X->marked_for_deletion = TRUE;
    else internal_error("tried to mark null entry for deletion");
}

int ActionNameLists::marked_for_deletion(anl_entry *X) {
    if (X) return X->marked_for_deletion;
    return FALSE;
}

§6. This function actually does two things: deletes unwanted entries, and deletes entries which fail to change the word position.

void ActionNameLists::remove_entries_marked_for_deletion(action_name_list *list) {
    if (list) {
        int pos = -2;
        for (anl_entry *entry = list->entries, *prev = NULL; entry; entry = entry->next_entry) {
            if (Log::aspect_switched_on(ACTION_PATTERN_PARSING_DA)) {
                ActionNameLists::log_entry(entry); LOG(" (wp %d)",
                    entry->parsing_data.word_position);
            }
            int delete = FALSE;
            if (entry->marked_for_deletion) {
                delete = TRUE;
                LOGIF(ACTION_PATTERN_PARSING, ": marked, so delete\n");
            } else if (pos == entry->parsing_data.word_position) {
                delete = TRUE;
                LOGIF(ACTION_PATTERN_PARSING, ": fails to advance, so delete\n");
            } else {
                LOGIF(ACTION_PATTERN_PARSING, ": retain\n");
            }
            if (delete) {
                if (prev == NULL) list->entries = entry->next_entry;
                else prev->next_entry = entry->next_entry;
            } else {
                prev = entry;
                pos = entry->parsing_data.word_position;
            }
        }
    }
}

§7. The list must be kept in a strict order, as can be seen:

void ActionNameLists::join_to(anl_entry *earlier, anl_entry *later) {
    if (later == earlier) internal_error("loop");
    if (ActionNameLists::precedes(later, earlier)) {
        WRITE_TO(STDERR, "Earlier: "); ActionNameLists::write_entry_briefly(STDERR, earlier); WRITE_TO(STDERR, "\n");
        WRITE_TO(STDERR, "Later:   "); ActionNameLists::write_entry_briefly(STDERR, later); WRITE_TO(STDERR, "\n");
        internal_error("misordering");
    }
    earlier->next_entry = later;
}

§8. Which uses the following function:

● Results in later word positions come first, and if that doesn't decide it
● NAPs come before actions, and if that doesn't decide it
● Older NAPs come before younger ones, and if that doesn't decide it
● Less abbreviated results come first, and if that doesn't decide it
● Action names with longer non-it-length come before shorter, and if that doesn't decide it
● Older action names come before younger, and if that doesn't decide it
● Later-discovered results come before earlier ones — which because of the way we parse them means that results with more fixed words in clauses come before those with fewer.

Here, "older" and "younger" mean how long ago the relevant named_action_pattern or action_name objects were created, and since this happens in source text order, we are really saying "closer to the top of the source text" when we say "older".

Note that this function is transitive (\(p(x, y)\) and \(p(y, z)\) implies \(p(x, z)\)) and antisymmetric (\(p(x, y)\) implies that \(p(y, x)\) is false). Strictly speaking it is not trichotomous, but if neither \(p(x, y)\) nor \(p(y, x)\) then \(x\) and \(y\) have identical item data; and in that case it doesn't matter which way round they are in the list.

int ActionNameLists::precedes(anl_entry *e1, anl_entry *e2) {
    if (e1 == NULL) return FALSE;
    if (e2 == NULL) return TRUE;

    int c = e1->parsing_data.word_position -
            e2->parsing_data.word_position;
    if (c > 0) return TRUE; if (c < 0) return FALSE;

    c = ((e1->item.nap_listed)?(e1->item.nap_listed->allocation_id):10000000) -
        ((e2->item.nap_listed)?(e2->item.nap_listed->allocation_id):10000000);
    if (c > 0) return TRUE; if (c < 0) return FALSE;

    c = e1->parsing_data.abbreviation_level -
        e2->parsing_data.abbreviation_level;
    if (c < 0) return TRUE; if (c > 0) return FALSE;

    c = ((e1->item.action_listed)?(ActionNameNames::non_it_length(e1->item.action_listed)):10000000) -
        ((e2->item.action_listed)?(ActionNameNames::non_it_length(e2->item.action_listed)):10000000);
    if (c > 0) return TRUE; if (c < 0) return FALSE;

    c = ((e1->item.action_listed)?(e1->item.action_listed->allocation_id):10000000) -
        ((e2->item.action_listed)?(e2->item.action_listed->allocation_id):10000000);
    if (c > 0) return TRUE; if (c < 0) return FALSE;

    return FALSE;
}

§9. These lists are never long, so we don't need to worry about running time here.

On entry main_list begins a linked list in which each entry ActionNameLists::precedes the next; while new_list can be in any sequence, but must be disjoint from the main_list, i.e., have no entries in common.

On exit the return value heads a linked list in which each entry precedes the next, and which includes exactly the members of the two lists passed to it.

anl_entry *ActionNameLists::join_entries(anl_entry *new_list, anl_entry *main_list) {
    if (new_list == NULL) return main_list;
    if (main_list == NULL) return new_list;
    for (anl_entry *X = new_list; X; X = X->next_entry)
        for (anl_entry *Y = main_list; Y; Y = Y->next_entry)
            if (X == Y) internal_error("ANLs not disjoint");
    anl_entry *new_entry = new_list;
    while (new_entry) {
        anl_entry *next_entry = new_entry->next_entry;
        new_entry->next_entry = NULL;
        Insertion-sort the new entry into the main list9.1;
        new_entry = next_entry;
    }
    return main_list;
}

§9.1. Insertion-sort the new entry into the main list9.1 =

    if (ActionNameLists::precedes(new_entry, main_list)) {
        ActionNameLists::join_to(new_entry, main_list);
        main_list = new_entry;
    } else {
        for (anl_entry *prev = NULL, *pos = main_list; pos; prev = pos, pos = pos->next_entry) {
            if (ActionNameLists::precedes(new_entry, pos)) {
                if (prev) ActionNameLists::join_to(prev, new_entry);
                ActionNameLists::join_to(new_entry, pos);
                break;
            }
            if (pos->next_entry == NULL) {
                ActionNameLists::join_to(pos, new_entry);
                break;
            }
        }
    }

This code is used in §9.

§10. When not pruning the list, this macro is useful for working through it:

define LOOP_THROUGH_ANL(var, list)
    for (anl_entry *var = (list)?(list->entries):NULL; var; var = var->next_entry)

int ActionNameLists::length(action_name_list *list) {
    int C = 0;
    LOOP_THROUGH_ANL(entry, list) C++;
    return C;
}

int ActionNameLists::nonempty(action_name_list *list) {
    if ((list) && (list->entries)) return TRUE;
    return FALSE;
}

§11. If an action list is a wrapper for a single named action pattern, the following function returns that NAP. Anything more complicated, NULL.

named_action_pattern *ActionNameLists::is_single_NAP(action_name_list *list) {
    if ((ActionNameLists::length(list) == 1) && (list->negation_state == ANL_POSITIVE)) {
        anl_item *item = ActionNameLists::first_item(list);
        return item->nap_listed;
    }
    return NULL;
}

§12. The anl_item material is the actual content we are trying to get at. Like life, items are a mixture of naps and actions. At most one of these fields is non-NULL. If they are both NULL, this represents "doing anything" — a completely unrestricted action.

typedef struct anl_item {
    struct action_name *action_listed;  the action in this ANL list entry
    struct named_action_pattern *nap_listed;  or a named pattern instead
} anl_item;

anl_item ActionNameLists::new_item(void) {
    anl_item item;
    item.action_listed = NULL;
    item.nap_listed = NULL;
    return item;
}

void ActionNameLists::clear_item_data(anl_entry *entry, action_name *an) {
    entry->item.action_listed = an;
    entry->item.nap_listed = NULL;
}

anl_item *ActionNameLists::first_item(action_name_list *list) {
    if ((list) && (list->entries)) return &(list->entries->item);
    return NULL;
}

The structure anl_item is accessed in 4/pc and here.

§13. 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 ActionNameLists::compare_specificity(action_name_list *anl1, action_name_list *anl2) {
    int count1, count2;
    count1 = ActionNameLists::count_actions_covered(anl1);
    count2 = ActionNameLists::count_actions_covered(anl2);
    if (count1 < count2) return 1;
    if (count1 > count2) return -1;
    return 0;
}

int ActionNameLists::count_actions_covered(action_name_list *list) {
    int k = 0, infinity = NUMBER_CREATED(action_name);
    if (list == NULL) return infinity;
    if (list->entries == NULL) return infinity;
    LOOP_THROUGH_ANL(entry, list) {
        if (entry->item.nap_listed) continue;
        if ((entry->item.action_listed) && (k < infinity)) k++;
        else k = infinity;
    }
    if (ActionNameLists::positive(list) == FALSE) k = infinity - k;
    return k;
}

named_action_pattern *ActionNameLists::nap(anl_entry *entry) {
    if (entry) return entry->item.nap_listed;
    return NULL;
}

action_name *ActionNameLists::action(anl_entry *entry) {
    if (entry) return entry->item.action_listed;
    return NULL;
}

§14. A given action an falls within the context of this list if it appears positively in the list, or negatively by not falling into a category excluded by it; for example, "examining" falls within "examining something", and also within "doing something other than looking at something" (a case of itemwise negation) but not "doing something other than looking".

int ActionNameLists::covers_action(action_name_list *list, action_name *an) {
    LOOP_THROUGH_ANL(entry, list) {
        anl_item *item = &(entry->item);
        int within = FALSE;
        if (item->action_listed == an) within = TRUE;
        else if (item->nap_listed) within =
            NamedActionPatterns::covers_action(item->nap_listed, an);
        if (((within) && (ActionNameLists::positive(list))) ||
            ((within == FALSE) && (ActionNameLists::positive(list) == FALSE)))
            return TRUE;
    }
    return FALSE;
}

§15. The anl_parsing_data material is needed on a temporary basis when parsing the text leading to a list:

typedef struct anl_parsing_data {
    int word_position;  and some values used temporarily during parsing
    int abbreviation_level;  number of words missing
    struct anl_clause *anl_clauses;  clauses in this reading
} anl_parsing_data;

anl_parsing_data ActionNameLists::new_parsing_data(int at) {
    anl_parsing_data parsing_data;
    parsing_data.anl_clauses = NULL;
    parsing_data.abbreviation_level = 0;
    parsing_data.word_position = -1;
    return parsing_data;
}

void ActionNameLists::clear_parsing_data(anl_entry *entry, wording W) {
    entry->parsing_data.anl_clauses = NULL;
    int at = -1;
    if (Wordings::nonempty(W)) at = Wordings::first_wn(W);
    entry->parsing_data.word_position = at;
    entry->parsing_data.abbreviation_level = 0;
}

The structure anl_parsing_data is accessed in 4/pc and here.

§16. Parsing data contains a linked list, sorted in ascending clause ID order, of the following structures, which are similar (but not identical) to ap_clause objects: and indeed, in the happy case where an entry in an action name list produces a successful parse in Parse Clauses, the anl_clause objects of that entry will be turned into a string of ap_clause objects in the final AP.

typedef struct anl_clause {
    int clause_ID;
    struct wording clause_text;
    struct anl_clause *next_clause;
    struct shared_variable *stv_to_match;
    struct parse_node *evaluation;
} anl_clause;

The structure anl_clause is accessed in 4/apc, 4/pc, 4/gng and here.

§17. And this is convenient for looking through them:

define LOOP_THROUGH_ANL_CLAUSES(c, entry)
    for (anl_clause *c = (entry)?(entry->parsing_data.anl_clauses):NULL; c; c = c->next_clause)

int ActionNameLists::noun_count(anl_entry *entry) {
    int p = 0;
    LOOP_THROUGH_ANL_CLAUSES(c, entry)
        if ((c->clause_ID == NOUN_AP_CLAUSE) || (c->clause_ID == SECOND_AP_CLAUSE))
            p++;
    return p;
}

int ActionNameLists::has_clause(anl_entry *entry, int C) {
    LOOP_THROUGH_ANL_CLAUSES(c, entry)
        if (c->clause_ID == C) return TRUE;
    return FALSE;
}

anl_clause *ActionNameLists::get_clause(anl_entry *entry, int C) {
    LOOP_THROUGH_ANL_CLAUSES(c, entry)
        if (c->clause_ID == C) return c;
    return NULL;
}

wording ActionNameLists::get_clause_wording(anl_entry *entry, int C) {
    LOOP_THROUGH_ANL_CLAUSES(c, entry)
        if (c->clause_ID == C) return c->clause_text;
    return EMPTY_WORDING;
}

§18. Note that it is legal to create an ANL clause with empty wording.

anl_entry *ActionNameLists::set_clause_wording(anl_entry *entry, int C, wording W) {
    if (entry == NULL) internal_error("no entry");
    anl_clause *prev = NULL;
    LOOP_THROUGH_ANL_CLAUSES(c, entry) {
        if (c->clause_ID == C) {
            c->clause_text = W; return entry;
        }
        if (c->clause_ID > C) Insert clause here18.1;
        prev = c;
    }
    anl_clause *c = NULL;
    Insert clause here18.1;
}

§18.1. Insert clause here18.1 =

    anl_clause *nc = CREATE(anl_clause);
    nc->clause_ID = C;
    nc->clause_text = W;
    nc->evaluation = NULL;
    nc->stv_to_match = NULL;
    if (prev) { nc->next_clause = c; prev->next_clause = nc; }
    else { nc->next_clause = c; entry->parsing_data.anl_clauses = nc; }
    return entry;

This code is used in §18 (twice).

§19. This is really the same function but setting variable as well as ID.

anl_entry *ActionNameLists::set_clause_wording_and_stv(anl_entry *entry, int C,
    wording W, shared_variable *stv) {
    if (entry == NULL) internal_error("no entry");
    anl_clause *prev = NULL;
    LOOP_THROUGH_ANL_CLAUSES(c, entry) {
        if (c->clause_ID == C) {
            c->clause_text = W; c->stv_to_match = stv; return entry;
        }
        if (c->clause_ID > C) Insert clause here with stv19.1;
        prev = c;
    }
    anl_clause *c = NULL;
    Insert clause here with stv19.1;
}

§19.1. Insert clause here with stv19.1 =

    anl_clause *nc = CREATE(anl_clause);
    nc->clause_ID = C;
    nc->clause_text = W;
    nc->evaluation = NULL;
    nc->stv_to_match = stv;
    if (prev) { nc->next_clause = c; prev->next_clause = nc; }
    else { nc->next_clause = c; entry->parsing_data.anl_clauses = nc; }
    return entry;

This code is used in §19 (twice).

§20. This truncates a clause so that it stops just before the given word number. For example, "croquet in the gardens" might be truncated to just "croquet". Here, if the text should become empty as a result then the clause is deleted from the list. This is important since it removes temporary TAIL_AP_CLAUSE clauses if they successfully convert into more permanent ones.

void ActionNameLists::truncate_clause(anl_entry *entry, int C, int wn) {
    anl_clause *prev = NULL;
    LOOP_THROUGH_ANL_CLAUSES(c, entry) {
        if (c->clause_ID == C) {
            c->clause_text = Wordings::up_to(c->clause_text, wn - 1);
            if (Wordings::empty(c->clause_text)) {
                if (prev) prev->next_clause = c->next_clause;
                else entry->parsing_data.anl_clauses = c->next_clause;
            }
            return;
        }
        prev = c;
    }
}

§21. Up to two "nouns" can be added to an entry; the first to be added is put into the NOUN_AP_CLAUSE clause, and the second to SECOND_AP_CLAUSE.

anl_entry *ActionNameLists::add_noun(anl_entry *entry, wording W) {
    switch (ActionNameLists::noun_count(entry)) {
        case 0: ActionNameLists::set_clause_wording(entry, NOUN_AP_CLAUSE, W); break;
        case 1: ActionNameLists::set_clause_wording(entry, SECOND_AP_CLAUSE, W); break;
        default: internal_error("too many nouns for ANL entry");
    }
    return entry;
}

§22. So much for clauses. Entries also record the word position at which the action (or named action pattern) text kicks in:

int ActionNameLists::word_position(anl_entry *entry) {
    if (entry) return entry->parsing_data.word_position;
    return -1;
}

§23. Single and best actions. This tests whether the list gives a single positive action:

action_name *ActionNameLists::single_positive_action(action_name_list *list) {
    if ((ActionNameLists::length(list) == 1) &&
        (ActionNameLists::positive(list)))
        return ActionNameLists::first_item(list)->action_listed;
    return NULL;
}

§24. This is used only when the list is part of an exactly known action, so that it should contain just one item, and this should be an actual action, not a NAP.

action_name *ActionNameLists::get_the_one_true_action(action_name_list *list) {
    action_name *an = ActionNameLists::single_positive_action(list);
    if (an == NULL) internal_error("Singleton ANL points to null AN");
    return an;
}

§25. The "best" action is the one maximising the number of words in the fixed part of an action name: thus if there are actions "throwing away" and "throwing", then in the ANL arising from the text "throwing away the fish", the action "throwing away" is better than the action "throwing".

This is easy to detect here, because the enforced ordering of the list ensures that earlier entries are (in this sense) better than later ones.

If the list includes actions at two different word positions, so that they are not alternate readings from the same point, then by definition there is no best action. (For example, in "throwing or removing something".)

action_name *ActionNameLists::get_best_action(action_name_list *list) {
    if (ActionNameLists::positive(list) == FALSE) return NULL;
    int posn = -1;
    action_name *choice = NULL;
    LOOP_THROUGH_ANL(entry, list) {
        if (entry->item.action_listed) {
            if (entry->parsing_data.word_position != posn) {
                if (posn >= 0) return NULL;
                posn = entry->parsing_data.word_position;
            }
            if (choice == NULL) choice = entry->item.action_listed;
        } else return NULL;
    }
    return choice;
}

§26. Duplication.

anl_entry *ActionNameLists::duplicate_entry(anl_entry *entry) {
    anl_entry *new_entry = ActionNameLists::new_entry_at(EMPTY_WORDING);
    new_entry->parsing_data = entry->parsing_data;
    new_entry->parsing_data.anl_clauses = NULL;
    LOOP_THROUGH_ANL_CLAUSES(c, entry)
        ActionNameLists::set_clause_wording_and_stv(new_entry,
            c->clause_ID, c->clause_text, c->stv_to_match);
    new_entry->item = entry->item;
    new_entry->next_entry = NULL;
    return new_entry;
}

§27. Logging.

void ActionNameLists::log(action_name_list *list) {
    if (list == NULL) {
        LOG("<null-anl>");
    } else {
        if (ActionNameLists::listwise_negated(list)) { LOG("L-NOT[\n"); LOG_INDENT; }
        if (ActionNameLists::itemwise_negated(list)) { LOG("I-NOT[\n"); LOG_INDENT; }
        int benchmark = 0;
        LOOP_THROUGH_ANL(entry, list) {
            if ((entry->parsing_data.word_position < benchmark) || (benchmark == 0))
                benchmark = entry->parsing_data.word_position;
        }
        int c = 1;
        LOOP_THROUGH_ANL(entry, list) {
            LOG("(%d). +%d ", c, entry->parsing_data.word_position - benchmark);
            ActionNameLists::log_entry(entry);
            LOG("\n");
            c++;
        }
        if (ActionNameLists::itemwise_negated(list)) { LOG_OUTDENT; LOG("\n]\n"); }
        if (ActionNameLists::listwise_negated(list)) { LOG_OUTDENT; LOG("\n]\n"); }
    }
}

void ActionNameLists::log_entry(anl_entry *entry) {
    if (entry == NULL) {
        LOG("<null-entry>");
    } else {
        if (entry->marked_for_deletion) LOG(" (to be deleted)");
        ActionNameLists::log_entry_briefly(entry);
        LOOP_THROUGH_ANL_CLAUSES(c, entry)
            if (Wordings::nonempty(c->clause_text)) {
                LOG(" "); ActionNameLists::log_clause(c);
            }
    }
}

void ActionNameLists::log_clause(anl_clause *c) {
    LOG("[");
    APClauses::write_clause_ID(DL, c->clause_ID, c->stv_to_match);
    LOG(": %W]", c->clause_text);
}

void ActionNameLists::log_briefly(action_name_list *list) {
    if (list == NULL) {
        LOG("<null-anl>");
    } else {
        if (ActionNameLists::listwise_negated(list)) LOG("L-NOT[ ");
        if (ActionNameLists::itemwise_negated(list)) LOG("I-NOT[ ");
        int c = 0;
        LOOP_THROUGH_ANL(entry, list) {
            if (c > 0) LOG(" / ");
            ActionNameLists::log_entry_briefly(entry);
            c++;
        }
        if (ActionNameLists::listwise_negated(list)) LOG(" ]");
    }
}

void ActionNameLists::log_entry_briefly(anl_entry *entry) {
    if (entry->item.nap_listed) {
        LOG("%W", Nouns::nominative_singular(entry->item.nap_listed->as_noun));
    } else if (entry->item.action_listed == NULL)
        LOG("ANY");
    else {
        LOG("%W", ActionNameNames::tensed(entry->item.action_listed, IS_TENSE));
    }
}

void ActionNameLists::write_entry_briefly(OUTPUT_STREAM, anl_entry *entry) {
    if (entry->item.nap_listed) {
        WRITE("%W", Nouns::nominative_singular(entry->item.nap_listed->as_noun));
    } else if (entry->item.action_listed == NULL)
        WRITE("ANY");
    else {
        WRITE("%W", ActionNameNames::tensed(entry->item.action_listed, IS_TENSE));
    }
}

§28. Parsing text to an ANL. Action name lists arise only for parsing text, and only from the function below; this might match, for example, "doing something other than waiting", or "dropping the box". We make no effort to understand the words which are not part of the action: "dropping the box" is just "dropping (two words)" here.

Note that it works in either IS_TENSE or HASBEEN_TENSE, and that sense is set to FALSE (if it is supplied) when the text had a negative sense — something other than something — or TRUE for a positive one.

The test group :actions is helpful in catching errors here.

§29.

int anl_parsing_tense = IS_TENSE;
action_name_list *ActionNameLists::parse(wording W, int tense, int *sense) {
    if (Wordings::mismatched_brackets(W)) return NULL;
    int t = anl_parsing_tense;
    anl_parsing_tense = tense;
    int r = <action-list>(W);
    anl_parsing_tense = t;
    if (r) {
        if (sense) *sense = <<r>>;
        return <<rp>>;
    }
    return NULL;
}

§30. The outer parts of the syntax are handled by a Preform grammar.

<action-list> ::=
    doing something/anything other than <excluded-list> | ==> { pass 1 }
    doing something/anything except <excluded-list> |     ==> { pass 1 }
    doing something/anything to/with {...} |    ==> { -, - }; wording TW = WR[1]; Something to30.1
    doing something/anything |                  ==> Something30.2
    doing something/anything {...} |            ==> { -, - }; TW = WR[1]; Something in30.3
    <anl>                                       ==> X30.4

<excluded-list> ::=
    <anl> to/with {<minimal-common-to-text>} |  ==> Something except X with30.6
    <anl>                                       ==> Something except X30.5

<minimal-common-to-text> ::=
    _,/or ... |                                 ==> { fail }
    ... to/with ... |                           ==> { fail }
    ...

This is Preform grammar, not regular C code.

§30.1. Something to30.1 =

    TW = GET_RW(<action-list>, 1);
    anl_entry *entry = ActionNameLists::new_entry_at(W);
    ActionNameLists::add_noun(entry, TW);
    ==> { TRUE, ActionNameLists::new_list(ActionNameLists::ramify(entry), ANL_POSITIVE) }

This code is used in §30.

§30.2. Something30.2 =

    anl_entry *entry = ActionNameLists::new_entry_at(W);
    ==> { TRUE, ActionNameLists::new_list(entry, ANL_POSITIVE) };

This code is used in §30.

§30.3. Something in30.3 =

    TW = GET_RW(<action-list>, 1);
    anl_entry *entry = ActionNameLists::new_entry_at(W);
    ActionNameLists::set_clause_wording(entry, TAIL_AP_CLAUSE, TW);
    ==> { TRUE, ActionNameLists::new_list(ActionNameLists::ramify(entry), ANL_POSITIVE) };

This code is used in §30.

§30.4. X30.4 =

    ==> { TRUE, ActionNameLists::new_list(RP[1], ANL_POSITIVE) };

This code is used in §30.

§30.5. Something except X30.5 =

    ==> { FALSE, ActionNameLists::new_list(RP[1], ANL_NEGATED_LISTWISE) };

This code is used in §30.

§30.6. Something except X with30.6 =

    anl_entry *entry = RP[1];
    if ((entry == NULL) ||
        (entry->item.action_listed == NULL) ||
        (ActionSemantics::can_have_noun(entry->item.action_listed) == FALSE)) {
        ==> { fail production };
    }
    ActionNameLists::add_noun(entry, GET_RW(<excluded-list>, 1));
    ==> { FALSE, ActionNameLists::new_list(ActionNameLists::ramify(entry), ANL_NEGATED_ITEMWISE) };

This code is used in §30.

§31. This matches a comma/or-separated list of items:

<anl> ::=
    <anl-entry> <anl-tail> |  ==> { -, ActionNameLists::join_entries(RP[1], RP[2]) }
    <anl-entry>               ==> { pass 1 }

<anl-tail> ::=
    , _or <anl> |             ==> { pass 1 }
    _,/or <anl>               ==> { pass 1 }

This is Preform grammar, not regular C code.

§32. Items can be named action patterns, so let's get those out of the way first:

<anl-entry> ::=
    <named-action-pattern> |        ==> { -, ActionNameLists::nap_entry(RP[1], W, EMPTY_WORDING) }
    <named-action-pattern-tailed> | ==> { pass 1 }
    <anl-entry-with-action>         ==> { pass 1 }

<named-action-pattern-tailed> internal {
    for (int i=Wordings::first_wn(W); i<= Wordings::last_wn(W) - 1; i++) {
        if (<named-action-pattern>(Wordings::up_to(W, i))) {
            ==> { -, ActionNameLists::nap_entry(<<rp>>, W, Wordings::from(W, i+1)) };
            return TRUE;
        }
    }
    return FALSE;
}

This is Preform grammar, not regular C code.

§33. Here TW is the "tail wording", that is, any text left over after the name itself. So, for "irreverent behaviour in the presence of the Bishop", the nap may be the "irreverent behaviour", and TW the text "in the presence of the Bishop". We put that temporarily into the TAIL_AP_CLAUSE, and then ramify what had been a single-entry list so that it may now have multiple entries — for example, "irreverent behaviour [in: the presence of the Bishop]" or "irreverent behaviour [in-presence: the Bishop]".

anl_entry *ActionNameLists::nap_entry(named_action_pattern *nap, wording W, wording TW) {
    anl_entry *entry = ActionNameLists::new_entry_at(W);
    entry->item.nap_listed = nap;
    if (Wordings::nonempty(TW))
        ActionNameLists::set_clause_wording(entry, TAIL_AP_CLAUSE, TW);
    return ActionNameLists::ramify(entry);
}

§34. If we aren't going to name an action pattern, we're going to have to spell out an actual choice of action.

<anl-entry-with-action> internal {
    anl_entry *results = NULL;
    Parse the wording into a list of results34.1;
    results = ActionNameLists::ramify(results);
    if (results) {
        ==> { -, results }; return TRUE;
    }
    ==> { fail nonterminal };
}

This is Preform grammar, not regular C code.

§34.1. The following makes a list of results before ramification.

For example, for the text "looking or taking inventory in the presence of Hans in the Laboratory", we get the following set of results:

(1). +2 taking inventory [tail: in the presence of hans in the laboratory]
(2). +2 taking [noun: inventory in the presence of hans in the laboratory]
(3). +0 looking

Parse the wording into a list of results34.1 =

    anl_entry *trial_entry = ActionNameLists::new_entry_at(EMPTY_WORDING);
    action_name *an;
    LOOP_OVER(an, action_name) {
        Ready the trial entry for another test34.1.1;
        wording RW = EMPTY_WORDING;
        int abbreviated_to_tail = FALSE;
        Make the trial entry fit this action, if possible, leaving remaining text in RW34.1.2;
        Transfer remaining words to a trailing clause34.1.3;
        Include the trial entry34.1.4;
        NoMatch: ;
    }

This code is used in §34.

§34.1.1. Ready the trial entry for another test34.1.1 =

    trial_entry->next_entry = NULL;
    ActionNameLists::clear_item_data(trial_entry, an);
    ActionNameLists::clear_parsing_data(trial_entry, W);

This code is used in §34.1.

§34.1.2. Here XW will be the wording of the action name, say "removing it from"; we try to fit W to this, say "removing a heavy thing from something in the Dining Room"; and if we cannot, we run away to the label NoMatch, which is inelegant, but there's no elegant way to break out of nested loops in C.

Make the trial entry fit this action, if possible, leaving remaining text in RW34.1.2 =

    int x_ended = FALSE;
    int it_optional = ActionNameNames::it_optional(an);
    int abbreviable = ActionNameNames::abbreviable(an);
    wording XW = ActionNameNames::tensed(an, anl_parsing_tense);
    int w_m = Wordings::first_wn(W), x_m = Wordings::first_wn(XW), n = 0;
    while ((w_m <= Wordings::last_wn(W)) && (x_m <= Wordings::last_wn(XW))) {
        if (Lexer::word(x_m++) != Lexer::word(w_m++)) {
            if ((abbreviable) && (it_optional) && (n >= 1)) {
                x_ended = TRUE; abbreviated_to_tail = TRUE; x_m--; w_m--;
            } else goto NoMatch;
            break;
        }
        n++;
        if (x_m > Wordings::last_wn(XW)) { x_ended = TRUE; break; }
        if (<object-pronoun>(Wordings::one_word(x_m))) {
            if (w_m > Wordings::last_wn(W)) x_ended = TRUE; else {
                int j = -1, k;
                for (k=(it_optional)?(w_m):(w_m+1); k<=Wordings::last_wn(W); k++)
                    if (Lexer::word(k) == Lexer::word(x_m+1)) { j = k; break; }
                if (j<0) goto NoMatch;
                if (j-1 >= w_m)
                    ActionNameLists::add_noun(trial_entry, Wordings::new(w_m, j-1));
                else
                    ActionNameLists::add_noun(trial_entry, EMPTY_WORDING);
                w_m = j; x_m++;
            }
        }
        if (x_ended) break;
    }
    if ((w_m > Wordings::last_wn(W)) && (x_ended == FALSE)) {
        if (abbreviable) x_ended = TRUE; else goto NoMatch;
    }
    if (x_m <= Wordings::last_wn(XW))
        trial_entry->parsing_data.abbreviation_level = Wordings::last_wn(XW)-x_m+1;
    RW = Wordings::from(W, w_m);

This code is used in §34.1.

§34.1.3. For example, in "looking or taking inventory in the presence of Hans in the Laboratory", after finding "taking inventory" as a possible action, we are left with "in the presence of Hans in the Laboratory". These have to be stored into TAIL_AP_CLAUSE, since they cannot be part of any noun — because taking inventory doesn't have a noun. But when finding "taking", the remaining words "inventory in the presence of Hans in the Laboratory" have to go into NOUN_AP_CLAUSE — the taking action does have a noun.

Transfer remaining words to a trailing clause34.1.3 =

    if (Wordings::nonempty(RW)) {
        if ((ActionSemantics::can_have_noun(an)) && (abbreviated_to_tail == FALSE)) {
            ActionNameLists::add_noun(trial_entry, RW);
        } else {
            ActionNameLists::set_clause_wording(trial_entry, TAIL_AP_CLAUSE, RW);
        }
    }

This code is used in §34.1.

§34.1.4. So this is the happy ending. We don't copy the trial entry; we insertion-sort the structure itself into the results list, and make a fresh structure to be the trial entry for future trials.

Include the trial entry34.1.4 =

    results = ActionNameLists::join_entries(trial_entry, results);
    trial_entry = ActionNameLists::new_entry_at(EMPTY_WORDING);

This code is used in §34.1.

§35. And now we get to ramification. This is what happens last, when a set of raw results has been produced. We "ramify" this by expanding it into multiple readings according to how the trailing clause might in fact be made up of sub-clauses. For example, our unramified set of results

(1). +2 taking inventory [tail: in the presence of hans in the laboratory]
(2). +2 taking [noun: inventory in the presence of hans in the laboratory]
(3). +0 looking

becomes the ramified set

(1). +2 taking inventory [in: the laboratory] [in-presence: hans]
(2). +2 taking inventory [in-presence: hans in the laboratory]
(3). +2 taking [noun: inventory] [in: the laboratory] [in-presence: hans]
(4). +2 taking [noun: inventory] [in-presence: hans in the laboratory]
(5). +2 taking [noun: inventory in the presence of hans] [in: the laboratory]
(6). +2 taking [noun: inventory in the presence of hans in the laboratory]
(7). +0 looking

Note that the TAIL_AP_CLAUSE clauses, which were just temporary holders for leftover text, have gone entirely. Had it been impossible to break them into legal subclauses, they would have caused the result to be struck out altogether. For example, this:

(1). +0 taking inventory [tail: book]
(2). +0 taking [noun: inventory book]

ramifies to just

(1). +0 taking [noun: inventory book]

because the tail wording "book" after "taking inventory" cannot be read as a sequence of subclauses.

Ramification thus consists of a sudden increase in the number of possible readings, which we call an explosion, followed by a cull of anything which still has a TAIL_AP_CLAUSE.

anl_entry *ActionNameLists::ramify(anl_entry *results) {
    for (anl_entry *entry = results; entry; ) {
        anl_entry *next = entry->next_entry;
        ActionNameLists::explode(entry);
        entry = next;
    }
    for (anl_entry *prev = NULL, *entry = results; entry; entry = entry->next_entry)
        if (Wordings::nonempty(ActionNameLists::get_clause_wording(entry, TAIL_AP_CLAUSE))) {
            if (prev) prev->next_entry = entry->next_entry;
            else results = entry->next_entry;
        } else {
            prev = entry;
        }
    return results;
}

§36. And here is the explosion part. tc here identifies what the trailing clause actually is:

void ActionNameLists::explode(anl_entry *entry) {
    int tc = -1;
    if (ActionNameLists::has_clause(entry, TAIL_AP_CLAUSE)) tc = TAIL_AP_CLAUSE;
    else if (ActionNameLists::has_clause(entry, SECOND_AP_CLAUSE)) tc = SECOND_AP_CLAUSE;
    else if (ActionNameLists::has_clause(entry, NOUN_AP_CLAUSE)) tc = NOUN_AP_CLAUSE;
    else return;
    wording TW = ActionNameLists::get_clause_wording(entry, tc);
    ActionNameLists::explode_clause(entry, tc, Wordings::first_wn(TW));
}

§37. The following is recursive and exhausts all possible readings of the trailing clause, exactly once each. We have to store what ought to be variables on the stack as globals because of the little dance via Preform nonterminals below, but the idea is simple enough. This function finds the first point at which the reading diverges: for example, if the text is "banjo in the Conservatoire", then "in" is the point of divergence. Maybe the text is all one clause, or maybe it's just "banjo" and there is then a second clause "in the Conservatoire".

anl_entry *currently_exploding_entry = NULL;
int currently_exploding_clause = -1;
int explosions_count = 0;
void ActionNameLists::explode_clause(anl_entry *entry, int tc, int from_wn) {
    anl_entry *saved = currently_exploding_entry;
    int saved_C = currently_exploding_clause;
    currently_exploding_entry = entry;
    currently_exploding_clause = tc;
    wording TW = ActionNameLists::get_clause_wording(entry, tc);
    explosions_count++;
    int start = explosions_count;
    LOG_INDENT;
    for (int w = from_wn; ((w < Wordings::last_wn(TW)) && (start == explosions_count)); w++) {
        if (<text-precluding-divergence>(Wordings::one_word(w))) { w++; continue; }
        <detonate-at-divergence-points>(Wordings::from(TW, w));
    }
    LOG_OUTDENT;
    currently_exploding_entry = saved;
    currently_exploding_clause = saved_C;
}

§38. Note that if we spot the <text-precluding-divergence> wording immediately before the point of divergence, we forbid divergence to occur. It must contain only single words.

This actually matters surprisingly rarely, but enables us to handle quite difficult cases like "Instead of buying the cheapest spice which is in the market" without typechecking errors occurring on the unwanted case of reading this as "Instead of buying the cheapest spice which is" plus the clause "in the market" — the latter being, in fact, valid.

<text-precluding-divergence> ::=
    is |
    not

This is Preform grammar, not regular C code.

§39. Divergence points, then, must not unexpectedly use upper case — so "taking Puss In Boots" is never read as possibly having an "in: Boots" clause; and they either use the two standard wordings "in the presence of" or "in". or else wording provided by a matching variable in an action declaration.

Note that this nonterminal never matches! It is parsed for its side-effect of calling ActionNameLists::detonate on every possibility.

<detonate-at-divergence-points> ::=
    _in _the _presence _of ... |  ==> Explode in-presence39.2
    _in ... |                     ==> Explode in39.1
    <clause-opening> ...

This is Preform grammar, not regular C code.

§39.1. Explode in39.1 =

    wording T = GET_RW(<detonate-at-divergence-points>, 1);
    ActionNameLists::detonate(IN_AP_CLAUSE, NULL, T, W);
    return FALSE;

This code is used in §39.

§39.2. Explode in-presence39.2 =

    wording T = GET_RW(<detonate-at-divergence-points>, 1);
    ActionNameLists::detonate(IN_THE_PRESENCE_OF_AP_CLAUSE, NULL, T, W);
    return FALSE;

This code is used in §39.

§40. For example, in the text "going from the Park to the Town", "from" and "to" are divergence points:

<clause-opening> internal ? {
    if (Word::unexpectedly_upper_case(Wordings::first_wn(W)) == FALSE) {
        action_name *chief_an = currently_exploding_entry->item.action_listed;
        if (chief_an) {
            shared_variable_set *stvo = chief_an->action_variables;
            if (stvo) {
                shared_variable *stv;
                LOOP_OVER_LINKED_LIST(stv, shared_variable, stvo->variables) {
                    wording VW = stv->match_wording_text;
                    if (Wordings::starts_with(W, VW)) {
                        wording T = Wordings::from(W, Wordings::first_wn(W) + Wordings::length(VW));
                        int potential_C = APClauses::clause_ID_for_action_variable(stv);
                        ActionNameLists::detonate(potential_C, stv, T, W);
                    }
                }
            }
        }
    }
    ==> { fail nonterminal };
}

This is Preform grammar, not regular C code.

§41. Finally the actual business of splitting our original entry into two, one (called X here) in which a new clause appears at the point of divergence, and one (called Y) where it does not.

The order is important here because this is why when there are multiple readings of clauses in the ramified list, the readings with more clauses come before the readings with fewer.

Note that we only diverge if the new clause is one which does not already exist for this entry. This is because it makes no sense to have the same clause twice, and also means that pathological text like "in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in" cannot cause a combinatorial nightmare; because each clause appears at most once in any entry, the number of entries produced by ramification is capped at \(2^n\), where \(n\) is the number of different clauses whose matching words appear somewhere in the text. As Inform ships with only seven different clauses anyway, this will never be too bad.

void ActionNameLists::detonate(int potential_C, shared_variable *stv, wording T, wording W) {
    if (ActionNameLists::has_clause(currently_exploding_entry, potential_C) == FALSE) {
        anl_entry *extra = ActionNameLists::duplicate_entry(currently_exploding_entry);

        anl_entry *Y = extra, *X = currently_exploding_entry;
        ActionNameLists::set_clause_wording(X, potential_C, T);
        anl_clause *extra_clause = ActionNameLists::get_clause(X, potential_C);
        extra_clause->stv_to_match = stv;
        ActionNameLists::truncate_clause(X, currently_exploding_clause, Wordings::first_wn(W));

        anl_entry *n = currently_exploding_entry->next_entry;
        currently_exploding_entry->next_entry = extra;
        extra->next_entry = n;
        ActionNameLists::explode_clause(Y, currently_exploding_clause, Wordings::first_wn(W)+1);
        ActionNameLists::explode_clause(X, potential_C, Wordings::first_wn(T)+1);
    }
}

§42. Lastly, this little function is provided for unit testing the above, and is otherwise never called.

void ActionNameLists::test_list(wording W) {
    LOG("Action name list for: %W\n", W);
    action_name_list *anl = ActionNameLists::parse(W, IS_TENSE, NULL);
    LOG("$L\n", anl);
}