void
SortColl_set_reader(SortCollector *self, SegReader *reader) {
SortCollectorIVARS *const ivars = SortColl_IVARS(self);
SortReader *sort_reader
= (SortReader*)SegReader_Fetch(reader, VTable_Get_Name(SORTREADER));
// Reset threshold variables and trigger auto-action behavior.
MatchDocIVARS *const bumped_ivars = MatchDoc_IVARS(ivars->bumped);
bumped_ivars->doc_id = INT32_MAX;
ivars->bubble_doc = INT32_MAX;
bumped_ivars->score = ivars->need_score ? F32_NEGINF : F32_NAN;
ivars->bubble_score = ivars->need_score ? F32_NEGINF : F32_NAN;
ivars->actions = ivars->auto_actions;
// Obtain sort caches. Derive actions array for this segment.
if (ivars->need_values && sort_reader) {
for (uint32_t i = 0, max = ivars->num_rules; i < max; i++) {
SortRule *rule = (SortRule*)VA_Fetch(ivars->rules, i);
CharBuf *field = SortRule_Get_Field(rule);
SortCache *cache = field
? SortReader_Fetch_Sort_Cache(sort_reader, field)
: NULL;
ivars->sort_caches[i] = cache;
ivars->derived_actions[i] = S_derive_action(rule, cache);
if (cache) { ivars->ord_arrays[i] = SortCache_Get_Ords(cache); }
else { ivars->ord_arrays[i] = NULL; }
}
}
ivars->seg_doc_max = reader ? SegReader_Doc_Max(reader) : 0;
Coll_set_reader((Collector*)self, reader);
}
HitQueue*
HitQ_init(HitQueue *self, Schema *schema, SortSpec *sort_spec,
uint32_t wanted) {
HitQueueIVARS *const ivars = HitQ_IVARS(self);
if (sort_spec) {
VArray *rules = SortSpec_Get_Rules(sort_spec);
uint32_t num_rules = VA_Get_Size(rules);
uint32_t action_num = 0;
if (!schema) {
THROW(ERR, "Can't supply sort_spec without schema");
}
ivars->need_values = false;
ivars->num_actions = num_rules;
ivars->actions = (uint8_t*)MALLOCATE(num_rules * sizeof(uint8_t));
ivars->field_types = (FieldType**)CALLOCATE(num_rules, sizeof(FieldType*));
for (uint32_t i = 0; i < num_rules; i++) {
SortRule *rule = (SortRule*)VA_Fetch(rules, i);
int32_t rule_type = SortRule_Get_Type(rule);
bool reverse = SortRule_Get_Reverse(rule);
if (rule_type == SortRule_SCORE) {
ivars->actions[action_num++] = reverse
? COMPARE_BY_SCORE_REV
: COMPARE_BY_SCORE;
}
else if (rule_type == SortRule_DOC_ID) {
ivars->actions[action_num++] = reverse
? COMPARE_BY_DOC_ID_REV
: COMPARE_BY_DOC_ID;
}
else if (rule_type == SortRule_FIELD) {
String *field = SortRule_Get_Field(rule);
FieldType *type = Schema_Fetch_Type(schema, field);
if (type) {
ivars->field_types[action_num] = (FieldType*)INCREF(type);
ivars->actions[action_num++] = reverse
? COMPARE_BY_VALUE_REV
: COMPARE_BY_VALUE;
ivars->need_values = true;
}
else {
// Skip over fields we don't know how to sort on.
continue;
}
}
else {
THROW(ERR, "Unknown SortRule type: %i32", rule_type);
}
}
}
else {
ivars->num_actions = 2;
ivars->actions = (uint8_t*)MALLOCATE(ivars->num_actions * sizeof(uint8_t));
ivars->actions[0] = COMPARE_BY_SCORE;
ivars->actions[1] = COMPARE_BY_DOC_ID;
}
return (HitQueue*)PriQ_init((PriorityQueue*)self, wanted);
}
SortCollector*
SortColl_init(SortCollector *self, Schema *schema, SortSpec *sort_spec,
uint32_t wanted) {
VArray *rules = sort_spec
? (VArray*)INCREF(SortSpec_Get_Rules(sort_spec))
: S_default_sort_rules();
uint32_t num_rules = VA_Get_Size(rules);
// Validate.
if (sort_spec && !schema) {
THROW(ERR, "Can't supply a SortSpec without a Schema.");
}
if (!num_rules) {
THROW(ERR, "Can't supply a SortSpec with no SortRules.");
}
// Init.
Coll_init((Collector*)self);
SortCollectorIVARS *const ivars = SortColl_IVARS(self);
ivars->total_hits = 0;
ivars->bubble_doc = INT32_MAX;
ivars->bubble_score = F32_NEGINF;
ivars->seg_doc_max = 0;
// Assign.
ivars->wanted = wanted;
// Derive.
ivars->hit_q = HitQ_new(schema, sort_spec, wanted);
ivars->rules = rules; // absorb refcount.
ivars->num_rules = num_rules;
ivars->sort_caches = (SortCache**)CALLOCATE(num_rules, sizeof(SortCache*));
ivars->ord_arrays = (void**)CALLOCATE(num_rules, sizeof(void*));
ivars->actions = (uint8_t*)CALLOCATE(num_rules, sizeof(uint8_t));
// Build up an array of "actions" which we will execute during each call
// to Collect(). Determine whether we need to track scores and field
// values.
ivars->need_score = false;
ivars->need_values = false;
for (uint32_t i = 0; i < num_rules; i++) {
SortRule *rule = (SortRule*)VA_Fetch(rules, i);
int32_t rule_type = SortRule_Get_Type(rule);
ivars->actions[i] = S_derive_action(rule, NULL);
if (rule_type == SortRule_SCORE) {
ivars->need_score = true;
}
else if (rule_type == SortRule_FIELD) {
CharBuf *field = SortRule_Get_Field(rule);
FieldType *type = Schema_Fetch_Type(schema, field);
if (!type || !FType_Sortable(type)) {
THROW(ERR, "'%o' isn't a sortable field", field);
}
ivars->need_values = true;
}
}
// Perform an optimization. So long as we always collect docs in
// ascending order, Collect() will favor lower doc numbers -- so we may
// not need to execute a final COMPARE_BY_DOC_ID action.
ivars->num_actions = num_rules;
if (ivars->actions[num_rules - 1] == COMPARE_BY_DOC_ID) {
ivars->num_actions--;
}
// Override our derived actions with an action which will be excecuted
// autmatically until the queue fills up.
ivars->auto_actions = (uint8_t*)MALLOCATE(1);
ivars->auto_actions[0] = wanted ? AUTO_ACCEPT : AUTO_REJECT;
ivars->derived_actions = ivars->actions;
ivars->actions = ivars->auto_actions;
// Prepare a MatchDoc-in-waiting.
VArray *values = ivars->need_values ? VA_new(num_rules) : NULL;
float score = ivars->need_score ? F32_NEGINF : F32_NAN;
ivars->bumped = MatchDoc_new(INT32_MAX, score, values);
DECREF(values);
return self;
}
