unsigned hash_size;
int num_records;
- struct reclist_bucket *sorted_list;
- struct reclist_bucket *sorted_ptr;
+ struct record_cluster *sorted_list;
+ struct record_cluster *sorted_ptr;
NMEM nmem;
YAZ_MUTEX mutex;
};
{
struct record_cluster *record;
struct reclist_bucket *hash_next;
- struct reclist_bucket *sorted_next;
- struct reclist_sortparms *sort_parms;
};
static void append_merge_keys(struct record_metadata_attr **p,
- struct record_metadata_attr *a,
+ const struct record_metadata_attr *a,
NMEM nmem)
{
- while (*p)
- p = &(*p)->next;
for (; a; a = a->next)
{
- *p = (struct record_metadata_attr *) nmem_malloc(nmem, sizeof(**p));
- (*p)->name = nmem_strdup_null(nmem, a->name);
- (*p)->value = nmem_strdup_null(nmem, a->value);
- p = &(*p)->next;
+ struct record_metadata_attr **pi = p;
+ for (; *pi; pi = &(*pi)->next)
+ if (!strcmp((*pi)->value, a->value))
+ break;
+ if (!*pi)
+ {
+ *pi = (struct record_metadata_attr *) nmem_malloc(nmem, sizeof(**p));
+ (*pi)->name = nmem_strdup_null(nmem, a->name);
+ (*pi)->value = nmem_strdup_null(nmem, a->value);
+ (*pi)->next = 0;
+ }
}
- *p = 0;
}
struct reclist_sortparms *reclist_parse_sortparms(NMEM nmem, const char *parms,
static int reclist_cmp(const void *p1, const void *p2)
{
- struct reclist_sortparms *sortparms =
- (*(struct reclist_bucket **) p1)->sort_parms;
- struct record_cluster *r1 = (*(struct reclist_bucket**) p1)->record;
- struct record_cluster *r2 = (*(struct reclist_bucket**) p2)->record;
+ struct record_cluster *r1 = (*(struct record_cluster**) p1);
+ struct record_cluster *r2 = (*(struct record_cluster**) p2);
+ struct reclist_sortparms *sortparms = r1->sort_parms;
struct reclist_sortparms *s;
int res = 0;
{
unsigned i;
int num = 0;
- struct reclist_bucket **pp = &l->sorted_list;
+ struct record_cluster **pp = &l->sorted_list;
reclist_enter(l);
{
struct reclist_bucket *p;
for (p = l->hashtable[i]; p; p = p->hash_next)
+ p->record->sorted_next = 0;
+ }
+ for (i = 0; i < l->hash_size; i++)
+ {
+ struct reclist_bucket *p;
+ for (p = l->hashtable[i]; p; p = p->hash_next)
{
if (session_check_cluster_limit(se, p->record))
{
- *pp = p;
- pp = &p->sorted_next;
- num++;
+ if (!p->record->sorted_next)
+ {
+ *pp = p->record;
+ pp = &p->record->sorted_next;
+ *pp = p->record; /* signal already in use */
+ num++;
+ }
}
}
}
void reclist_sort(struct reclist *l, struct reclist_sortparms *parms)
{
- struct reclist_bucket **flatlist = xmalloc(sizeof(*flatlist) * l->num_records);
- struct reclist_bucket *ptr;
- struct reclist_bucket **prev;
+ struct record_cluster **flatlist = xmalloc(sizeof(*flatlist) * l->num_records);
+ struct record_cluster *ptr;
+ struct record_cluster **prev;
int i = 0;
reclist_enter(l);
{
if (l && l->sorted_ptr)
{
- struct record_cluster *t = l->sorted_ptr->record;
+ struct record_cluster *t = l->sorted_ptr;
l->sorted_ptr = l->sorted_ptr->sorted_next;
return t;
}
{
wrbuf_destroy(p->record->relevance_explain1);
wrbuf_destroy(p->record->relevance_explain2);
+ p->record->relevance_explain1 = 0;
+ p->record->relevance_explain2 = 0;
}
}
yaz_mutex_destroy(&l->mutex);
return 0;
}
+static void merge_cluster(struct reclist *l,
+ struct relevance *r,
+ struct record_cluster *dst,
+ struct record_cluster *src)
+{
+ struct record_metadata_attr *mkl;
+ struct record **rp = &dst->records;
+ for (; *rp; rp = &(*rp)->next)
+ ;
+ *rp = src->records;
+
+ for (mkl = src->merge_keys; mkl; mkl = mkl->next)
+ {
+ const char *merge_key = mkl->value;
+ unsigned int bucket =
+ jenkins_hash((unsigned char*) merge_key) % l->hash_size;
+ struct reclist_bucket *p;
+
+ for (p = l->hashtable[bucket]; p; p = p->hash_next)
+ if (p->record == src)
+ p->record = dst;
+ }
+
+ /* not merging metadata and sortkeys yet */
+
+ relevance_mergerec(r, dst, src);
+
+ wrbuf_puts(dst->relevance_explain1, wrbuf_cstr(src->relevance_explain1));
+ wrbuf_puts(dst->relevance_explain2, wrbuf_cstr(src->relevance_explain2));
+
+ wrbuf_destroy(src->relevance_explain1);
+ src->relevance_explain1 = 0;
+ wrbuf_destroy(src->relevance_explain2);
+ src->relevance_explain2 = 0;
+
+ append_merge_keys(&dst->merge_keys, src->merge_keys, l->nmem);
+}
+
+static struct record_cluster *new_cluster(
+ struct reclist *l,
+ struct relevance *r,
+ struct conf_service *service,
+ struct record *record,
+ struct record_metadata_attr *merge_keys
+ )
+{
+ struct record_cluster *cluster;
+ cluster = nmem_malloc(l->nmem, sizeof(*cluster));
+
+ record->next = 0;
+ cluster->records = record;
+ cluster->merge_keys = 0;
+ append_merge_keys(&cluster->merge_keys, merge_keys, l->nmem);
+ cluster->relevance_score = 0;
+ cluster->recid = cluster->merge_keys->value;
+ cluster->metadata =
+ nmem_malloc(l->nmem,
+ sizeof(struct record_metadata*) * service->num_metadata);
+ memset(cluster->metadata, 0,
+ sizeof(struct record_metadata*) * service->num_metadata);
+ cluster->sortkeys =
+ nmem_malloc(l->nmem, sizeof(struct record_metadata*) * service->num_sortkeys);
+ memset(cluster->sortkeys, 0,
+ sizeof(union data_types*) * service->num_sortkeys);
+ relevance_newrec(r, cluster);
+ cluster->relevance_explain1 = wrbuf_alloc();
+ cluster->relevance_explain2 = wrbuf_alloc();
+ /* attach to hash list */
+ l->num_records++;
+ l->sorted_list = l->sorted_ptr = 0;
+ return cluster;
+}
+
// Insert a record. Return record cluster (newly formed or pre-existing)
struct record_cluster *reclist_insert(struct reclist *l,
+ struct relevance *r,
struct conf_service *service,
struct record *record,
struct record_metadata_attr *merge_keys,
{
struct record_cluster *cluster = 0;
struct record_metadata_attr *mkl = merge_keys;
- struct reclist_bucket **p;
assert(service);
assert(l);
const char *merge_key = mkl->value;
unsigned int bucket =
jenkins_hash((unsigned char*) merge_key) % l->hash_size;
+ struct reclist_bucket **p;
+ struct reclist_bucket *rb = 0;
for (p = &l->hashtable[bucket]; *p; p = &(*p)->hash_next)
{
{
struct record **re;
- cluster = (*p)->record;
- for (re = &cluster->records; *re; re = &(*re)->next)
+ rb = *p;
+ for (re = &rb->record->records; *re; re = &(*re)->next)
{
if ((*re)->client == record->client &&
record_compare(record, *re, service))
return 0;
}
}
- *re = record;
- record->next = 0;
- goto out;
+
+ if (!cluster)
+ {
+ cluster = rb->record;
+ *re = record;
+ record->next = 0;
+ yaz_log(YLOG_LOG, "reclist: record insert %p", cluster);
+ }
+ else
+ {
+ if (cluster != rb->record)
+ {
+ assert(rb->record->relevance_explain1);
+ yaz_log(YLOG_LOG, "reclist: cluster merge %p %p", cluster, rb->record);
+ merge_cluster(l, r, cluster, rb->record);
+ (*total)--;
+ }
+ }
}
}
}
- }
-out:
- if (!cluster)
- {
- struct reclist_bucket *new =
- nmem_malloc(l->nmem, sizeof(*new));
-
- cluster = nmem_malloc(l->nmem, sizeof(*cluster));
-
- record->next = 0;
- new->record = cluster;
- new->hash_next = 0;
- cluster->records = record;
-
- cluster->merge_keys = 0;
- append_merge_keys(&cluster->merge_keys, merge_keys, l->nmem);
-
- cluster->relevance_score = 0;
- cluster->term_frequency_vec = 0;
- cluster->recid = merge_keys->value;
- (*total)++;
- cluster->metadata =
- nmem_malloc(l->nmem,
- sizeof(struct record_metadata*) * service->num_metadata);
- memset(cluster->metadata, 0,
- sizeof(struct record_metadata*) * service->num_metadata);
- cluster->sortkeys =
- nmem_malloc(l->nmem, sizeof(struct record_metadata*) * service->num_sortkeys);
- memset(cluster->sortkeys, 0,
- sizeof(union data_types*) * service->num_sortkeys);
-
- cluster->relevance_explain1 = wrbuf_alloc();
- cluster->relevance_explain2 = wrbuf_alloc();
- /* attach to hash list */
- *p = new;
- l->num_records++;
- l->sorted_list = l->sorted_ptr = 0;
+ if (!cluster)
+ {
+ (*total)++;
+ cluster = new_cluster(l, r, service, record, merge_keys);
+ yaz_log(YLOG_LOG, "reclist: new cluster p=%p", cluster);
+ }
+
+ if (!rb)
+ {
+ rb = nmem_malloc(l->nmem, sizeof(*rb));
+ rb->record = cluster;
+ rb->hash_next = 0;
+
+ *p = rb;
+ }
}
yaz_mutex_leave(l->mutex);
return cluster;