-/* $Id: reclists.c,v 1.11 2007-04-17 07:52:03 marc Exp $
- Copyright (c) 2006-2007, Index Data.
-
-This file is part of Pazpar2.
+/* This file is part of Pazpar2.
+ Copyright (C) Index Data
Pazpar2 is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
for more details.
You should have received a copy of the GNU General Public License
-along with Pazpar2; see the file LICENSE. If not, write to the
-Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA
-02111-1307, USA.
- */
+along with this program; if not, write to the Free Software
+Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+
+*/
#include <assert.h>
#if HAVE_CONFIG_H
-#include <cconfig.h>
+#include <config.h>
#endif
#include <yaz/yaz-util.h>
-#include "pazpar2.h"
+#include "ppmutex.h"
+#include "session.h"
#include "reclists.h"
+#include "jenkins_hash.h"
-extern struct parameters global_parameters;
-
-// Not threadsafe
-static struct reclist_sortparms *sortparms = 0;
+struct reclist
+{
+ struct reclist_bucket **hashtable;
+ unsigned hash_size;
+
+ int num_records;
+ struct record_cluster *sorted_list;
+ struct record_cluster *sorted_ptr;
+ NMEM nmem;
+ YAZ_MUTEX mutex;
+};
struct reclist_bucket
{
struct record_cluster *record;
- struct reclist_bucket *next;
+ struct reclist_bucket *hash_next;
};
-struct reclist_sortparms *reclist_parse_sortparms(NMEM nmem, const char *parms)
+static void append_merge_keys(struct record_metadata_attr **p,
+ const struct record_metadata_attr *a,
+ NMEM nmem)
+{
+ for (; a; a = a->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;
+ }
+ }
+}
+
+struct reclist_sortparms *reclist_parse_sortparms(NMEM nmem, const char *parms,
+ struct conf_service *service)
{
struct reclist_sortparms *res = 0;
struct reclist_sortparms **rp = &res;
- struct conf_service *service = config->servers->service;
if (strlen(parms) > 256)
return 0;
char parm[256];
char *pp;
const char *cpp;
- int increasing;
+ int increasing = 0;
int i;
- int offset;
- enum conf_sortkey_type type;
+ int offset = 0;
+ enum conf_metadata_type type = Metadata_type_generic;
struct reclist_sortparms *new;
if (!(cpp = strchr(parms, ',')))
cpp = parms + strlen(parms);
- strncpy(parm, parms, cpp - parms);
+ strncpy(parm, parms, cpp - parms);
parm[cpp-parms] = '\0';
if ((pp = strchr(parm, ':')))
{
- increasing = pp[1] == '1' ? 1 : 0;
+ if (pp[1] == '1')
+ increasing = 1;
+ else if (pp[1] == '0')
+ increasing = 0;
+ else
+ {
+ yaz_log(YLOG_FATAL, "Bad sortkey modifier: %s", parm);
+ return 0;
+ }
+
+ if (pp[2])
+ {
+ if (pp[2] == 'p')
+ type = Metadata_type_position;
+ else
+ yaz_log(YLOG_FATAL, "Bad sortkey modifier: %s", parm);
+ }
*pp = '\0';
}
- else
- increasing = 0;
- if (!strcmp(parm, "relevance"))
- {
- type = Metadata_sortkey_relevance;
- offset = -1;
- }
- else
+ if (type != Metadata_type_position)
{
- for (i = 0; i < service->num_sortkeys; i++)
+ if (!strcmp(parm, "relevance"))
+ {
+ type = Metadata_type_relevance;
+ }
+ else if (!strcmp(parm, "position"))
+ {
+ type = Metadata_type_position;
+ }
+ else if (!strcmp(parm, "retrieval"))
{
- struct conf_sortkey *sk = &service->sortkeys[i];
- if (!strcmp(sk->name, parm))
+ type = Metadata_type_retrieval;
+ }
+ else if (service)
+ {
+ for (i = 0; i < service->num_sortkeys; i++)
+ {
+ struct conf_sortkey *sk = &service->sortkeys[i];
+ if (!strcmp(sk->name, parm))
+ {
+ type = sk->type;
+ break;
+ }
+ }
+ if (i >= service->num_sortkeys)
{
- type = sk->type;
- if (type == Metadata_sortkey_skiparticle)
- type = Metadata_sortkey_string;
- break;
+ yaz_log(YLOG_FATAL, "Sortkey not defined in service: %s",
+ parm);
+ return 0;
}
+ offset = i;
}
- if (i >= service->num_sortkeys)
+ else
{
- yaz_log(YLOG_FATAL, "Bad sortkey: %s", parm);
+ yaz_log(YLOG_FATAL, "Sortkey not defined in service: %s",
+ parm);
return 0;
}
- else
- offset = i;
}
new = *rp = nmem_malloc(nmem, sizeof(struct reclist_sortparms));
new->next = 0;
new->offset = offset;
new->type = type;
new->increasing = increasing;
+ new->name = nmem_strdup(nmem, parm);
rp = &new->next;
if (*(parms = cpp))
parms++;
{
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;
- for (s = sortparms; s; s = s->next)
+ for (s = sortparms; s && res == 0; s = s->next)
{
- int res;
+ union data_types *ut1 = r1->sortkeys[s->offset];
+ union data_types *ut2 = r2->sortkeys[s->offset];
+ const char *s1, *s2;
switch (s->type)
{
- case Metadata_sortkey_relevance:
- res = r2->relevance - r1->relevance;
- break;
- case Metadata_sortkey_string:
- res = strcmp(r2->sortkeys[s->offset]->text, r1->sortkeys[s->offset]->text);
- break;
- case Metadata_sortkey_numeric:
+ case Metadata_type_relevance:
+ res = r1->relevance_score - r2->relevance_score;
+ break;
+ case Metadata_type_generic:
+ case Metadata_type_skiparticle:
+ s1 = ut1 ? ut1->text.sort : "";
+ s2 = ut2 ? ut2->text.sort : "";
+ res = strcmp(s1, s2);
+ break;
+ case Metadata_type_year:
+ case Metadata_type_date:
+ if (ut1 && ut2)
+ {
+ if (s->increasing)
+ res = ut1->number.min - ut2->number.min;
+ else
+ res = ut1->number.max - ut2->number.max;
+ }
+ else if (ut1 && !ut2)
+ {
+ res = -1; /* without date/year: last! */
+ continue;
+ }
+ else if (!ut1 && ut2)
+ {
+ res = 1; /* without date/year: last! */
+ continue;
+ }
+ else
res = 0;
- break;
- default:
- yaz_log(YLOG_FATAL, "Bad sort type: %d", s->type);
- exit(1);
+ break;
+ case Metadata_type_position:
+ if (r1->records && r2->records)
+ {
+ int pos1 = 0, pos2 = 0;
+ struct record *rec;
+ for (rec = r1->records; rec; rec = rec->next)
+ if (pos1 == 0 || rec->position < pos1)
+ pos1 = rec->position;
+ for (rec = r2->records; rec; rec = rec->next)
+ if (pos2 == 0 || rec->position < pos2)
+ pos2 = rec->position;
+ res = pos1 - pos2;
+ continue;
+ }
+ break;
+ case Metadata_type_float:
+ if (ut1 && ut2)
+ {
+ if (ut1->fnumber == ut2->fnumber)
+ res = 0;
+ else if (ut1->fnumber > ut2->fnumber)
+ res = 1;
+ else
+ res = -1;
+ }
+ else if (ut1)
+ res = 1;
+ else if (ut2)
+ res = -1;
+ else
+ res = 0;
+ break;
+ case Metadata_type_retrieval:
+ res = r1->retrieval_position - r2->retrieval_position;
+ break;
}
- if (res)
+ if (res && !s->increasing)
+ res *= -1;
+ }
+ if (res == 0)
+ res = strcmp(r1->recid, r2->recid);
+ return res;
+}
+
+void reclist_limit(struct reclist *l, struct session *se, int lazy)
+{
+ unsigned i;
+ int num = 0;
+ struct record_cluster **pp = &l->sorted_list;
+
+ reclist_enter(l);
+
+ if (!lazy || !*pp)
+ {
+ for (i = 0; i < l->hash_size; i++)
+ {
+ 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++)
{
- if (s->increasing)
- res *= -1;
- return res;
+ struct reclist_bucket *p;
+ for (p = l->hashtable[i]; p; p = p->hash_next)
+ {
+ if (session_check_cluster_limit(se, p->record))
+ {
+ if (!p->record->sorted_next)
+ {
+ *pp = p->record;
+ pp = &p->record->sorted_next;
+ *pp = p->record; /* signal already in use */
+ num++;
+ }
+ }
+ }
}
+ *pp = 0;
}
- return 0;
+ l->num_records = num;
+ reclist_leave(l);
}
void reclist_sort(struct reclist *l, struct reclist_sortparms *parms)
{
- sortparms = parms;
- qsort(l->flatlist, l->num_records, sizeof(struct record_cluster*), reclist_cmp);
- reclist_rewind(l);
+ struct record_cluster **flatlist = xmalloc(sizeof(*flatlist) * l->num_records);
+ struct record_cluster *ptr;
+ struct record_cluster **prev;
+ int i = 0;
+
+ reclist_enter(l);
+
+ ptr = l->sorted_list;
+ prev = &l->sorted_list;
+ while (ptr)
+ {
+ ptr->sort_parms = parms;
+ flatlist[i] = ptr;
+ ptr = ptr->sorted_next;
+ i++;
+ }
+ assert(i == l->num_records);
+
+ qsort(flatlist, l->num_records, sizeof(*flatlist), reclist_cmp);
+ for (i = 0; i < l->num_records; i++)
+ {
+ *prev = flatlist[i];
+ prev = &flatlist[i]->sorted_next;
+ }
+ *prev = 0;
+
+ xfree(flatlist);
+
+ reclist_leave(l);
}
struct record_cluster *reclist_read_record(struct reclist *l)
{
- if (l->pointer < l->num_records)
- return l->flatlist[l->pointer++];
+ if (l && l->sorted_ptr)
+ {
+ struct record_cluster *t = l->sorted_ptr;
+ l->sorted_ptr = l->sorted_ptr->sorted_next;
+ return t;
+ }
else
return 0;
}
+void reclist_enter(struct reclist *l)
+{
+ yaz_mutex_enter(l->mutex);
+ l->sorted_ptr = l->sorted_list;
+}
+
+
+void reclist_leave(struct reclist *l)
+{
+ yaz_mutex_leave(l->mutex);
+ l->sorted_ptr = l->sorted_list;
+}
+
void reclist_rewind(struct reclist *l)
{
- l->pointer = 0;
+ l->sorted_ptr = l->sorted_list;
}
-// Jenkins one-at-a-time hash (from wikipedia)
-static unsigned int hash(const unsigned char *key)
+
+struct reclist *reclist_create(NMEM nmem)
{
- unsigned int hash = 0;
+ struct reclist *res = nmem_malloc(nmem, sizeof(struct reclist));
+ res->hash_size = 399;
+ res->hashtable
+ = nmem_malloc(nmem, res->hash_size * sizeof(struct reclist_bucket*));
+ memset(res->hashtable, 0, res->hash_size * sizeof(struct reclist_bucket*));
+ res->nmem = nmem;
+
+ res->sorted_ptr = 0;
+ res->sorted_list = 0;
+
+ res->num_records = 0;
+ res->mutex = 0;
+ pazpar2_mutex_create(&res->mutex, "reclist");
+ return res;
+}
- while (*key)
+void reclist_destroy(struct reclist *l)
+{
+ if (l)
{
- hash += *(key++);
- hash += (hash << 10);
- hash ^= (hash >> 6);
+ unsigned i;
+ for (i = 0; i < l->hash_size; i++)
+ {
+ struct reclist_bucket *p;
+ for (p = l->hashtable[i]; p; p = p->hash_next)
+ {
+ 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);
}
- hash += (hash << 3);
- hash ^= (hash >> 11);
- hash += (hash << 15);
- return hash;
}
-struct reclist *reclist_create(NMEM nmem, int numrecs)
+int reclist_get_num_records(struct reclist *l)
{
- int hashsize = 1;
- struct reclist *res;
-
- assert(numrecs);
- while (hashsize < numrecs)
- hashsize <<= 1;
- res = nmem_malloc(nmem, sizeof(struct reclist));
- res->hashtable = nmem_malloc(nmem, hashsize * sizeof(struct reclist_bucket*));
- memset(res->hashtable, 0, hashsize * sizeof(struct reclist_bucket*));
- res->hashtable_size = hashsize;
- res->nmem = nmem;
- res->hashmask = hashsize - 1; // Creates a bitmask
+ if (l)
+ return l->num_records;
+ return 0;
+}
- res->num_records = 0;
- res->flatlist = nmem_malloc(nmem, numrecs * sizeof(struct record_cluster*));
- res->flatlist_size = numrecs;
+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;
- return res;
+ 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->retrieval_position = l->num_records;
+ 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 conf_service *service,
- struct record *record,
- char *merge_key, int *total)
+struct record_cluster *reclist_insert(struct reclist *l,
+ struct relevance *r,
+ struct conf_service *service,
+ struct record *record,
+ struct record_metadata_attr *merge_keys,
+ int *total)
{
- unsigned int bucket;
- struct reclist_bucket **p;
struct record_cluster *cluster = 0;
-
+ struct record_metadata_attr *mkl = merge_keys;
+
assert(service);
- assert(service->num_metadata);
- assert(service->num_sortkeys);
- assert(merge_key);
+ assert(l);
+ assert(record);
+ assert(merge_keys);
+ assert(total);
+ yaz_mutex_enter(l->mutex);
- bucket = hash((unsigned char*) merge_key) & l->hashmask;
- for (p = &l->hashtable[bucket]; *p; p = &(*p)->next)
+ for (; mkl; mkl = mkl->next)
{
- // We found a matching record. Merge them
- if (!strcmp(merge_key, (*p)->record->merge_key))
+ 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_cluster *existing = (*p)->record;
- record->next = existing->records;
- existing->records = record;
- cluster = existing;
- break;
+ struct record_metadata_attr *mkr = (*p)->record->merge_keys;
+ for (; mkr; mkr = mkr->next)
+ {
+ // We found a matching record. Merge them
+ if (!strcmp(merge_key, mkr->value))
+ {
+ rb = *p;
+ if (!cluster)
+ {
+ struct record **re;
+ for (re = &rb->record->records; *re; re = &(*re)->next)
+ {
+ if ((*re)->client == record->client &&
+ record_compare(record, *re, service))
+ {
+ yaz_mutex_leave(l->mutex);
+ return 0;
+ }
+ }
+ cluster = rb->record;
+ *re = record;
+ record->next = 0;
+ }
+ else
+ {
+ if (cluster != rb->record)
+ {
+ assert(rb->record->relevance_explain1);
+ merge_cluster(l, r, cluster, rb->record);
+ (*total)--;
+ }
+ }
+ }
+ }
+ }
+ if (!cluster)
+ {
+ (*total)++;
+ cluster = new_cluster(l, r, service, record, merge_keys);
+ }
+
+ if (!rb)
+ {
+ rb = nmem_malloc(l->nmem, sizeof(*rb));
+ rb->record = cluster;
+ rb->hash_next = 0;
+
+ *p = rb;
}
}
- if (!cluster && l->num_records < l->flatlist_size)
- {
- struct reclist_bucket *new =
- nmem_malloc(l->nmem, sizeof(struct reclist_bucket));
- struct record_cluster *newc =
- nmem_malloc(l->nmem, sizeof(struct record_cluster));
-
- record->next = 0;
- new->record = newc;
- new->next = 0;
- newc->records = record;
- newc->merge_key = merge_key;
- newc->relevance = 0;
- newc->term_frequency_vec = 0;
- newc->recid = (*total)++;
- newc->metadata = nmem_malloc(l->nmem,
- sizeof(struct record_metadata*) * service->num_metadata);
- memset(newc->metadata, 0, sizeof(struct record_metadata*) * service->num_metadata);
- newc->sortkeys = nmem_malloc(l->nmem,
- sizeof(struct record_metadata*) * service->num_sortkeys);
- memset(newc->sortkeys, 0, sizeof(union data_types*) * service->num_sortkeys);
-
- *p = new;
- l->flatlist[l->num_records++] = newc;
- cluster = newc;
- }
+ yaz_mutex_leave(l->mutex);
return cluster;
}
-
+int reclist_sortparms_cmp(struct reclist_sortparms *sort1, struct reclist_sortparms *sort2)
+{
+ int rc;
+ if (sort1 == sort2)
+ return 0;
+ if (sort1 == 0 || sort2 == 0)
+ return 1;
+ rc = strcmp(sort1->name, sort2->name) || sort1->increasing != sort2->increasing || sort1->type != sort2->type;
+ return rc;
+}
/*
* Local variables:
* c-basic-offset: 4
+ * c-file-style: "Stroustrup"
* indent-tabs-mode: nil
* End:
* vim: shiftwidth=4 tabstop=8 expandtab
*/
+
projects / pazpar2-moved-to-github.git / blobdiff