X-Git-Url: http://git.indexdata.com/?a=blobdiff_plain;f=src%2Freclists.c;h=245d06187192dd1f46473169dc2046f6ab09ab99;hb=a3f57f20fb1ade51cc5a99aafcc16af1a1f5b0c8;hp=40f1d76e7695b85dc4e7f5abd353ad1e3e9c386a;hpb=fe3383c0559a453df1e5076fc6faac6d1a11685c;p=pazpar2-moved-to-github.git

diff --git a/src/reclists.c b/src/reclists.c
index 40f1d76..245d061 100644
--- a/src/reclists.c
+++ b/src/reclists.c
@@ -1,110 +1,576 @@
-/*
- * $Id: reclists.c,v 1.1 2006-12-20 20:47:16 quinn Exp $
- */
+/* 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
+Software Foundation; either version 2, or (at your option) any later
+version.
+
+Pazpar2 is distributed in the hope that it will be useful, but WITHOUT ANY
+WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+for more details.
+
+You should have received a copy of the GNU General Public License
+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 <config.h>
+#endif
+
 #include <yaz/yaz-util.h>
 
-#include "pazpar2.h"
+#include "ppmutex.h"
+#include "session.h"
 #include "reclists.h"
+#include "jenkins_hash.h"
+
+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 *record;
-    struct reclist_bucket *next;
+    struct record_cluster *record;
+    struct reclist_bucket *hash_next;
 };
 
-struct record *reclist_read_record(struct reclist *l)
+static void append_merge_keys(struct record_metadata_attr **p,
+                              const struct record_metadata_attr *a,
+                              NMEM nmem)
 {
-    if (l->pointer < l->num_records)
-        return l->flatlist[l->pointer++];
-    else
+    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;
+
+    if (strlen(parms) > 256)
         return 0;
+    while (*parms)
+    {
+        char parm[256];
+        char *pp;
+        const char *cpp;
+        int increasing = 0;
+        int i;
+        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);
+        parm[cpp-parms] = '\0';
+
+        if ((pp = strchr(parm, ':')))
+        {
+            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';
+        }
+        if (type != Metadata_type_position)
+        {
+            if (!strcmp(parm, "relevance"))
+            {
+                type = Metadata_type_relevance;
+            }
+            else if (!strcmp(parm, "position"))
+            {
+                type = Metadata_type_position;
+            }
+            else if (!strcmp(parm, "retrieval"))
+            {
+                type = Metadata_type_retrieval;
+            }
+            else
+            {
+                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)
+                {
+                    yaz_log(YLOG_FATAL, "Sortkey not defined in service: %s",
+                            parm);
+                    return 0;
+                }
+                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++;
+    }
+    return res;
 }
 
-void reclist_rewind(struct reclist *l)
+static int reclist_cmp(const void *p1, const void *p2)
 {
-    l->pointer = 0;
+    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 && res == 0; s = s->next)
+    {
+        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_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;
+        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 && !s->increasing)
+            res *= -1;
+    }
+    if (res == 0)
+        res = strcmp(r1->recid, r2->recid);
+    return res;
 }
 
-// Jenkins one-at-a-time hash (from wikipedia)
-static unsigned int hash(const unsigned char *key)
+void reclist_limit(struct reclist *l, struct session *se, int lazy)
 {
-    unsigned int hash = 0;
+    unsigned i;
+    int num = 0;
+    struct record_cluster **pp = &l->sorted_list;
+
+    reclist_enter(l);
 
-    while (*key)
+    if (!lazy || !*pp)
     {
-        hash += *(key++);
-        hash += (hash << 10);
-        hash ^= (hash >> 6);
+        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++)
+        {
+            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;
     }
-    hash += (hash << 3);
-    hash ^= (hash >> 11);
-    hash += (hash << 15);
-    return hash;
+    l->num_records = num;
+    reclist_leave(l);
 }
 
-struct reclist *reclist_create(NMEM nmem, int numrecs)
+void reclist_sort(struct reclist *l, struct reclist_sortparms *parms)
 {
-    int hashsize = 1;
-    struct reclist *res;
+    struct record_cluster **flatlist = xmalloc(sizeof(*flatlist) * l->num_records);
+    struct record_cluster *ptr;
+    struct record_cluster **prev;
+    int i = 0;
+
+    reclist_enter(l);
 
-    assert(numrecs);
-    while (hashsize < numrecs)
-        hashsize <<= 1;
-    res = nmem_malloc(nmem, sizeof(struct reclist));
-    res->hashtable = nmem_malloc(nmem, hashsize * sizeof(struct reclist_bucket*));
-    bzero(res->hashtable, hashsize * sizeof(struct reclist_bucket*));
-    res->hashtable_size = hashsize;
+    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 && 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->sorted_ptr = l->sorted_list;
+}
+
+
+struct reclist *reclist_create(NMEM nmem)
+{
+    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->hashmask = hashsize - 1; // Creates a bitmask
 
-    res->num_records = 0;
-    res->flatlist = nmem_malloc(nmem, numrecs * sizeof(struct record*));
-    res->flatlist_size = numrecs;
+    res->sorted_ptr = 0;
+    res->sorted_list = 0;
 
+    res->num_records = 0;
+    res->mutex = 0;
+    pazpar2_mutex_create(&res->mutex, "reclist");
     return res;
 }
 
-struct record *reclist_insert(struct reclist *l, struct record  *record)
+void reclist_destroy(struct reclist *l)
 {
-    unsigned int bucket;
-    struct reclist_bucket **p;
-    struct record *head;
-
-    bucket = hash((unsigned char*) record->merge_key) & l->hashmask;
-    for (p = &l->hashtable[bucket]; *p; p = &(*p)->next)
+    if (l)
     {
-        // We found a matching record. Merge them
-        if (!strcmp(record->merge_key, (*p)->record->merge_key))
+        unsigned i;
+        for (i = 0; i < l->hash_size; i++)
         {
-            struct record *existing = (*p)->record;
-            record->next_cluster = existing->next_cluster;
-            existing->next_cluster = record;
-            head = existing;
-            break;
+            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);
     }
-    if (!*p) // We made it to the end of the bucket without finding match
+}
+
+int reclist_get_num_records(struct reclist *l)
+{
+    if (l)
+        return l->num_records;
+    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)
     {
-        struct reclist_bucket *new = nmem_malloc(l->nmem,
-                sizeof(struct reclist_bucket));
-        new->record = record;
-        record->next_cluster = 0;
-        new->next = 0;
-        *p = new;
-        l->flatlist[l->num_records++] = record;
-        head = record;
+        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;
     }
-    return head;
+
+    /* 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 relevance *r,
+                                      struct conf_service *service,
+                                      struct record *record,
+                                      struct record_metadata_attr *merge_keys,
+                                      int *total)
+{
+    struct record_cluster *cluster = 0;
+    struct record_metadata_attr *mkl = merge_keys;
+
+    assert(service);
+    assert(l);
+    assert(record);
+    assert(merge_keys);
+    assert(total);
+
+    yaz_mutex_enter(l->mutex);
+
+    for (; 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;
+        struct reclist_bucket *rb = 0;
+
+        for (p = &l->hashtable[bucket]; *p; p = &(*p)->hash_next)
+        {
+            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;
+        }
+    }
+    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
  */
+