Fix session_nmem may be used by multiple threads PAZ-962
[pazpar2-moved-to-github.git] / src / session.c
1 /* This file is part of Pazpar2.
2    Copyright (C) Index Data
3
4 Pazpar2 is free software; you can redistribute it and/or modify it under
5 the terms of the GNU General Public License as published by the Free
6 Software Foundation; either version 2, or (at your option) any later
7 version.
8
9 Pazpar2 is distributed in the hope that it will be useful, but WITHOUT ANY
10 WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
12 for more details.
13
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17
18 */
19
20 /** \file session.c
21     \brief high-level logic; mostly user sessions and settings
22 */
23
24 #if HAVE_CONFIG_H
25 #include <config.h>
26 #endif
27
28 #include <time.h>
29 #include <stdlib.h>
30 #include <stdio.h>
31 #include <string.h>
32 #if HAVE_SYS_TIME_H
33 #include <sys/time.h>
34 #endif
35 #if HAVE_UNISTD_H
36 #include <unistd.h>
37 #endif
38 #ifdef WIN32
39 #include <windows.h>
40 #endif
41 #include <signal.h>
42 #include <ctype.h>
43 #include <assert.h>
44 #include <math.h>
45
46 #include <yaz/marcdisp.h>
47 #include <yaz/comstack.h>
48 #include <yaz/tcpip.h>
49 #include <yaz/proto.h>
50 #include <yaz/readconf.h>
51 #include <yaz/pquery.h>
52 #include <yaz/otherinfo.h>
53 #include <yaz/yaz-util.h>
54 #include <yaz/nmem.h>
55 #include <yaz/query-charset.h>
56 #include <yaz/querytowrbuf.h>
57 #include <yaz/oid_db.h>
58 #include <yaz/snprintf.h>
59
60 #define USE_TIMING 0
61 #if USE_TIMING
62 #include <yaz/timing.h>
63 #endif
64
65 #include "ppmutex.h"
66 #include "parameters.h"
67 #include "session.h"
68 #include "eventl.h"
69 #include "http.h"
70 #include "termlists.h"
71 #include "reclists.h"
72 #include "relevance.h"
73 #include "database.h"
74 #include "client.h"
75 #include "settings.h"
76 #include "normalize7bit.h"
77
78 #include <libxml/tree.h>
79
80 #define MAX_CHUNK 15
81
82 #define MAX(a,b) ((a)>(b)?(a):(b))
83
84 // Note: Some things in this structure will eventually move to configuration
85 struct parameters global_parameters =
86 {
87     0,   // dump_records
88     0,   // debug_mode
89     0,   // predictable sessions
90 };
91
92 struct client_list {
93     struct client *client;
94     struct client_list *next;
95 };
96
97 /* session counting (1) , disable client counting (0) */
98 static YAZ_MUTEX g_session_mutex = 0;
99 static int no_sessions = 0;
100
101 static int session_use(int delta)
102 {
103     int sessions;
104     if (!g_session_mutex)
105         yaz_mutex_create(&g_session_mutex);
106     yaz_mutex_enter(g_session_mutex);
107     no_sessions += delta;
108     sessions = no_sessions;
109     yaz_mutex_leave(g_session_mutex);
110     yaz_log(YLOG_DEBUG, "%s sessions=%d", delta == 0 ? "" :
111             (delta > 0 ? "INC" : "DEC"), no_sessions);
112     return sessions;
113 }
114
115 int sessions_count(void)
116 {
117     return session_use(0);
118 }
119
120 static void log_xml_doc(xmlDoc *doc)
121 {
122     FILE *lf = yaz_log_file();
123     xmlChar *result = 0;
124     int len = 0;
125 #if LIBXML_VERSION >= 20600
126     xmlDocDumpFormatMemory(doc, &result, &len, 1);
127 #else
128     xmlDocDumpMemory(doc, &result, &len);
129 #endif
130     if (lf && len)
131     {
132         (void) fwrite(result, 1, len, lf);
133         fprintf(lf, "\n");
134     }
135     xmlFree(result);
136 }
137
138 static void session_enter(struct session *s, const char *caller)
139 {
140     if (caller)
141         session_log(s, YLOG_DEBUG, "Session lock by %s", caller);
142     yaz_mutex_enter(s->session_mutex);
143 }
144
145 static void session_leave(struct session *s, const char *caller)
146 {
147     yaz_mutex_leave(s->session_mutex);
148     if (caller)
149         session_log(s, YLOG_DEBUG, "Session unlock by %s", caller);
150 }
151
152 static void session_normalize_facet(struct session *s,
153                                     const char *type, const char *value,
154                                     WRBUF display_wrbuf, WRBUF facet_wrbuf)
155 {
156     struct conf_service *service = s->service;
157     pp2_charset_token_t prt;
158     const char *facet_component;
159     int i;
160     const char *icu_chain_id = 0;
161
162     for (i = 0; i < service->num_metadata; i++)
163         if (!strcmp((service->metadata + i)->name, type))
164             icu_chain_id = (service->metadata + i)->facetrule;
165     if (!icu_chain_id)
166         icu_chain_id = "facet";
167     prt = pp2_charset_token_create(service->charsets, icu_chain_id);
168     if (!prt)
169     {
170         session_log(s, YLOG_FATAL,
171                     "Unknown ICU chain '%s' for facet of type '%s'",
172                 icu_chain_id, type);
173         wrbuf_destroy(facet_wrbuf);
174         wrbuf_destroy(display_wrbuf);
175         return;
176     }
177     pp2_charset_token_first(prt, value, 0);
178     while ((facet_component = pp2_charset_token_next(prt)))
179     {
180         const char *display_component;
181         if (*facet_component)
182         {
183             if (wrbuf_len(facet_wrbuf))
184                 wrbuf_puts(facet_wrbuf, " ");
185             wrbuf_puts(facet_wrbuf, facet_component);
186         }
187         display_component = pp2_get_display(prt);
188         if (display_component)
189         {
190             if (wrbuf_len(display_wrbuf))
191                 wrbuf_puts(display_wrbuf, " ");
192             wrbuf_puts(display_wrbuf, display_component);
193         }
194     }
195     pp2_charset_token_destroy(prt);
196 }
197
198 void add_facet(struct session *s, const char *type, const char *value, int count)
199 {
200     WRBUF facet_wrbuf = wrbuf_alloc();
201     WRBUF display_wrbuf = wrbuf_alloc();
202
203     session_normalize_facet(s, type, value, display_wrbuf, facet_wrbuf);
204
205     if (wrbuf_len(facet_wrbuf))
206     {
207         int i;
208         for (i = 0; i < s->num_termlists; i++)
209             if (!strcmp(s->termlists[i].name, type))
210                 break;
211         if (i == s->num_termlists)
212         {
213             if (i == SESSION_MAX_TERMLISTS)
214             {
215                 session_log(s, YLOG_FATAL, "Too many termlists");
216                 wrbuf_destroy(facet_wrbuf);
217                 wrbuf_destroy(display_wrbuf);
218                 return;
219             }
220
221             s->termlists[i].name = nmem_strdup(s->nmem, type);
222             s->termlists[i].termlist = termlist_create(s->nmem);
223             s->num_termlists = i + 1;
224         }
225
226 #if 0
227         session_log(s, YLOG_LOG, "Facets for %s: %s norm:%s (%d)", type, value, wrbuf_cstr(facet_wrbuf), count);
228 #endif
229         termlist_insert(s->termlists[i].termlist, wrbuf_cstr(display_wrbuf),
230                         wrbuf_cstr(facet_wrbuf), count);
231     }
232     wrbuf_destroy(facet_wrbuf);
233     wrbuf_destroy(display_wrbuf);
234 }
235
236 static xmlDoc *record_to_xml(struct session *se,
237                              struct session_database *sdb, const char *rec)
238 {
239     struct database *db = sdb->database;
240     xmlDoc *rdoc = 0;
241
242     rdoc = xmlParseMemory(rec, strlen(rec));
243
244     if (!rdoc)
245     {
246         session_log(se, YLOG_WARN, "Non-wellformed XML");
247         return 0;
248     }
249
250     if (global_parameters.dump_records)
251     {
252         session_log(se, YLOG_LOG, "Un-normalized record from %s", db->id);
253         log_xml_doc(rdoc);
254     }
255
256     return rdoc;
257 }
258
259 #define MAX_XSLT_ARGS 16
260
261 // Add static values from session database settings if applicable
262 static void insert_settings_parameters(struct session_database *sdb,
263                                        struct conf_service *service,
264                                        char **parms,
265                                        NMEM nmem)
266 {
267     int i;
268     int nparms = 0;
269     int offset = 0;
270
271     for (i = 0; i < service->num_metadata; i++)
272     {
273         struct conf_metadata *md = &service->metadata[i];
274         int setting;
275
276         if (md->setting == Metadata_setting_parameter &&
277             (setting = settings_lookup_offset(service, md->name)) >= 0)
278         {
279             const char *val = session_setting_oneval(sdb, setting);
280             if (val && nparms < MAX_XSLT_ARGS)
281             {
282                 char *buf;
283                 int len = strlen(val);
284                 buf = nmem_malloc(nmem, len + 3);
285                 buf[0] = '\'';
286                 strcpy(buf + 1, val);
287                 buf[len+1] = '\'';
288                 buf[len+2] = '\0';
289                 parms[offset++] = md->name;
290                 parms[offset++] = buf;
291                 nparms++;
292             }
293         }
294     }
295     parms[offset] = 0;
296 }
297
298 // Add static values from session database settings if applicable
299 static void insert_settings_values(struct session_database *sdb, xmlDoc *doc,
300                                    xmlNode *root,
301                                    struct conf_service *service)
302 {
303     int i;
304
305     for (i = 0; i < service->num_metadata; i++)
306     {
307         struct conf_metadata *md = &service->metadata[i];
308         int offset;
309
310         if (md->setting == Metadata_setting_postproc &&
311             (offset = settings_lookup_offset(service, md->name)) >= 0)
312         {
313             const char *val = session_setting_oneval(sdb, offset);
314             if (val)
315             {
316                 xmlNode *n = xmlNewTextChild(root, 0, (xmlChar *) "metadata",
317                                              (xmlChar *) val);
318                 xmlSetProp(n, (xmlChar *) "type", (xmlChar *) md->name);
319             }
320         }
321     }
322 }
323
324 static xmlDoc *normalize_record(struct session *se,
325                                 struct session_database *sdb,
326                                 struct conf_service *service,
327                                 const char *rec, NMEM nmem)
328 {
329     xmlDoc *rdoc = record_to_xml(se, sdb, rec);
330
331     if (rdoc)
332     {
333         char *parms[MAX_XSLT_ARGS*2+1];
334
335         insert_settings_parameters(sdb, service, parms, nmem);
336
337         if (normalize_record_transform(sdb->map, &rdoc, (const char **)parms))
338         {
339             session_log(se, YLOG_WARN, "Normalize failed");
340         }
341     }
342     return rdoc;
343 }
344
345 void session_settings_dump(struct session *se,
346                            struct session_database *db,
347                            WRBUF w)
348 {
349     if (db->settings)
350     {
351         int i, num = db->num_settings;
352         for (i = 0; i < num; i++)
353         {
354             struct setting *s = db->settings[i];
355             for (;s ; s = s->next)
356             {
357                 wrbuf_puts(w, "<set name=\"");
358                 wrbuf_xmlputs(w, s->name);
359                 wrbuf_puts(w, "\" value=\"");
360                 wrbuf_xmlputs(w, s->value);
361                 wrbuf_puts(w, "\"/>");
362             }
363             if (db->settings[i])
364                 wrbuf_puts(w, "\n");
365         }
366     }
367 }
368
369 // Retrieve first defined value for 'name' for given database.
370 // Will be extended to take into account user associated with session
371 const char *session_setting_oneval(struct session_database *db, int offset)
372 {
373     if (offset >= db->num_settings || !db->settings[offset])
374         return "";
375     return db->settings[offset]->value;
376 }
377
378 // Prepare XSLT stylesheets for record normalization
379 // Structures are allocated on the session_wide nmem to avoid having
380 // to recompute this for every search. This would lead
381 // to leaking if a single session was to repeatedly change the PZ_XSLT
382 // setting. However, this is not a realistic use scenario.
383 static int prepare_map(struct session *se, struct session_database *sdb)
384 {
385     if (sdb->settings && !sdb->map)
386     {
387         const char *s;
388
389         if (sdb->settings[PZ_XSLT] &&
390             (s = session_setting_oneval(sdb, PZ_XSLT)))
391         {
392             char auto_stylesheet[256];
393
394             if (!strcmp(s, "auto"))
395             {
396                 const char *request_syntax = session_setting_oneval(
397                     sdb, PZ_REQUESTSYNTAX);
398                 if (request_syntax)
399                 {
400                     char *cp;
401                     yaz_snprintf(auto_stylesheet, sizeof(auto_stylesheet),
402                                  "%s.xsl", request_syntax);
403                     for (cp = auto_stylesheet; *cp; cp++)
404                     {
405                         /* deliberately only consider ASCII */
406                         if (*cp > 32 && *cp < 127)
407                             *cp = tolower(*cp);
408                     }
409                     s = auto_stylesheet;
410                 }
411                 else
412                 {
413                     session_log(se, YLOG_WARN,
414                                 "No pz:requestsyntax for auto stylesheet");
415                 }
416             }
417             sdb->map = normalize_cache_get(se->normalize_cache,
418                                            se->service, s);
419             if (!sdb->map)
420                 return -1;
421         }
422     }
423     return 0;
424 }
425
426 // called if watch should be removed because http_channel is to be destroyed
427 static void session_watch_cancel(void *data, struct http_channel *c,
428                                  void *data2)
429 {
430     struct session_watchentry *ent = data;
431
432     ent->fun = 0;
433     ent->data = 0;
434     ent->obs = 0;
435 }
436
437 // set watch. Returns 0=OK, -1 if watch is already set
438 int session_set_watch(struct session *s, int what,
439                       session_watchfun fun, void *data,
440                       struct http_channel *chan)
441 {
442     int ret;
443     session_enter(s, "session_set_watch");
444     if (s->watchlist[what].fun)
445         ret = -1;
446     else
447     {
448
449         s->watchlist[what].fun = fun;
450         s->watchlist[what].data = data;
451         s->watchlist[what].obs = http_add_observer(chan, &s->watchlist[what],
452                                                    session_watch_cancel);
453         ret = 0;
454     }
455     session_leave(s, "session_set_watch");
456     return ret;
457 }
458
459 void session_alert_watch(struct session *s, int what)
460 {
461     assert(s);
462     session_enter(s, "session_alert_watch");
463     if (s->watchlist[what].fun)
464     {
465         /* our watch is no longer associated with http_channel */
466         void *data;
467         session_watchfun fun;
468
469         http_remove_observer(s->watchlist[what].obs);
470         fun  = s->watchlist[what].fun;
471         data = s->watchlist[what].data;
472
473         /* reset watch before fun is invoked - in case fun wants to set
474            it again */
475         s->watchlist[what].fun = 0;
476         s->watchlist[what].data = 0;
477         s->watchlist[what].obs = 0;
478
479         session_leave(s, "session_alert_watch");
480         session_log(s, YLOG_DEBUG,
481                     "Alert Watch: %d calling function: %p", what, fun);
482         fun(data);
483     }
484     else
485         session_leave(s,"session_alert_watch");
486 }
487
488 //callback for grep_databases
489 static void select_targets_callback(struct session *se,
490                                     struct session_database *db)
491 {
492     struct client *cl;
493     struct client_list *l;
494
495     for (l = se->clients_cached; l; l = l->next)
496         if (client_get_database(l->client) == db)
497             break;
498
499     if (l)
500         cl = l->client;
501     else
502     {
503         cl = client_create(db->database->id);
504         client_set_database(cl, db);
505
506         l = xmalloc(sizeof(*l));
507         l->client = cl;
508         l->next = se->clients_cached;
509         se->clients_cached = l;
510     }
511     /* set session always. If may be 0 if client is not active */
512     client_set_session(cl, se);
513
514     l = xmalloc(sizeof(*l));
515     l->client = cl;
516     l->next = se->clients_active;
517     se->clients_active = l;
518 }
519
520 static void session_reset_active_clients(struct session *se,
521                                          struct client_list *new_list)
522 {
523     struct client_list *l;
524
525     session_enter(se, "session_reset_active_clients");
526     l = se->clients_active;
527     se->clients_active = new_list;
528     session_leave(se, "session_reset_active_clients");
529
530     while (l)
531     {
532         struct client_list *l_next = l->next;
533
534         client_lock(l->client);
535         client_set_session(l->client, 0); /* mark client inactive */
536         client_unlock(l->client);
537
538         xfree(l);
539         l = l_next;
540     }
541 }
542
543 static void session_remove_cached_clients(struct session *se)
544 {
545     struct client_list *l;
546
547     session_reset_active_clients(se, 0);
548
549     session_enter(se, "session_remove_cached_clients");
550     l = se->clients_cached;
551     se->clients_cached = 0;
552     session_leave(se, "session_remove_cached_clients");
553
554     while (l)
555     {
556         struct client_list *l_next = l->next;
557         client_lock(l->client);
558         client_set_session(l->client, 0);
559         client_set_database(l->client, 0);
560         client_unlock(l->client);
561         client_destroy(l->client);
562         xfree(l);
563         l = l_next;
564     }
565 }
566
567 // Associates a set of clients with a session;
568 // Note: Session-databases represent databases with per-session
569 // setting overrides
570 static int select_targets(struct session *se, const char *filter)
571 {
572     return session_grep_databases(se, filter, select_targets_callback);
573 }
574
575 int session_active_clients(struct session *s)
576 {
577     struct client_list *l;
578     int res = 0;
579
580     for (l = s->clients_active; l; l = l->next)
581         if (client_is_active(l->client))
582             res++;
583
584     return res;
585 }
586
587 int session_is_preferred_clients_ready(struct session *s)
588 {
589     struct client_list *l;
590     int res = 0;
591
592     for (l = s->clients_active; l; l = l->next)
593         if (client_is_active_preferred(l->client))
594             res++;
595     session_log(s, YLOG_DEBUG, "Has %d active preferred clients.", res);
596     return res == 0;
597 }
598
599 static void session_clear_set(struct session *se, struct reclist_sortparms *sp)
600 {
601     reclist_destroy(se->reclist);
602     if (nmem_total(se->nmem))
603         session_log(se, YLOG_DEBUG, "NMEN operation usage %zd",
604                     nmem_total(se->nmem));
605     nmem_reset(se->nmem);
606     se->total_records = se->total_merged = 0;
607     se->num_termlists = 0;
608
609     relevance_clear(se->relevance);
610
611     /* reset list of sorted results and clear to relevance search */
612     se->sorted_results = nmem_malloc(se->nmem, sizeof(*se->sorted_results));
613     se->sorted_results->name = nmem_strdup(se->nmem, sp->name);
614     se->sorted_results->increasing = sp->increasing;
615     se->sorted_results->type = sp->type;
616     se->sorted_results->next = 0;
617
618     session_log(se, YLOG_DEBUG, "clear_set session_sort: field=%s increasing=%d type=%d configured",
619             sp->name, sp->increasing, sp->type);
620
621     se->reclist = reclist_create(se->nmem);
622 }
623
624 static void session_sort_unlocked(struct session *se,
625                                   struct reclist_sortparms *sp,
626                                   const char *mergekey,
627                                   const char *rank)
628 {
629     struct client_list *l;
630     const char *field = sp->name;
631     int increasing = sp->increasing;
632     int type  = sp->type;
633     int clients_research = 0;
634
635     session_log(se, YLOG_DEBUG, "session_sort field=%s increasing=%d type=%d",
636                 field, increasing, type);
637
638     if (rank && (!se->rank || strcmp(se->rank, rank)))
639     {
640         /* new rank must research/reingest anyway */
641         assert(rank);
642         xfree(se->rank);
643         se->rank = *rank ? xstrdup(rank) : 0;
644         clients_research = 1;
645         session_log(se, YLOG_DEBUG, "session_sort: new rank = %s",
646                     rank);
647     }
648     if (mergekey && (!se->mergekey || strcmp(se->mergekey, mergekey)))
649     {
650         /* new mergekey must research/reingest anyway */
651         assert(mergekey);
652         xfree(se->mergekey);
653         se->mergekey = *mergekey ? xstrdup(mergekey) : 0;
654         clients_research = 1;
655         session_log(se, YLOG_DEBUG, "session_sort: new mergekey = %s",
656                     mergekey);
657     }
658     if (clients_research == 0)
659     {
660         struct reclist_sortparms *sr;
661         for (sr = se->sorted_results; sr; sr = sr->next)
662             if (!reclist_sortparms_cmp(sr, sp))
663                 break;
664         if (sr)
665         {
666             session_log(se, YLOG_DEBUG, "session_sort: field=%s increasing=%d type=%d already fetched",
667                         field, increasing, type);
668             return;
669         }
670     }
671     session_log(se, YLOG_DEBUG, "session_sort: field=%s increasing=%d type=%d must fetch",
672                 field, increasing, type);
673
674     // We need to reset reclist on every sort that changes the records, not just for position
675     // So if just one client requires new searching, we need to clear set.
676     // Ask each of the client if sorting requires re-search due to native sort
677     // If it does it will require us to
678     for (l = se->clients_active; l; l = l->next)
679     {
680         struct client *cl = l->client;
681         // Assume no re-search is required.
682         client_parse_init(cl, 1);
683         clients_research += client_parse_sort(cl, sp);
684     }
685     if (clients_research)
686     {
687         session_log(se, YLOG_DEBUG,
688                     "session_sort: reset results due to %d clients researching",
689                     clients_research);
690         session_clear_set(se, sp);
691     }
692     else
693     {
694         // A new sorting based on same record set
695         struct reclist_sortparms *sr = nmem_malloc(se->nmem, sizeof(*sr));
696         sr->name = nmem_strdup(se->nmem, field);
697         sr->increasing = increasing;
698         sr->type = type;
699         sr->next = se->sorted_results;
700         se->sorted_results = sr;
701         session_log(se, YLOG_DEBUG, "session_sort: no research/ingesting done");
702         return ;
703     }
704     session_log(se, YLOG_DEBUG, "Re- search/ingesting for clients due to change in sort order");
705
706     for (l = se->clients_active; l; l = l->next)
707     {
708         struct client *cl = l->client;
709         if (client_get_state(cl) == Client_Connecting ||
710             client_get_state(cl) == Client_Idle ||
711             client_get_state(cl) == Client_Working) {
712             client_start_search(cl);
713         }
714         else
715         {
716             session_log(se, YLOG_DEBUG,
717                         "session_sort: %s: No re-start/ingest in show. "
718                         "Wrong client state: %d",
719                         client_get_id(cl), client_get_state(cl));
720         }
721     }
722 }
723
724 void session_sort(struct session *se, struct reclist_sortparms *sp,
725                   const char *mergekey, const char *rank)
726 {
727     //session_enter(se, "session_sort");
728     session_sort_unlocked(se, sp, mergekey, rank);
729     //session_leave(se, "session_sort");
730 }
731
732
733 enum pazpar2_error_code session_search(struct session *se,
734                                        const char *query,
735                                        const char *startrecs,
736                                        const char *maxrecs,
737                                        const char *filter,
738                                        const char *limit,
739                                        const char **addinfo,
740                                        const char **addinfo2,
741                                        struct reclist_sortparms *sp,
742                                        const char *mergekey,
743                                        const char *rank)
744 {
745     int live_channels = 0;
746     int no_working = 0;
747     int no_failed_query = 0;
748     int no_failed_limit = 0;
749     struct client_list *l, *l0;
750
751     session_alert_watch(se, SESSION_WATCH_SHOW);
752     session_alert_watch(se, SESSION_WATCH_BYTARGET);
753     session_alert_watch(se, SESSION_WATCH_TERMLIST);
754     session_alert_watch(se, SESSION_WATCH_SHOW_PREF);
755
756     session_log(se, YLOG_DEBUG, "Search");
757
758     *addinfo = 0;
759
760     if (se->settings_modified) {
761         session_remove_cached_clients(se);
762     }
763     else
764         session_reset_active_clients(se, 0);
765
766     session_enter(se, "session_search");
767     se->settings_modified = 0;
768
769     if (mergekey)
770     {
771         xfree(se->mergekey);
772         se->mergekey = *mergekey ? xstrdup(mergekey) : 0;
773     }
774     if (rank)
775     {
776         xfree(se->rank);
777         se->rank = *rank ? xstrdup(rank) : 0;
778     }
779
780     session_clear_set(se, sp);
781     relevance_destroy(&se->relevance);
782
783     live_channels = select_targets(se, filter);
784     if (!live_channels)
785     {
786         session_leave(se, "session_search");
787         return PAZPAR2_NO_TARGETS;
788     }
789
790     facet_limits_destroy(se->facet_limits);
791     se->facet_limits = facet_limits_create(limit);
792     if (!se->facet_limits)
793     {
794         *addinfo = "limit";
795         session_leave(se, "session_search");
796         return PAZPAR2_MALFORMED_PARAMETER_VALUE;
797     }
798
799     l0 = se->clients_active;
800     se->clients_active = 0;
801     session_leave(se, "session_search");
802
803     for (l = l0; l; l = l->next)
804     {
805         int parse_ret;
806         struct client *cl = l->client;
807         client_parse_init(cl, 1);
808         if (prepare_map(se, client_get_database(cl)) < 0)
809             continue;
810
811         parse_ret = client_parse_query(cl, query, se->facet_limits, addinfo2);
812         if (parse_ret == -1)
813             no_failed_query++;
814         else if (parse_ret == -2)
815             no_failed_limit++;
816         else if (parse_ret < 0)
817             no_working++; /* other error, such as bad CCL map */
818         else
819         {
820             client_parse_range(cl, startrecs, maxrecs);
821             client_parse_sort(cl, sp);
822             client_start_search(cl);
823             no_working++;
824         }
825     }
826     session_reset_active_clients(se, l0);
827
828     if (no_working == 0)
829     {
830         if (no_failed_query > 0)
831         {
832             *addinfo = "query";
833             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
834         }
835         else if (no_failed_limit > 0)
836         {
837             *addinfo = "limit";
838             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
839         }
840         else
841             return PAZPAR2_NO_TARGETS;
842     }
843     return PAZPAR2_NO_ERROR;
844 }
845
846 // Creates a new session_database object for a database
847 static void session_init_databases_fun(void *context, struct database *db)
848 {
849     struct session *se = (struct session *) context;
850     struct session_database *new = nmem_malloc(se->session_nmem, sizeof(*new));
851     int i;
852
853     new->database = db;
854
855     new->map = 0;
856     assert(db->settings);
857     new->settings = nmem_malloc(se->session_nmem,
858                                 sizeof(struct settings *) * db->num_settings);
859     new->num_settings = db->num_settings;
860     for (i = 0; i < db->num_settings; i++)
861     {
862         struct setting *setting = db->settings[i];
863         new->settings[i] = setting;
864     }
865     new->next = se->databases;
866     se->databases = new;
867 }
868
869 // Doesn't free memory associated with sdb -- nmem takes care of that
870 static void session_database_destroy(struct session_database *sdb)
871 {
872     sdb->map = 0;
873 }
874
875 // Initialize session_database list -- this represents this session's view
876 // of the database list -- subject to modification by the settings ws command
877 void session_init_databases(struct session *se)
878 {
879     se->databases = 0;
880     predef_grep_databases(se, se->service, session_init_databases_fun);
881 }
882
883 // Probably session_init_databases_fun should be refactored instead of
884 // called here.
885 static struct session_database *load_session_database(struct session *se,
886                                                       const char *id)
887 {
888     struct database *db = new_database_inherit_settings(id, se->session_nmem, se->service->settings);
889     session_init_databases_fun((void*) se, db);
890
891     // New sdb is head of se->databases list
892     return se->databases;
893 }
894
895 // Find an existing session database. If not found, load it
896 static struct session_database *find_session_database(struct session *se,
897                                                       const char *id)
898 {
899     struct session_database *sdb;
900
901     for (sdb = se->databases; sdb; sdb = sdb->next)
902         if (!strcmp(sdb->database->id, id))
903             return sdb;
904     return load_session_database(se, id);
905 }
906
907 // Apply a session override to a database
908 void session_apply_setting(struct session *se, const char *dbname,
909                            const char *name, const char *value)
910 {
911     session_enter(se, "session_apply_setting");
912     {
913         struct session_database *sdb = find_session_database(se, dbname);
914         struct conf_service *service = se->service;
915         struct setting *s;
916         int offset = settings_create_offset(service, name);
917
918         expand_settings_array(&sdb->settings, &sdb->num_settings, offset,
919                               se->session_nmem);
920         // Force later recompute of settings-driven data structures
921         // (happens when a search starts and client connections are prepared)
922         if (offset == PZ_XSLT)
923             sdb->map = 0;
924         se->settings_modified = 1;
925         for (s = sdb->settings[offset]; s; s = s->next)
926             if (!strcmp(s->name, name) &&
927                 dbname && s->target && !strcmp(dbname, s->target))
928                 break;
929         if (!s)
930         {
931             s = nmem_malloc(se->session_nmem, sizeof(*s));
932             s->precedence = 0;
933             s->target = nmem_strdup(se->session_nmem, dbname);
934             s->name = nmem_strdup(se->session_nmem, name);
935             s->next = sdb->settings[offset];
936             sdb->settings[offset] = s;
937         }
938         s->value = nmem_strdup(se->session_nmem, value);
939     }
940     session_leave(se, "session_apply_setting");
941 }
942
943 void session_destroy(struct session *se)
944 {
945     struct session_database *sdb;
946     session_log(se, YLOG_LOG, "destroy");
947     session_use(-1);
948     session_remove_cached_clients(se);
949
950     for (sdb = se->databases; sdb; sdb = sdb->next)
951         session_database_destroy(sdb);
952     normalize_cache_destroy(se->normalize_cache);
953     relevance_destroy(&se->relevance);
954     reclist_destroy(se->reclist);
955     xfree(se->mergekey);
956     xfree(se->rank);
957     if (nmem_total(se->nmem))
958         session_log(se, YLOG_DEBUG, "NMEN operation usage %zd", nmem_total(se->nmem));
959     if (nmem_total(se->session_nmem))
960         session_log(se, YLOG_DEBUG, "NMEN session usage %zd", nmem_total(se->session_nmem));
961     facet_limits_destroy(se->facet_limits);
962     nmem_destroy(se->nmem);
963     service_destroy(se->service);
964     yaz_mutex_destroy(&se->session_mutex);
965 }
966
967 size_t session_get_memory_status(struct session *session) {
968     size_t session_nmem;
969     if (session == 0)
970         return 0;
971     session_enter(session, "session_get_memory_status");
972     session_nmem = nmem_total(session->nmem);
973     session_leave(session, "session_get_memory_status");
974     return session_nmem;
975 }
976
977
978 struct session *new_session(NMEM nmem, struct conf_service *service,
979                             unsigned session_id)
980 {
981     int i;
982     struct session *session = nmem_malloc(nmem, sizeof(*session));
983
984     char tmp_str[50];
985
986     sprintf(tmp_str, "session#%u", session_id);
987
988     session->session_id = session_id;
989     session_log(session, YLOG_DEBUG, "New");
990     session->service = service;
991     session->relevance = 0;
992     session->total_records = 0;
993     session->number_of_warnings_unknown_elements = 0;
994     session->number_of_warnings_unknown_metadata = 0;
995     session->num_termlists = 0;
996     session->reclist = reclist_create(nmem);
997     session->clients_active = 0;
998     session->clients_cached = 0;
999     session->settings_modified = 0;
1000     session->session_nmem = nmem;
1001     session->nmem = nmem_create();
1002     session->databases = 0;
1003     session->sorted_results = 0;
1004     session->facet_limits = 0;
1005     session->mergekey = 0;
1006     session->rank = 0;
1007
1008     for (i = 0; i <= SESSION_WATCH_MAX; i++)
1009     {
1010         session->watchlist[i].data = 0;
1011         session->watchlist[i].fun = 0;
1012     }
1013     session->normalize_cache = normalize_cache_create();
1014     session->session_mutex = 0;
1015     pazpar2_mutex_create(&session->session_mutex, tmp_str);
1016     session_log(session, YLOG_LOG, "create");
1017
1018     session_use(1);
1019     return session;
1020 }
1021
1022 const char * client_get_suggestions_xml(struct client *cl, WRBUF wrbuf);
1023
1024 static struct hitsbytarget *hitsbytarget_nb(struct session *se,
1025                                             int *count, NMEM nmem)
1026 {
1027     struct hitsbytarget *res = 0;
1028     struct client_list *l;
1029     size_t sz = 0;
1030
1031     for (l = se->clients_active; l; l = l->next)
1032         sz++;
1033
1034     res = nmem_malloc(nmem, sizeof(*res) * sz);
1035     *count = 0;
1036     for (l = se->clients_active; l; l = l->next)
1037     {
1038         struct client *cl = l->client;
1039         WRBUF w = wrbuf_alloc();
1040         const char *name = session_setting_oneval(client_get_database(cl),
1041                                                   PZ_NAME);
1042         res[*count].id = client_get_id(cl);
1043         res[*count].name = *name ? name : "Unknown";
1044         res[*count].hits = client_get_hits(cl);
1045         res[*count].approximation = client_get_approximation(cl);
1046         res[*count].records = client_get_num_records(cl,
1047                                                      &res[*count].filtered,
1048                                                      0, 0);
1049         res[*count].diagnostic =
1050             client_get_diagnostic(cl, &res[*count].message,
1051                                   &res[*count].addinfo);
1052         res[*count].state = client_get_state_str(cl);
1053         res[*count].connected  = client_get_connection(cl) ? 1 : 0;
1054         session_settings_dump(se, client_get_database(cl), w);
1055         res[*count].settings_xml = nmem_strdup(nmem, wrbuf_cstr(w));
1056         wrbuf_rewind(w);
1057         wrbuf_puts(w, "");
1058         res[*count].suggestions_xml = nmem_strdup(nmem, client_get_suggestions_xml(cl, w));
1059         wrbuf_destroy(w);
1060         (*count)++;
1061     }
1062     return res;
1063 }
1064
1065 struct hitsbytarget *get_hitsbytarget(struct session *se, int *count, NMEM nmem)
1066 {
1067     struct hitsbytarget *p;
1068     session_enter(se, "get_hitsbytarget");
1069     p = hitsbytarget_nb(se, count, nmem);
1070     session_leave(se, "get_hitsbytarget");
1071     return p;
1072 }
1073
1074 // Compares two hitsbytarget nodes by hitcount
1075 static int cmp_ht(const void *p1, const void *p2)
1076 {
1077     const struct hitsbytarget *h1 = p1;
1078     const struct hitsbytarget *h2 = p2;
1079     return h2->hits - h1->hits;
1080 }
1081
1082 // Compares two hitsbytarget nodes by hitcount
1083 static int cmp_ht_approx(const void *p1, const void *p2)
1084 {
1085     const struct hitsbytarget *h1 = p1;
1086     const struct hitsbytarget *h2 = p2;
1087     return h2->approximation - h1->approximation;
1088 }
1089
1090 static int targets_termlist_nb(WRBUF wrbuf, struct session *se, int num,
1091                                NMEM nmem, int version)
1092 {
1093     struct hitsbytarget *ht;
1094     int count, i;
1095
1096     ht = hitsbytarget_nb(se, &count, nmem);
1097     if (version >= 2)
1098         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht_approx);
1099     else
1100         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht);
1101     for (i = 0; i < count && i < num && ht[i].hits > 0; i++)
1102     {
1103
1104         // do only print terms which have display names
1105
1106         wrbuf_puts(wrbuf, "<term>\n");
1107
1108         wrbuf_puts(wrbuf, "<id>");
1109         wrbuf_xmlputs(wrbuf, ht[i].id);
1110         wrbuf_puts(wrbuf, "</id>\n");
1111
1112         wrbuf_puts(wrbuf, "<name>");
1113         if (!ht[i].name || !ht[i].name[0])
1114             wrbuf_xmlputs(wrbuf, "NO TARGET NAME");
1115         else
1116             wrbuf_xmlputs(wrbuf, ht[i].name);
1117         wrbuf_puts(wrbuf, "</name>\n");
1118
1119         wrbuf_printf(wrbuf, "<frequency>" ODR_INT_PRINTF "</frequency>\n",
1120                      ht[i].hits);
1121
1122         if (version >= 2) {
1123             // Should not print if we know it isn't a approximation.
1124             wrbuf_printf(wrbuf, "<approximation>" ODR_INT_PRINTF "</approximation>\n", ht[i].approximation);
1125             wrbuf_printf(wrbuf, "<records>%d</records>\n", ht[i].records - ht[i].filtered);
1126             wrbuf_printf(wrbuf, "<filtered>%d</filtered>\n", ht[i].filtered);
1127         }
1128
1129         wrbuf_puts(wrbuf, "<state>");
1130         wrbuf_xmlputs(wrbuf, ht[i].state);
1131         wrbuf_puts(wrbuf, "</state>\n");
1132
1133         wrbuf_printf(wrbuf, "<diagnostic>%d</diagnostic>\n",
1134                      ht[i].diagnostic);
1135         wrbuf_puts(wrbuf, "</term>\n");
1136     }
1137     return count;
1138 }
1139
1140 void perform_termlist(struct http_channel *c, struct session *se,
1141                       const char *name, int num, int version)
1142 {
1143     int i, j;
1144     NMEM nmem_tmp = nmem_create();
1145     char **names;
1146     int num_names = 0;
1147
1148     if (!name)
1149         name = "*";
1150
1151     nmem_strsplit(nmem_tmp, ",", name, &names, &num_names);
1152
1153     session_enter(se, "perform_termlist");
1154
1155     for (j = 0; j < num_names; j++)
1156     {
1157         const char *tname;
1158         int must_generate_empty = 1; /* bug 5350 */
1159
1160         for (i = 0; i < se->num_termlists; i++)
1161         {
1162             tname = se->termlists[i].name;
1163             if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1164             {
1165                 struct termlist_score **p = 0;
1166                 int len;
1167
1168                 wrbuf_puts(c->wrbuf, "<list name=\"");
1169                 wrbuf_xmlputs(c->wrbuf, tname);
1170                 wrbuf_puts(c->wrbuf, "\">\n");
1171                 must_generate_empty = 0;
1172
1173                 p = termlist_highscore(se->termlists[i].termlist, &len,
1174                                        nmem_tmp);
1175                 if (p)
1176                 {
1177                     int i;
1178                     for (i = 0; i < len && i < num; i++)
1179                     {
1180                         // prevent sending empty term elements
1181                         if (!p[i]->display_term || !p[i]->display_term[0])
1182                             continue;
1183
1184                         wrbuf_puts(c->wrbuf, "<term>");
1185                         wrbuf_puts(c->wrbuf, "<name>");
1186                         wrbuf_xmlputs(c->wrbuf, p[i]->display_term);
1187                         wrbuf_puts(c->wrbuf, "</name>");
1188
1189                         wrbuf_printf(c->wrbuf,
1190                                      "<frequency>%d</frequency>",
1191                                      p[i]->frequency);
1192                         wrbuf_puts(c->wrbuf, "</term>\n");
1193                     }
1194                 }
1195                 wrbuf_puts(c->wrbuf, "</list>\n");
1196             }
1197         }
1198         tname = "xtargets";
1199         if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1200         {
1201             wrbuf_puts(c->wrbuf, "<list name=\"");
1202             wrbuf_xmlputs(c->wrbuf, tname);
1203             wrbuf_puts(c->wrbuf, "\">\n");
1204
1205             targets_termlist_nb(c->wrbuf, se, num, c->nmem, version);
1206             wrbuf_puts(c->wrbuf, "</list>\n");
1207             must_generate_empty = 0;
1208         }
1209         if (must_generate_empty)
1210         {
1211             wrbuf_puts(c->wrbuf, "<list name=\"");
1212             wrbuf_xmlputs(c->wrbuf, names[j]);
1213             wrbuf_puts(c->wrbuf, "\"/>\n");
1214         }
1215     }
1216     session_leave(se, "perform_termlist");
1217     nmem_destroy(nmem_tmp);
1218 }
1219
1220 #ifdef MISSING_HEADERS
1221 void report_nmem_stats(void)
1222 {
1223     size_t in_use, is_free;
1224
1225     nmem_get_memory_in_use(&in_use);
1226     nmem_get_memory_free(&is_free);
1227
1228     yaz_log(YLOG_LOG, "nmem stat: use=%ld free=%ld",
1229             (long) in_use, (long) is_free);
1230 }
1231 #endif
1232
1233 struct record_cluster *show_single_start(struct session *se, const char *id,
1234                                          struct record_cluster **prev_r,
1235                                          struct record_cluster **next_r)
1236 {
1237     struct record_cluster *r = 0;
1238
1239     session_enter(se, "show_single_start");
1240     *prev_r = 0;
1241     *next_r = 0;
1242     reclist_limit(se->reclist, se, 1);
1243
1244     reclist_enter(se->reclist);
1245     while ((r = reclist_read_record(se->reclist)))
1246     {
1247         if (!strcmp(r->recid, id))
1248         {
1249             *next_r = reclist_read_record(se->reclist);
1250             break;
1251         }
1252         *prev_r = r;
1253     }
1254     reclist_leave(se->reclist);
1255     if (!r)
1256         session_leave(se, "show_single_start");
1257     return r;
1258 }
1259
1260 void show_single_stop(struct session *se, struct record_cluster *rec)
1261 {
1262     session_leave(se, "show_single_stop");
1263 }
1264
1265
1266 int session_fetch_more(struct session *se)
1267 {
1268     struct client_list *l;
1269     int ret = 0;
1270
1271     for (l = se->clients_active; l; l = l->next)
1272     {
1273         struct client *cl = l->client;
1274         if (client_get_state(cl) == Client_Idle)
1275         {
1276             if (client_fetch_more(cl))
1277             {
1278                 session_log(se, YLOG_LOG, "%s: more to fetch",
1279                             client_get_id(cl));
1280                 ret = 1;
1281             }
1282             else
1283             {
1284                 int filtered;
1285                 int ingest_failures;
1286                 int record_failures;
1287                 int num = client_get_num_records(
1288                     cl, &filtered, &ingest_failures, &record_failures);
1289
1290                 session_log(se, YLOG_LOG, "%s: hits=" ODR_INT_PRINTF
1291                             " fetched=%d filtered=%d",
1292                             client_get_id(cl),
1293                             client_get_hits(cl),
1294                             num, filtered);
1295                 if (ingest_failures || record_failures)
1296                 {
1297                     session_log(se, YLOG_WARN, "%s:"
1298                                 " ingest failures=%d record failures=%d",
1299                                 client_get_id(cl),
1300                                 ingest_failures, record_failures);
1301                 }
1302             }
1303         }
1304         else
1305         {
1306             session_log(se, YLOG_LOG, "%s: no fetch due to state=%s",
1307                         client_get_id(cl), client_get_state_str(cl));
1308         }
1309
1310     }
1311     return ret;
1312 }
1313
1314 struct record_cluster **show_range_start(struct session *se,
1315                                          struct reclist_sortparms *sp,
1316                                          int start, int *num, int *total,
1317                                          Odr_int *sumhits, Odr_int *approx_hits,
1318                                          void (*show_records_ready)(void *data),
1319                                          struct http_channel *chan)
1320 {
1321     struct record_cluster **recs = 0;
1322     struct reclist_sortparms *spp;
1323     struct client_list *l;
1324     int i;
1325 #if USE_TIMING
1326     yaz_timing_t t = yaz_timing_create();
1327 #endif
1328     session_enter(se, "show_range_start");
1329     *sumhits = 0;
1330     *approx_hits = 0;
1331     *total = 0;
1332     reclist_limit(se->reclist, se, 0);
1333     if (se->relevance)
1334     {
1335         for (spp = sp; spp; spp = spp->next)
1336             if (spp->type == Metadata_type_relevance)
1337             {
1338                 relevance_prepare_read(se->relevance, se->reclist);
1339                 break;
1340             }
1341         for (l = se->clients_active; l; l = l->next) {
1342             *sumhits += client_get_hits(l->client);
1343             *approx_hits += client_get_approximation(l->client);
1344         }
1345     }
1346     reclist_sort(se->reclist, sp);
1347
1348     reclist_enter(se->reclist);
1349     *total = reclist_get_num_records(se->reclist);
1350
1351     for (l = se->clients_active; l; l = l->next)
1352         client_update_show_stat(l->client, 0);
1353
1354     for (i = 0; i < start; i++)
1355     {
1356         struct record_cluster *r = reclist_read_record(se->reclist);
1357         if (!r)
1358         {
1359             *num = 0;
1360             break;
1361         }
1362         else
1363         {
1364             struct record *rec = r->records;
1365             for (;rec; rec = rec->next)
1366                 client_update_show_stat(rec->client, 1);
1367         }
1368     }
1369     recs = nmem_malloc(se->nmem, (*num > 0 ? *num : 1) * sizeof(*recs));
1370     for (i = 0; i < *num; i++)
1371     {
1372         struct record_cluster *r = reclist_read_record(se->reclist);
1373         if (!r)
1374         {
1375             *num = i;
1376             break;
1377         }
1378         else
1379         {
1380             struct record *rec = r->records;
1381             for (;rec; rec = rec->next)
1382                 client_update_show_stat(rec->client, 1);
1383             recs[i] = r;
1384         }
1385     }
1386     reclist_leave(se->reclist);
1387 #if USE_TIMING
1388     yaz_timing_stop(t);
1389     session_log(se, YLOG_LOG, "show %6.5f %3.2f %3.2f",
1390             yaz_timing_get_real(t), yaz_timing_get_user(t),
1391             yaz_timing_get_sys(t));
1392     yaz_timing_destroy(&t);
1393 #endif
1394
1395     if (!session_fetch_more(se))
1396         session_log(se, YLOG_LOG, "can not fetch more");
1397     else
1398     {
1399         show_range_stop(se, recs);
1400         session_log(se, YLOG_LOG, "fetching more in progress");
1401         if (session_set_watch(se, SESSION_WATCH_SHOW,
1402                               show_records_ready, chan, chan))
1403         {
1404             session_log(se, YLOG_WARN, "Ignoring show block");
1405             session_enter(se, "show_range_start");
1406         }
1407         else
1408         {
1409             session_log(se, YLOG_LOG, "session watch OK");
1410             return 0;
1411         }
1412     }
1413     return recs;
1414 }
1415
1416 void show_range_stop(struct session *se, struct record_cluster **recs)
1417 {
1418     session_leave(se, "show_range_stop");
1419 }
1420
1421 void statistics(struct session *se, struct statistics *stat)
1422 {
1423     struct client_list *l;
1424     int count = 0;
1425
1426     memset(stat, 0, sizeof(*stat));
1427     stat->num_hits = 0;
1428     for (l = se->clients_active; l; l = l->next)
1429     {
1430         struct client *cl = l->client;
1431         if (!client_get_connection(cl))
1432             stat->num_no_connection++;
1433         stat->num_hits += client_get_hits(cl);
1434         switch (client_get_state(cl))
1435         {
1436         case Client_Connecting: stat->num_connecting++; break;
1437         case Client_Working: stat->num_working++; break;
1438         case Client_Idle: stat->num_idle++; break;
1439         case Client_Failed: stat->num_failed++; break;
1440         case Client_Error: stat->num_error++; break;
1441         default: break;
1442         }
1443         count++;
1444     }
1445     stat->num_records = se->total_records;
1446
1447     stat->num_clients = count;
1448 }
1449
1450 static struct record_metadata *record_metadata_init(
1451     NMEM nmem, const char *value, enum conf_metadata_type type,
1452     struct _xmlAttr *attr)
1453 {
1454     struct record_metadata *rec_md = record_metadata_create(nmem);
1455     struct record_metadata_attr **attrp = &rec_md->attributes;
1456
1457     for (; attr; attr = attr->next)
1458     {
1459         if (attr->children && attr->children->content)
1460         {
1461             if (strcmp((const char *) attr->name, "type")
1462                 && strcmp((const char *) attr->name, "empty"))
1463             {  /* skip the "type" + "empty" attribute..
1464                   The "Type" is already part of the element in output
1465                   (md-%s) and so repeating it here is redundant */
1466                 *attrp = nmem_malloc(nmem, sizeof(**attrp));
1467                 (*attrp)->name =
1468                     nmem_strdup(nmem, (const char *) attr->name);
1469                 (*attrp)->value =
1470                     nmem_strdup(nmem, (const char *) attr->children->content);
1471                 attrp = &(*attrp)->next;
1472             }
1473         }
1474     }
1475     *attrp = 0;
1476
1477     switch (type)
1478     {
1479     case Metadata_type_generic:
1480     case Metadata_type_skiparticle:
1481         if (strstr(value, "://")) /* looks like a URL */
1482             rec_md->data.text.disp = nmem_strdup(nmem, value);
1483         else
1484             rec_md->data.text.disp =
1485                 normalize7bit_generic(nmem_strdup(nmem, value), " ,/.:([");
1486         rec_md->data.text.sort = 0;
1487         rec_md->data.text.snippet = 0;
1488         break;
1489     case Metadata_type_year:
1490     case Metadata_type_date:
1491     {
1492         int first, last;
1493         int longdate = 0;
1494
1495         if (type == Metadata_type_date)
1496             longdate = 1;
1497         if (extract7bit_dates((char *) value, &first, &last, longdate) < 0)
1498             return 0;
1499
1500         rec_md->data.number.min = first;
1501         rec_md->data.number.max = last;
1502     }
1503     break;
1504     case Metadata_type_float:
1505         rec_md->data.fnumber = atof(value);
1506         break;
1507     case Metadata_type_relevance:
1508     case Metadata_type_position:
1509         return 0;
1510     }
1511     return rec_md;
1512 }
1513
1514 static void mergekey_norm_wr(pp2_charset_fact_t charsets,
1515                              WRBUF norm_wr, const char *value)
1516 {
1517     const char *norm_str;
1518     pp2_charset_token_t prt =
1519         pp2_charset_token_create(charsets, "mergekey");
1520
1521     pp2_charset_token_first(prt, value, 0);
1522     while ((norm_str = pp2_charset_token_next(prt)))
1523     {
1524         if (*norm_str)
1525         {
1526             if (wrbuf_len(norm_wr))
1527                 wrbuf_puts(norm_wr, " ");
1528             wrbuf_puts(norm_wr, norm_str);
1529         }
1530     }
1531     pp2_charset_token_destroy(prt);
1532 }
1533
1534 static int get_mergekey_from_doc(xmlDoc *doc, xmlNode *root, const char *name,
1535                                  struct conf_service *service, WRBUF norm_wr)
1536 {
1537     xmlNode *n;
1538     int no_found = 0;
1539     for (n = root->children; n; n = n->next)
1540     {
1541         if (n->type != XML_ELEMENT_NODE)
1542             continue;
1543         if (!strcmp((const char *) n->name, "metadata"))
1544         {
1545             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1546             if (type == NULL) {
1547                 yaz_log(YLOG_FATAL, "Missing type attribute on metadata element. Skipping!");
1548             }
1549             else if (!strcmp(name, (const char *) type))
1550             {
1551                 xmlChar *value = xmlNodeListGetString(doc, n->children, 1);
1552                 if (value && *value)
1553                 {
1554                     if (wrbuf_len(norm_wr) > 0)
1555                         wrbuf_puts(norm_wr, " ");
1556                     wrbuf_puts(norm_wr, name);
1557                     mergekey_norm_wr(service->charsets, norm_wr,
1558                                      (const char *) value);
1559                     no_found++;
1560                 }
1561                 if (value)
1562                     xmlFree(value);
1563             }
1564             xmlFree(type);
1565         }
1566     }
1567     return no_found;
1568 }
1569
1570 static const char *get_mergekey(xmlDoc *doc, xmlNode *root, 
1571                                 struct client *cl, int record_no,
1572                                 struct conf_service *service, NMEM nmem,
1573                                 const char *session_mergekey)
1574 {
1575     char *mergekey_norm = 0;
1576     WRBUF norm_wr = wrbuf_alloc();
1577     xmlChar *mergekey;
1578
1579     if (session_mergekey)
1580     {
1581         int i, num = 0;
1582         char **values = 0;
1583         nmem_strsplit_escape2(nmem, ",", session_mergekey, &values,
1584                               &num, 1, '\\', 1);
1585
1586         for (i = 0; i < num; i++)
1587             get_mergekey_from_doc(doc, root, values[i], service, norm_wr);
1588     }
1589     else if ((mergekey = xmlGetProp(root, (xmlChar *) "mergekey")))
1590     {
1591         mergekey_norm_wr(service->charsets, norm_wr, (const char *) mergekey);
1592         xmlFree(mergekey);
1593     }
1594     else
1595     {
1596         /* no mergekey defined in XSL. Look for mergekey metadata instead */
1597         int field_id;
1598         for (field_id = 0; field_id < service->num_metadata; field_id++)
1599         {
1600             struct conf_metadata *ser_md = &service->metadata[field_id];
1601             if (ser_md->mergekey != Metadata_mergekey_no)
1602             {
1603                 int r = get_mergekey_from_doc(doc, root, ser_md->name,
1604                                               service, norm_wr);
1605                 if (r == 0 && ser_md->mergekey == Metadata_mergekey_required)
1606                 {
1607                     /* no mergekey on this one and it is required..
1608                        Generate unique key instead */
1609                     wrbuf_rewind(norm_wr);
1610                     break;
1611                 }
1612             }
1613         }
1614     }
1615
1616     /* generate unique key if none is not generated already or is empty */
1617     if (wrbuf_len(norm_wr) == 0)
1618     {
1619         wrbuf_printf(norm_wr, "position: %s-%d",
1620                      client_get_id(cl), record_no);
1621     }
1622     else
1623     {
1624         const char *lead = "content: ";
1625         wrbuf_insert(norm_wr, 0, lead, strlen(lead));
1626     }
1627     if (wrbuf_len(norm_wr) > 0)
1628         mergekey_norm = nmem_strdup(nmem, wrbuf_cstr(norm_wr));
1629     wrbuf_destroy(norm_wr);
1630     return mergekey_norm;
1631 }
1632
1633 /** \brief see if metadata for pz:recordfilter exists
1634     \param root xml root element of normalized record
1635     \param sdb session database for client
1636     \retval 0 if there is no metadata for pz:recordfilter
1637     \retval 1 if there is metadata for pz:recordfilter
1638
1639     If there is no pz:recordfilter defined, this function returns 1
1640     as well.
1641 */
1642
1643 static int check_record_filter(xmlNode *root, struct session_database *sdb)
1644 {
1645     int match = 0;
1646     xmlNode *n;
1647     const char *s;
1648     s = session_setting_oneval(sdb, PZ_RECORDFILTER);
1649
1650     if (!s || !*s)
1651         return 1;
1652
1653     for (n = root->children; n; n = n->next)
1654     {
1655         if (n->type != XML_ELEMENT_NODE)
1656             continue;
1657         if (!strcmp((const char *) n->name, "metadata"))
1658         {
1659             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1660             if (type)
1661             {
1662                 size_t len;
1663                 int substring;
1664                 const char *eq;
1665
1666                 if ((eq = strchr(s, '=')))
1667                     substring = 0;
1668                 else if ((eq = strchr(s, '~')))
1669                     substring = 1;
1670                 if (eq)
1671                     len = eq - s;
1672                 else
1673                     len = strlen(s);
1674                 if (len == strlen((const char *)type) &&
1675                     !memcmp((const char *) type, s, len))
1676                 {
1677                     xmlChar *value = xmlNodeGetContent(n);
1678                     if (value && *value)
1679                     {
1680                         if (!eq ||
1681                             (substring && strstr((const char *) value, eq+1)) ||
1682                             (!substring && !strcmp((const char *) value, eq + 1)))
1683                             match = 1;
1684                     }
1685                     xmlFree(value);
1686                 }
1687                 xmlFree(type);
1688             }
1689         }
1690     }
1691     return match;
1692 }
1693
1694 static int ingest_to_cluster(struct client *cl,
1695                              xmlDoc *xdoc,
1696                              xmlNode *root,
1697                              int record_no,
1698                              struct record_metadata_attr *mergekey);
1699
1700 static int ingest_sub_record(struct client *cl, xmlDoc *xdoc, xmlNode *root,
1701                              int record_no, NMEM nmem,
1702                              struct session_database *sdb,
1703                              struct record_metadata_attr *mergekeys)
1704 {
1705     int ret = 0;
1706     struct session *se = client_get_session(cl);
1707
1708     if (!check_record_filter(root, sdb))
1709     {
1710         session_log(se, YLOG_LOG,
1711                     "Filtered out record no %d from %s",
1712                     record_no, sdb->database->id);
1713         return 0;
1714     }
1715     session_enter(se, "ingest_sub_record");
1716     if (client_get_session(cl) == se && se->relevance)
1717         ret = ingest_to_cluster(cl, xdoc, root, record_no, mergekeys);
1718     session_leave(se, "ingest_sub_record");
1719
1720     return ret;
1721 }
1722
1723 /** \brief ingest XML record
1724     \param cl client holds the result set for record
1725     \param rec record buffer (0 terminated)
1726     \param record_no record position (1, 2, ..)
1727     \param nmem working NMEM
1728     \retval 0 OK
1729     \retval -1 failure
1730     \retval -2 Filtered
1731 */
1732 int ingest_record(struct client *cl, const char *rec,
1733                   int record_no, NMEM nmem)
1734 {
1735     struct session *se = client_get_session(cl);
1736     struct session_database *sdb = client_get_database(cl);
1737     struct conf_service *service = se->service;
1738     xmlDoc *xdoc = normalize_record(se, sdb, service, rec, nmem);
1739     int r = ingest_xml_record(cl, xdoc, record_no, nmem, 0);
1740     client_store_xdoc(cl, record_no, xdoc);
1741     return r;
1742 }
1743
1744 int ingest_xml_record(struct client *cl, xmlDoc *xdoc,
1745                       int record_no, NMEM nmem, int cached_copy)
1746 {
1747     struct session *se = client_get_session(cl);
1748     struct session_database *sdb = client_get_database(cl);
1749     struct conf_service *service = se->service;
1750     xmlNode *root;
1751     int r = 0;
1752     if (!xdoc)
1753         return -1;
1754
1755     if (global_parameters.dump_records)
1756     {
1757         session_log(se, YLOG_LOG, "Normalized record from %s",
1758                     sdb->database->id);
1759         log_xml_doc(xdoc);
1760     }
1761
1762     root = xmlDocGetRootElement(xdoc);
1763
1764     if (!strcmp((const char *) root->name, "cluster"))
1765     {
1766         int no_merge_keys = 0;
1767         int no_merge_dups = 0;
1768         xmlNode *sroot;
1769         struct record_metadata_attr *mk = 0;
1770
1771         for (sroot = root->children; sroot; sroot = sroot->next)
1772             if (sroot->type == XML_ELEMENT_NODE &&
1773                 !strcmp((const char *) sroot->name, "record"))
1774             {
1775                 struct record_metadata_attr **mkp;
1776                 const char *mergekey_norm =
1777                     get_mergekey(xdoc, sroot, cl, record_no, service, nmem,
1778                                  se->mergekey);
1779                 if (!mergekey_norm)
1780                 {
1781                     r = -1;
1782                     break;
1783                 }
1784                 for (mkp = &mk; *mkp; mkp = &(*mkp)->next)
1785                     if (!strcmp((*mkp)->value, mergekey_norm))
1786                         break;
1787                 if (!*mkp)
1788                 {
1789                     *mkp = (struct record_metadata_attr*)
1790                         nmem_malloc(nmem, sizeof(**mkp));
1791                     (*mkp)->name = 0;
1792                     (*mkp)->value = nmem_strdup(nmem, mergekey_norm);
1793                     (*mkp)->next = 0;
1794                     no_merge_keys++;
1795                 }
1796                 else
1797                     no_merge_dups++;
1798             }
1799         if (no_merge_keys > 1 || no_merge_dups > 0)
1800         {
1801             yaz_log(YLOG_LOG, "Got %d mergekeys, %d dups for position %d",
1802                     no_merge_keys, no_merge_dups, record_no);
1803         }
1804         for (sroot = root->children; !r && sroot; sroot = sroot->next)
1805             if (sroot->type == XML_ELEMENT_NODE &&
1806                 !strcmp((const char *) sroot->name, "record"))
1807             {
1808                 if (!cached_copy)
1809                     insert_settings_values(sdb, xdoc, root, service);
1810                 r = ingest_sub_record(cl, xdoc, sroot, record_no, nmem, sdb,
1811                                       mk);
1812             }
1813     }
1814     else if (!strcmp((const char *) root->name, "record"))
1815     {
1816         const char *mergekey_norm =
1817             get_mergekey(xdoc, root, cl, record_no, service, nmem,
1818                          se->mergekey);
1819         if (mergekey_norm)
1820         {
1821             struct record_metadata_attr *mk = (struct record_metadata_attr*)
1822                 nmem_malloc(nmem, sizeof(*mk));
1823             mk->name = 0;
1824             mk->value = nmem_strdup(nmem, mergekey_norm);
1825             mk->next = 0;
1826
1827             if (!cached_copy)
1828                 insert_settings_values(sdb, xdoc, root, service);
1829             r = ingest_sub_record(cl, xdoc, root, record_no, nmem, sdb, mk);
1830         }
1831     }
1832     else
1833     {
1834         session_log(se, YLOG_WARN, "Bad pz root element: %s",
1835                     (const char *) root->name);
1836         r = -1;
1837     }
1838     return r;
1839 }
1840
1841
1842 //    struct conf_metadata *ser_md = &service->metadata[md_field_id];
1843 //    struct record_metadata *rec_md = record->metadata[md_field_id];
1844 static int match_metadata_local(struct conf_service *service,
1845                                 struct conf_metadata *ser_md,
1846                                 struct record_metadata *rec_md0,
1847                                 char **values, int num_v)
1848 {
1849     int i;
1850     struct record_metadata *rec_md = rec_md0;
1851     WRBUF val_wr = 0;
1852     WRBUF text_wr = wrbuf_alloc();
1853     for (i = 0; i < num_v; )
1854     {
1855         if (rec_md)
1856         {
1857             if (ser_md->type == Metadata_type_year
1858                 || ser_md->type == Metadata_type_date)
1859             {
1860                 int y = atoi(values[i]);
1861                 if (y >= rec_md->data.number.min
1862                     && y <= rec_md->data.number.max)
1863                     break;
1864             }
1865             else
1866             {
1867                 if (!val_wr)
1868                 {
1869                     val_wr = wrbuf_alloc();
1870                     mergekey_norm_wr(service->charsets, val_wr, values[i]);
1871                 }
1872                 wrbuf_rewind(text_wr);
1873                 mergekey_norm_wr(service->charsets, text_wr,
1874                                  rec_md->data.text.disp);
1875                 if (!strcmp(wrbuf_cstr(val_wr), wrbuf_cstr(text_wr)))
1876                     break;
1877             }
1878             rec_md = rec_md->next;
1879         }
1880         else
1881         {
1882             rec_md = rec_md0;
1883             wrbuf_destroy(val_wr);
1884             val_wr = 0;
1885             i++;
1886         }
1887     }
1888     wrbuf_destroy(val_wr);
1889     wrbuf_destroy(text_wr);
1890     return i < num_v ? 1 : 0;
1891 }
1892
1893 int session_check_cluster_limit(struct session *se, struct record_cluster *rec)
1894 {
1895     int i;
1896     struct conf_service *service = se->service;
1897     int ret = 1;
1898     const char *name;
1899     const char *value;
1900     NMEM nmem_tmp = nmem_create();
1901
1902     for (i = 0; (name = facet_limits_get(se->facet_limits, i, &value)); i++)
1903     {
1904         int j;
1905         for (j = 0; j < service->num_metadata; j++)
1906         {
1907             struct conf_metadata *md = service->metadata + j;
1908             if (!strcmp(md->name, name) && md->limitcluster)
1909             {
1910                 char **values = 0;
1911                 int num = 0;
1912                 int md_field_id =
1913                     conf_service_metadata_field_id(service,
1914                                                    md->limitcluster);
1915
1916                 if (md_field_id < 0)
1917                 {
1918                     ret = 0;
1919                     break;
1920                 }
1921
1922                 nmem_strsplit_escape2(nmem_tmp, "|", value, &values,
1923                                       &num, 1, '\\', 1);
1924
1925                 if (!match_metadata_local(service,
1926                                           &service->metadata[md_field_id],
1927                                           rec->metadata[md_field_id],
1928                                           values, num))
1929                 {
1930                     ret = 0;
1931                     break;
1932                 }
1933             }
1934         }
1935     }
1936     nmem_destroy(nmem_tmp);
1937     return ret;
1938 }
1939
1940 // Skip record on non-zero
1941 static int check_limit_local(struct client *cl,
1942                              struct record *record,
1943                              int record_no)
1944 {
1945     int skip_record = 0;
1946     struct session *se = client_get_session(cl);
1947     struct conf_service *service = se->service;
1948     NMEM nmem_tmp = nmem_create();
1949     struct session_database *sdb = client_get_database(cl);
1950     int l = 0;
1951     while (!skip_record)
1952     {
1953         int md_field_id;
1954         char **values = 0;
1955         int num_v = 0;
1956         const char *name =
1957             client_get_facet_limit_local(cl, sdb, &l, nmem_tmp,
1958                                          &num_v, &values);
1959         if (!name)
1960             break;
1961
1962         if (!strcmp(name, "*"))
1963         {
1964             for (md_field_id = 0; md_field_id < service->num_metadata;
1965                  md_field_id++)
1966             {
1967                 if (match_metadata_local(
1968                         service,
1969                         &service->metadata[md_field_id],
1970                         record->metadata[md_field_id],
1971                         values, num_v))
1972                     break;
1973             }
1974             if (md_field_id == service->num_metadata)
1975                 skip_record = 1;
1976         }
1977         else
1978         {
1979             md_field_id = conf_service_metadata_field_id(service, name);
1980             if (md_field_id < 0)
1981             {
1982                 skip_record = 1;
1983                 break;
1984             }
1985             if (!match_metadata_local(
1986                     service,
1987                     &service->metadata[md_field_id],
1988                     record->metadata[md_field_id],
1989                     values, num_v))
1990             {
1991                 skip_record = 1;
1992             }
1993         }
1994     }
1995     nmem_destroy(nmem_tmp);
1996     return skip_record;
1997 }
1998
1999 static int ingest_to_cluster(struct client *cl,
2000                              xmlDoc *xdoc,
2001                              xmlNode *root,
2002                              int record_no,
2003                              struct record_metadata_attr *merge_keys)
2004 {
2005     xmlNode *n;
2006     xmlChar *type = 0;
2007     xmlChar *value = 0;
2008     struct session *se = client_get_session(cl);
2009     struct conf_service *service = se->service;
2010     int term_factor = 1;
2011     struct record_cluster *cluster;
2012     struct record_metadata **metadata0;
2013     struct session_database *sdb = client_get_database(cl);
2014     NMEM ingest_nmem = 0;
2015     char **rank_values = 0;
2016     int rank_num = 0;
2017     struct record *record = record_create(se->nmem,
2018                                           service->num_metadata,
2019                                           service->num_sortkeys, cl,
2020                                           record_no);
2021
2022     for (n = root->children; n; n = n->next)
2023     {
2024         if (type)
2025             xmlFree(type);
2026         if (value)
2027             xmlFree(value);
2028         type = value = 0;
2029
2030         if (n->type != XML_ELEMENT_NODE)
2031             continue;
2032         if (!strcmp((const char *) n->name, "metadata"))
2033         {
2034             struct conf_metadata *ser_md = 0;
2035             struct record_metadata **wheretoput = 0;
2036             struct record_metadata *rec_md = 0;
2037             int md_field_id = -1;
2038
2039             type = xmlGetProp(n, (xmlChar *) "type");
2040             value = xmlNodeListGetString(xdoc, n->children, 1);
2041             if (!type)
2042                 continue;
2043             if (!value || !*value)
2044             {
2045                 xmlChar *empty = xmlGetProp(n, (xmlChar *) "empty");
2046                 if (!empty)
2047                     continue;
2048                 if (value)
2049                     xmlFree(value);
2050                 value = empty;
2051             }
2052             md_field_id
2053                 = conf_service_metadata_field_id(service, (const char *) type);
2054             if (md_field_id < 0)
2055             {
2056                 if (se->number_of_warnings_unknown_metadata == 0)
2057                 {
2058                     session_log(se, YLOG_WARN,
2059                             "Ignoring unknown metadata element: %s", type);
2060                 }
2061                 se->number_of_warnings_unknown_metadata++;
2062                 continue;
2063             }
2064
2065             ser_md = &service->metadata[md_field_id];
2066
2067             // non-merged metadata
2068             rec_md = record_metadata_init(se->nmem, (const char *) value,
2069                                           ser_md->type, n->properties);
2070             if (!rec_md)
2071             {
2072                 session_log(se, YLOG_WARN, "bad metadata data '%s' "
2073                             "for element '%s'", value, type);
2074                 continue;
2075             }
2076
2077             if (ser_md->type == Metadata_type_generic)
2078             {
2079                 WRBUF w = wrbuf_alloc();
2080                 if (relevance_snippet(se->relevance,
2081                                       (char*) value, ser_md->name, w))
2082                     rec_md->data.text.snippet = nmem_strdup(se->nmem,
2083                                                             wrbuf_cstr(w));
2084                 wrbuf_destroy(w);
2085             }
2086
2087
2088             wheretoput = &record->metadata[md_field_id];
2089             while (*wheretoput)
2090                 wheretoput = &(*wheretoput)->next;
2091             *wheretoput = rec_md;
2092         }
2093     }
2094
2095     if (check_limit_local(cl, record, record_no))
2096     {
2097         if (type)
2098             xmlFree(type);
2099         if (value)
2100             xmlFree(value);
2101         return -2;
2102     }
2103     cluster = reclist_insert(se->reclist, se->relevance, service, record,
2104                              merge_keys, &se->total_merged);
2105     if (!cluster)
2106     {
2107         if (type)
2108             xmlFree(type);
2109         if (value)
2110             xmlFree(value);
2111         return 0; // complete match with existing record
2112     }
2113
2114     {
2115         const char *use_term_factor_str =
2116             session_setting_oneval(sdb, PZ_TERMLIST_TERM_FACTOR);
2117         if (use_term_factor_str && use_term_factor_str[0] == '1')
2118         {
2119             int maxrecs = client_get_maxrecs(cl);
2120             int hits = (int) client_get_hits(cl);
2121             term_factor = MAX(hits, maxrecs) /  MAX(1, maxrecs);
2122             assert(term_factor >= 1);
2123             session_log(se, YLOG_DEBUG, "Using term factor: %d (%d / %d)",
2124                         term_factor, MAX(hits, maxrecs), MAX(1, maxrecs));
2125         }
2126     }
2127
2128     if (global_parameters.dump_records)
2129         session_log(se, YLOG_LOG, "Cluster id %s from %s (#%d)", cluster->recid,
2130                     sdb->database->id, record_no);
2131
2132     // original metadata, to check if first existence of a field
2133     metadata0 = xmalloc(sizeof(*metadata0) * service->num_metadata);
2134     memcpy(metadata0, cluster->metadata,
2135            sizeof(*metadata0) * service->num_metadata);
2136
2137     ingest_nmem = nmem_create();
2138     if (se->rank)
2139     {
2140         yaz_log(YLOG_LOG, "local in sort : %s", se->rank);
2141         nmem_strsplit_escape2(ingest_nmem, ",", se->rank, &rank_values,
2142                               &rank_num, 1, '\\', 1);
2143     }
2144
2145     // now parsing XML record and adding data to cluster or record metadata
2146     for (n = root->children; n; n = n->next)
2147     {
2148         pp2_charset_token_t prt;
2149         if (type)
2150             xmlFree(type);
2151         if (value)
2152             xmlFree(value);
2153         type = value = 0;
2154
2155         if (n->type != XML_ELEMENT_NODE)
2156             continue;
2157         if (!strcmp((const char *) n->name, "metadata"))
2158         {
2159             struct conf_metadata *ser_md = 0;
2160             struct conf_sortkey *ser_sk = 0;
2161             struct record_metadata **wheretoput = 0;
2162             struct record_metadata *rec_md = 0;
2163             int md_field_id = -1;
2164             int sk_field_id = -1;
2165             const char *rank = 0;
2166             xmlChar *xml_rank = 0;
2167
2168             type = xmlGetProp(n, (xmlChar *) "type");
2169             value = xmlNodeListGetString(xdoc, n->children, 1);
2170
2171             if (!type || !value || !*value)
2172                 continue;
2173
2174             md_field_id
2175                 = conf_service_metadata_field_id(service, (const char *) type);
2176             if (md_field_id < 0)
2177                 continue;
2178
2179             ser_md = &service->metadata[md_field_id];
2180
2181             if (ser_md->sortkey_offset >= 0)
2182             {
2183                 sk_field_id = ser_md->sortkey_offset;
2184                 ser_sk = &service->sortkeys[sk_field_id];
2185             }
2186
2187             // merged metadata
2188             rec_md = record_metadata_init(se->nmem, (const char *) value,
2189                                           ser_md->type, 0);
2190
2191             // see if the field was not in cluster already (from beginning)
2192
2193             if (!rec_md)
2194                 continue;
2195
2196             if (rank_num)
2197             {
2198                 int i;
2199                 for (i = 0; i < rank_num; i++)
2200                 {
2201                     const char *val = rank_values[i];
2202                     const char *cp = strchr(val, '=');
2203                     if (!cp)
2204                         continue;
2205                     if ((cp - val) == strlen((const char *) type)
2206                         && !memcmp(val, type, cp - val))
2207                     {
2208                         rank = cp + 1;
2209                         break;
2210                     }
2211                 }
2212             }
2213             else
2214             {
2215                 xml_rank = xmlGetProp(n, (xmlChar *) "rank");
2216                 rank = xml_rank ? (const char *) xml_rank : ser_md->rank;
2217             }
2218
2219             wheretoput = &cluster->metadata[md_field_id];
2220
2221             if (ser_md->merge == Metadata_merge_first)
2222             {
2223                 if (!metadata0[md_field_id])
2224                 {
2225                     while (*wheretoput)
2226                         wheretoput = &(*wheretoput)->next;
2227                     *wheretoput = rec_md;
2228                 }
2229             }
2230             else if (ser_md->merge == Metadata_merge_unique)
2231             {
2232                 while (*wheretoput)
2233                 {
2234                     if (!strcmp((const char *) (*wheretoput)->data.text.disp,
2235                                 rec_md->data.text.disp))
2236                         break;
2237                     wheretoput = &(*wheretoput)->next;
2238                 }
2239                 if (!*wheretoput)
2240                     *wheretoput = rec_md;
2241             }
2242             else if (ser_md->merge == Metadata_merge_longest)
2243             {
2244                 if (!*wheretoput
2245                     || strlen(rec_md->data.text.disp)
2246                     > strlen((*wheretoput)->data.text.disp))
2247                 {
2248                     *wheretoput = rec_md;
2249                     if (ser_sk)
2250                     {
2251                         const char *sort_str = 0;
2252                         int skip_article =
2253                             ser_sk->type == Metadata_type_skiparticle;
2254
2255                         if (!cluster->sortkeys[sk_field_id])
2256                             cluster->sortkeys[sk_field_id] =
2257                                 nmem_malloc(se->nmem,
2258                                             sizeof(union data_types));
2259
2260                         prt =
2261                             pp2_charset_token_create(service->charsets, "sort");
2262
2263                         pp2_charset_token_first(prt, rec_md->data.text.disp,
2264                                                 skip_article);
2265
2266                         pp2_charset_token_next(prt);
2267
2268                         sort_str = pp2_get_sort(prt);
2269
2270                         cluster->sortkeys[sk_field_id]->text.disp =
2271                             rec_md->data.text.disp;
2272                         if (!sort_str)
2273                         {
2274                             sort_str = rec_md->data.text.disp;
2275                             session_log(se, YLOG_WARN,
2276                                     "Could not make sortkey. Bug #1858");
2277                         }
2278                         cluster->sortkeys[sk_field_id]->text.sort =
2279                             nmem_strdup(se->nmem, sort_str);
2280                         pp2_charset_token_destroy(prt);
2281                     }
2282                 }
2283             }
2284             else if (ser_md->merge == Metadata_merge_all)
2285             {
2286                 while (*wheretoput)
2287                     wheretoput = &(*wheretoput)->next;
2288                 *wheretoput = rec_md;
2289             }
2290             else if (ser_md->merge == Metadata_merge_range)
2291             {
2292                 if (!*wheretoput)
2293                 {
2294                     *wheretoput = rec_md;
2295                     if (ser_sk)
2296                         cluster->sortkeys[sk_field_id]
2297                             = &rec_md->data;
2298                 }
2299                 else
2300                 {
2301                     int this_min = rec_md->data.number.min;
2302                     int this_max = rec_md->data.number.max;
2303                     if (this_min < (*wheretoput)->data.number.min)
2304                         (*wheretoput)->data.number.min = this_min;
2305                     if (this_max > (*wheretoput)->data.number.max)
2306                         (*wheretoput)->data.number.max = this_max;
2307                 }
2308             }
2309
2310             // ranking of _all_ fields enabled ...
2311             if (rank)
2312             {
2313                 relevance_countwords(se->relevance, cluster,
2314                                      (char *) value, rank, ser_md->name);
2315             }
2316             // construct facets ... unless the client already has reported them
2317             if (ser_md->termlist && !client_has_facet(cl, (char *) type))
2318             {
2319                 if (ser_md->type == Metadata_type_year)
2320                 {
2321                     char year[64];
2322                     sprintf(year, "%d", rec_md->data.number.max);
2323
2324                     add_facet(se, (char *) type, year, term_factor);
2325                     if (rec_md->data.number.max != rec_md->data.number.min)
2326                     {
2327                         sprintf(year, "%d", rec_md->data.number.min);
2328                         add_facet(se, (char *) type, year, term_factor);
2329                     }
2330                 }
2331                 else
2332                     add_facet(se, (char *) type, (char *) value, term_factor);
2333             }
2334
2335             // cleaning up
2336             if (xml_rank)
2337                 xmlFree(xml_rank);
2338             xmlFree(type);
2339             xmlFree(value);
2340             type = value = 0;
2341         }
2342         else
2343         {
2344             if (se->number_of_warnings_unknown_elements == 0)
2345                 session_log(se, YLOG_WARN,
2346                         "Unexpected element in internal record: %s", n->name);
2347             se->number_of_warnings_unknown_elements++;
2348         }
2349     }
2350     if (type)
2351         xmlFree(type);
2352     if (value)
2353         xmlFree(value);
2354
2355     nmem_destroy(ingest_nmem);
2356     xfree(metadata0);
2357     relevance_donerecord(se->relevance, cluster);
2358     se->total_records++;
2359
2360     return 0;
2361 }
2362
2363 void session_log(struct session *s, int level, const char *fmt, ...)
2364 {
2365     char buf[1024];
2366     va_list ap;
2367     va_start(ap, fmt);
2368
2369     yaz_vsnprintf(buf, sizeof(buf)-30, fmt, ap);
2370     yaz_log(level, "Session %u: %s", s ? s->session_id : 0, buf);
2371
2372     va_end(ap);
2373 }
2374
2375 /*
2376  * Local variables:
2377  * c-basic-offset: 4
2378  * c-file-style: "Stroustrup"
2379  * indent-tabs-mode: nil
2380  * End:
2381  * vim: shiftwidth=4 tabstop=8 expandtab
2382  */
2383