Comments
[pazpar2-moved-to-github.git] / src / session.c
1 /* This file is part of Pazpar2.
2    Copyright (C) 2006-2012 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 #include <yaz/gettimeofday.h>
60
61 #define USE_TIMING 0
62 #if USE_TIMING
63 #include <yaz/timing.h>
64 #endif
65
66 #include "ppmutex.h"
67 #include "parameters.h"
68 #include "session.h"
69 #include "eventl.h"
70 #include "http.h"
71 #include "termlists.h"
72 #include "reclists.h"
73 #include "relevance.h"
74 #include "database.h"
75 #include "client.h"
76 #include "settings.h"
77 #include "normalize7bit.h"
78
79 #include <libxml/tree.h>
80
81 #define TERMLIST_HIGH_SCORE 25
82
83 #define MAX_CHUNK 15
84
85 #define MAX(a,b) ((a)>(b)?(a):(b))
86
87 // Note: Some things in this structure will eventually move to configuration
88 struct parameters global_parameters = 
89 {
90     0,   // dump_records
91     0,   // debug_mode
92     0,   // predictable sessions
93 };
94
95 struct client_list {
96     struct client *client;
97     struct client_list *next;
98 };
99
100 /* session counting (1) , disable client counting (0) */
101 static YAZ_MUTEX g_session_mutex = 0;
102 static int no_sessions = 0;
103 static int no_session_total = 0;
104
105 static int session_use(int delta)
106 {
107     int sessions;
108     if (!g_session_mutex)
109         yaz_mutex_create(&g_session_mutex);
110     yaz_mutex_enter(g_session_mutex);
111     no_sessions += delta;
112     if (delta > 0)
113         no_session_total += delta;
114     sessions = no_sessions;
115     yaz_mutex_leave(g_session_mutex);
116     yaz_log(YLOG_DEBUG, "%s sessions=%d", delta == 0 ? "" : (delta > 0 ? "INC" : "DEC"), no_sessions);
117     return sessions;
118 }
119
120 int sessions_count(void)
121 {
122     return session_use(0);
123 }
124
125 int session_count_total(void)
126 {
127     int total = 0;
128     if (!g_session_mutex)
129         return 0;
130     yaz_mutex_enter(g_session_mutex);
131     total = no_session_total;
132     yaz_mutex_leave(g_session_mutex);
133     return total;
134 }
135
136 static void log_xml_doc(xmlDoc *doc)
137 {
138     FILE *lf = yaz_log_file();
139     xmlChar *result = 0;
140     int len = 0;
141 #if LIBXML_VERSION >= 20600
142     xmlDocDumpFormatMemory(doc, &result, &len, 1);
143 #else
144     xmlDocDumpMemory(doc, &result, &len);
145 #endif
146     if (lf && len)
147     {
148         (void) fwrite(result, 1, len, lf);
149         fprintf(lf, "\n");
150     }
151     xmlFree(result);
152 }
153
154 static void session_enter(struct session *s)
155 {
156     yaz_mutex_enter(s->session_mutex);
157 }
158
159 static void session_leave(struct session *s)
160 {
161     yaz_mutex_leave(s->session_mutex);
162 }
163
164 static void session_normalize_facet(struct session *s, const char *type,
165                                     const char *value,
166                                     WRBUF display_wrbuf,
167                                     WRBUF facet_wrbuf)
168 {
169     struct conf_service *service = s->service;
170     pp2_charset_token_t prt;
171     const char *facet_component;
172     int i;
173     const char *icu_chain_id = 0;
174
175     for (i = 0; i < service->num_metadata; i++)
176         if (!strcmp((service->metadata + i)->name, type))
177             icu_chain_id = (service->metadata + i)->facetrule;
178     if (!icu_chain_id)
179         icu_chain_id = "facet";
180     prt = pp2_charset_token_create(service->charsets, icu_chain_id);
181     if (!prt)
182     {
183         yaz_log(YLOG_FATAL, "Unknown ICU chain '%s' for facet of type '%s'",
184                 icu_chain_id, type);
185         wrbuf_destroy(facet_wrbuf);
186         wrbuf_destroy(display_wrbuf);
187         return;
188     }
189     pp2_charset_token_first(prt, value, 0);
190     while ((facet_component = pp2_charset_token_next(prt)))
191     {
192         const char *display_component;
193         if (*facet_component)
194         {
195             if (wrbuf_len(facet_wrbuf))
196                 wrbuf_puts(facet_wrbuf, " ");
197             wrbuf_puts(facet_wrbuf, facet_component);
198         }
199         display_component = pp2_get_display(prt);
200         if (display_component)
201         {
202             if (wrbuf_len(display_wrbuf))
203                 wrbuf_puts(display_wrbuf, " ");
204             wrbuf_puts(display_wrbuf, display_component);
205         }
206     }
207     pp2_charset_token_destroy(prt);
208 }
209
210 void add_facet(struct session *s, const char *type, const char *value, int count)
211 {
212     WRBUF facet_wrbuf = wrbuf_alloc();
213     WRBUF display_wrbuf = wrbuf_alloc();
214
215     session_normalize_facet(s, type, value, display_wrbuf, facet_wrbuf);
216  
217     if (wrbuf_len(facet_wrbuf))
218     {
219         int i;
220         for (i = 0; i < s->num_termlists; i++)
221             if (!strcmp(s->termlists[i].name, type))
222                 break;
223         if (i == s->num_termlists)
224         {
225             if (i == SESSION_MAX_TERMLISTS)
226             {
227                 session_log(s, YLOG_FATAL, "Too many termlists");
228                 wrbuf_destroy(facet_wrbuf);
229                 wrbuf_destroy(display_wrbuf);
230                 return;
231             }
232             
233             s->termlists[i].name = nmem_strdup(s->nmem, type);
234             s->termlists[i].termlist 
235                 = termlist_create(s->nmem, TERMLIST_HIGH_SCORE);
236             s->num_termlists = i + 1;
237         }
238         
239 #if 0
240         session_log(s, YLOG_DEBUG, "Facets for %s: %s norm:%s (%d)", type, value, wrbuf_cstr(facet_wrbuf), count);
241 #endif
242         termlist_insert(s->termlists[i].termlist, wrbuf_cstr(display_wrbuf),
243                         wrbuf_cstr(facet_wrbuf), count);
244     }
245     wrbuf_destroy(facet_wrbuf);
246     wrbuf_destroy(display_wrbuf);
247 }
248
249 static xmlDoc *record_to_xml(struct session *se,
250                              struct session_database *sdb, const char *rec)
251 {
252     struct database *db = sdb->database;
253     xmlDoc *rdoc = 0;
254
255     rdoc = xmlParseMemory(rec, strlen(rec));
256
257     if (!rdoc)
258     {
259         session_log(se, YLOG_FATAL, "Non-wellformed XML received from %s",
260                     db->id);
261         return 0;
262     }
263
264     if (global_parameters.dump_records)
265     {
266         session_log(se, YLOG_LOG, "Un-normalized record from %s", db->id);
267         log_xml_doc(rdoc);
268     }
269
270     return rdoc;
271 }
272
273 #define MAX_XSLT_ARGS 16
274
275 // Add static values from session database settings if applicable
276 static void insert_settings_parameters(struct session_database *sdb,
277                                        struct conf_service *service,
278                                        char **parms,
279                                        NMEM nmem)
280 {
281     int i;
282     int nparms = 0;
283     int offset = 0;
284
285     for (i = 0; i < service->num_metadata; i++)
286     {
287         struct conf_metadata *md = &service->metadata[i];
288         int setting;
289
290         if (md->setting == Metadata_setting_parameter &&
291             (setting = settings_lookup_offset(service, md->name)) >= 0)
292         {
293             const char *val = session_setting_oneval(sdb, setting);
294             if (val && nparms < MAX_XSLT_ARGS)
295             {
296                 char *buf;
297                 int len = strlen(val);
298                 buf = nmem_malloc(nmem, len + 3);
299                 buf[0] = '\'';
300                 strcpy(buf + 1, val);
301                 buf[len+1] = '\'';
302                 buf[len+2] = '\0';
303                 parms[offset++] = md->name;
304                 parms[offset++] = buf;
305                 nparms++;
306             }
307         }
308     }
309     parms[offset] = 0;
310 }
311
312 // Add static values from session database settings if applicable
313 static void insert_settings_values(struct session_database *sdb, xmlDoc *doc,
314     struct conf_service *service)
315 {
316     int i;
317
318     for (i = 0; i < service->num_metadata; i++)
319     {
320         struct conf_metadata *md = &service->metadata[i];
321         int offset;
322
323         if (md->setting == Metadata_setting_postproc &&
324             (offset = settings_lookup_offset(service, md->name)) >= 0)
325         {
326             const char *val = session_setting_oneval(sdb, offset);
327             if (val)
328             {
329                 xmlNode *r = xmlDocGetRootElement(doc);
330                 xmlNode *n = xmlNewTextChild(r, 0, (xmlChar *) "metadata",
331                                              (xmlChar *) val);
332                 xmlSetProp(n, (xmlChar *) "type", (xmlChar *) md->name);
333             }
334         }
335     }
336 }
337
338 static xmlDoc *normalize_record(struct session *se,
339                                 struct session_database *sdb,
340                                 struct conf_service *service,
341                                 const char *rec, NMEM nmem)
342 {
343     xmlDoc *rdoc = record_to_xml(se, sdb, rec);
344
345     if (rdoc)
346     {
347         char *parms[MAX_XSLT_ARGS*2+1];
348         
349         insert_settings_parameters(sdb, service, parms, nmem);
350         
351         if (normalize_record_transform(sdb->map, &rdoc, (const char **)parms))
352         {
353             session_log(se, YLOG_WARN, "Normalize failed from %s",
354                         sdb->database->id);
355         }
356         else
357         {
358             insert_settings_values(sdb, rdoc, service);
359             
360             if (global_parameters.dump_records)
361             {
362                 session_log(se, YLOG_LOG, "Normalized record from %s", 
363                             sdb->database->id);
364                 log_xml_doc(rdoc);
365             }
366         }
367     }
368     return rdoc;
369 }
370
371 void session_settings_dump(struct session *se,
372                            struct session_database *db,
373                            WRBUF w)
374 {
375     if (db->settings)
376     {
377         int i, num = db->num_settings;
378         for (i = 0; i < num; i++)
379         {
380             struct setting *s = db->settings[i];
381             for (;s ; s = s->next)
382             {
383                 wrbuf_puts(w, "<set name=\"");
384                 wrbuf_xmlputs(w, s->name);
385                 wrbuf_puts(w, "\" value=\"");
386                 wrbuf_xmlputs(w, s->value);
387                 wrbuf_puts(w, "\"/>");
388             }
389             if (db->settings[i])
390                 wrbuf_puts(w, "\n");
391         }
392     }
393 }
394
395 // Retrieve first defined value for 'name' for given database.
396 // Will be extended to take into account user associated with session
397 const char *session_setting_oneval(struct session_database *db, int offset)
398 {
399     if (offset >= db->num_settings || !db->settings[offset])
400         return "";
401     return db->settings[offset]->value;
402 }
403
404 // Prepare XSLT stylesheets for record normalization
405 // Structures are allocated on the session_wide nmem to avoid having
406 // to recompute this for every search. This would lead
407 // to leaking if a single session was to repeatedly change the PZ_XSLT
408 // setting. However, this is not a realistic use scenario.
409 static int prepare_map(struct session *se, struct session_database *sdb)
410 {
411     const char *s;
412
413     if (sdb->settings && sdb->settings[PZ_XSLT] && !sdb->map &&
414         (s = session_setting_oneval(sdb, PZ_XSLT)))        
415     {
416         char auto_stylesheet[256];
417
418         if (!strcmp(s, "auto"))
419         {
420             const char *request_syntax = session_setting_oneval(
421                 sdb, PZ_REQUESTSYNTAX);
422             if (request_syntax)
423             {
424                 char *cp;
425                 yaz_snprintf(auto_stylesheet, sizeof(auto_stylesheet),
426                              "%s.xsl", request_syntax);
427                 for (cp = auto_stylesheet; *cp; cp++)
428                 {
429                     /* deliberately only consider ASCII */
430                     if (*cp > 32 && *cp < 127)
431                         *cp = tolower(*cp);
432                 }
433                 s = auto_stylesheet;
434             }
435             else
436             {
437                 session_log(se, YLOG_WARN,
438                             "No pz:requestsyntax for auto stylesheet");
439             }
440         }
441         sdb->map = normalize_cache_get(se->normalize_cache,
442                                        se->service, s);
443         if (!sdb->map)
444             return -1;
445     }
446     return 0;
447 }
448
449 // called if watch should be removed because http_channel is to be destroyed
450 static void session_watch_cancel(void *data, struct http_channel *c,
451                                  void *data2)
452 {
453     struct session_watchentry *ent = data;
454
455     ent->fun = 0;
456     ent->data = 0;
457     ent->obs = 0;
458 }
459
460 // set watch. Returns 0=OK, -1 if watch is already set
461 int session_set_watch(struct session *s, int what, 
462                       session_watchfun fun, void *data,
463                       struct http_channel *chan)
464 {
465     int ret;
466     session_enter(s);
467     if (s->watchlist[what].fun)
468         ret = -1;
469     else
470     {
471         
472         s->watchlist[what].fun = fun;
473         s->watchlist[what].data = data;
474         s->watchlist[what].obs = http_add_observer(chan, &s->watchlist[what],
475                                                    session_watch_cancel);
476         ret = 0;
477     }
478     session_leave(s);
479     return ret;
480 }
481
482 void session_alert_watch(struct session *s, int what)
483 {
484     assert(s);
485     session_enter(s);
486     if (s->watchlist[what].fun)
487     {
488         /* our watch is no longer associated with http_channel */
489         void *data;
490         session_watchfun fun;
491
492         http_remove_observer(s->watchlist[what].obs);
493         fun  = s->watchlist[what].fun;
494         data = s->watchlist[what].data;
495
496         /* reset watch before fun is invoked - in case fun wants to set
497            it again */
498         s->watchlist[what].fun = 0;
499         s->watchlist[what].data = 0;
500         s->watchlist[what].obs = 0;
501
502         session_leave(s);
503         session_log(s, YLOG_DEBUG,
504                     "Alert Watch: %d calling function: %p", what, fun);
505         fun(data);
506     }
507     else
508         session_leave(s);
509 }
510
511 //callback for grep_databases
512 static void select_targets_callback(struct session *se,
513                                     struct session_database *db)
514 {
515     struct client *cl;
516     struct client_list *l;
517
518     for (l = se->clients_cached; l; l = l->next)
519         if (client_get_database(l->client) == db)
520             break;
521
522     if (l)
523         cl = l->client;
524     else
525     {
526         cl = client_create(db->database->id);
527         client_set_database(cl, db);
528
529         l = xmalloc(sizeof(*l));
530         l->client = cl;
531         l->next = se->clients_cached;
532         se->clients_cached = l;
533     }
534     /* set session always. If may be 0 if client is not active */
535     client_set_session(cl, se);
536
537     l = xmalloc(sizeof(*l));
538     l->client = cl;
539     l->next = se->clients_active;
540     se->clients_active = l;
541 }
542
543 static void session_reset_active_clients(struct session *se,
544                                          struct client_list *new_list)
545 {
546     struct client_list *l;
547
548     session_enter(se);
549     l = se->clients_active;
550     se->clients_active = new_list;
551     session_leave(se);
552
553     while (l)
554     {
555         struct client_list *l_next = l->next;
556
557         client_lock(l->client); 
558         client_set_session(l->client, 0); /* mark client inactive */
559         client_unlock(l->client);
560
561         xfree(l);
562         l = l_next;
563     }
564 }
565
566 static void session_remove_cached_clients(struct session *se)
567 {
568     struct client_list *l;
569
570     session_reset_active_clients(se, 0);
571
572     session_enter(se);
573     l = se->clients_cached;
574     se->clients_cached = 0;
575     session_leave(se);
576
577     while (l)
578     {
579         struct client_list *l_next = l->next;
580         client_lock(l->client);
581         client_set_session(l->client, 0);
582         client_set_database(l->client, 0);
583         client_unlock(l->client);
584         client_destroy(l->client);
585         xfree(l);
586         l = l_next;
587     }
588 }
589
590 // Associates a set of clients with a session;
591 // Note: Session-databases represent databases with per-session 
592 // setting overrides
593 static int select_targets(struct session *se, const char *filter)
594 {
595     return session_grep_databases(se, filter, select_targets_callback);
596 }
597
598 int session_active_clients(struct session *s)
599 {
600     struct client_list *l;
601     int res = 0;
602
603     for (l = s->clients_active; l; l = l->next)
604         if (client_is_active(l->client))
605             res++;
606
607     return res;
608 }
609
610 int session_is_preferred_clients_ready(struct session *s)
611 {
612     struct client_list *l;
613     int res = 0;
614
615     for (l = s->clients_active; l; l = l->next)
616         if (client_is_active_preferred(l->client))
617             res++;
618     session_log(s, YLOG_DEBUG, "Has %d active preferred clients.", res);
619     return res == 0;
620 }
621
622 void session_sort(struct session *se, const char *field, int increasing)
623 {
624     struct session_sorted_results *sr;
625     struct client_list *l;
626
627     session_enter(se);
628
629     /* see if we already have sorted for this critieria */
630     for (sr = se->sorted_results; sr; sr = sr->next)
631     {
632         if (!strcmp(field, sr->field) && increasing == sr->increasing)
633             break;
634     }
635     if (sr)
636     {
637         yaz_log(YLOG_LOG, "search_sort: field=%s increasing=%d already fetched",
638                 field, increasing);
639         session_leave(se);
640         return;
641     }
642     yaz_log(YLOG_LOG, "search_sort: field=%s increasing=%d must fetch",
643             field, increasing);
644     sr = nmem_malloc(se->nmem, sizeof(*sr));
645     sr->field = nmem_strdup(se->nmem, field);
646     sr->increasing = increasing;
647     sr->next = se->sorted_results;
648     se->sorted_results = sr;
649     
650     for (l = se->clients_active; l; l = l->next)
651     {
652         struct client *cl = l->client;
653         if (client_get_state(cl) == Client_Connecting ||
654             client_get_state(cl) == Client_Idle ||
655             client_get_state(cl) == Client_Working)
656             client_start_search(cl);
657     }
658     session_leave(se);
659 }
660
661 enum pazpar2_error_code session_search(struct session *se,
662                                        const char *query,
663                                        const char *startrecs,
664                                        const char *maxrecs,
665                                        const char *filter,
666                                        const char *limit,
667                                        const char **addinfo,
668                                        const char *sort_field, int increasing)
669 {
670     int live_channels = 0;
671     int no_working = 0;
672     int no_failed_query = 0;
673     int no_failed_limit = 0;
674     struct client_list *l, *l0;
675     struct timeval tval;
676     facet_limits_t facet_limits;
677
678     session_log(se, YLOG_DEBUG, "Search");
679
680     *addinfo = 0;
681
682     if (se->settings_modified)
683         session_remove_cached_clients(se);
684     else
685         session_reset_active_clients(se, 0);
686     
687     session_enter(se);
688     reclist_destroy(se->reclist);
689     se->reclist = 0;
690     se->settings_modified = 0;
691     relevance_destroy(&se->relevance);
692     nmem_reset(se->nmem);
693     se->total_records = se->total_merged = 0;
694     se->num_termlists = 0;
695
696     /* reset list of sorted results and clear to relevance search */
697     se->sorted_results = nmem_malloc(se->nmem, sizeof(*se->sorted_results));
698     se->sorted_results->field = nmem_strdup(se->nmem, sort_field);
699     se->sorted_results->increasing = increasing;
700     se->sorted_results->next = 0;
701     
702     live_channels = select_targets(se, filter);
703     if (!live_channels)
704     {
705         session_leave(se);
706         return PAZPAR2_NO_TARGETS;
707     }
708     se->reclist = reclist_create(se->nmem);
709
710     yaz_gettimeofday(&tval);
711     
712     tval.tv_sec += 5;
713
714     facet_limits = facet_limits_create(limit);
715     if (!facet_limits)
716     {
717         *addinfo = "limit";
718         session_leave(se);
719         return PAZPAR2_MALFORMED_PARAMETER_VALUE;
720     }
721
722     l0 = se->clients_active;
723     se->clients_active = 0;
724     session_leave(se);
725
726     for (l = l0; l; l = l->next)
727     {
728         int parse_ret;
729         struct client *cl = l->client;
730
731         if (prepare_map(se, client_get_database(cl)) < 0)
732             continue;
733
734         parse_ret = client_parse_query(cl, query, facet_limits, startrecs,
735             maxrecs);
736         if (parse_ret == -1)
737             no_failed_query++;
738         else if (parse_ret == -2)
739             no_failed_limit++;
740         else
741         {
742             int r =
743                 client_prep_connection(cl, se->service->z3950_operation_timeout,
744                                        se->service->z3950_session_timeout,
745                                        se->service->server->iochan_man,
746                                        &tval);
747             if (parse_ret == 1 && r == 2)
748             {
749                 session_log(se, YLOG_LOG, "client REUSE %s", client_get_id(cl));
750                 client_reingest(cl);
751             }
752             else if (r)
753             {
754                 session_log(se, YLOG_LOG, "client NEW %s", client_get_id(cl));
755                 client_start_search(cl);
756             }
757             no_working++;
758         }
759     }
760     facet_limits_destroy(facet_limits);
761     session_reset_active_clients(se, l0);
762
763     if (no_working == 0)
764     {
765         if (no_failed_query > 0)
766         {
767             *addinfo = "query";
768             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
769         }
770         else if (no_failed_limit > 0)
771         {
772             *addinfo = "limit";
773             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
774         }
775         else
776             return PAZPAR2_NO_TARGETS;
777     }
778     yaz_log(YLOG_LOG, "session_start_search done");
779     return PAZPAR2_NO_ERROR;
780 }
781
782 // Creates a new session_database object for a database
783 static void session_init_databases_fun(void *context, struct database *db)
784 {
785     struct session *se = (struct session *) context;
786     struct session_database *new = nmem_malloc(se->session_nmem, sizeof(*new));
787     int i;
788
789     new->database = db;
790     
791     new->map = 0;
792     assert(db->settings);
793     new->settings = nmem_malloc(se->session_nmem,
794                                 sizeof(struct settings *) * db->num_settings);
795     new->num_settings = db->num_settings;
796     for (i = 0; i < db->num_settings; i++)
797     {
798         struct setting *setting = db->settings[i];
799         new->settings[i] = setting;
800     }
801     new->next = se->databases;
802     se->databases = new;
803 }
804
805 // Doesn't free memory associated with sdb -- nmem takes care of that
806 static void session_database_destroy(struct session_database *sdb)
807 {
808     sdb->map = 0;
809 }
810
811 // Initialize session_database list -- this represents this session's view
812 // of the database list -- subject to modification by the settings ws command
813 void session_init_databases(struct session *se)
814 {
815     se->databases = 0;
816     predef_grep_databases(se, se->service, session_init_databases_fun);
817 }
818
819 // Probably session_init_databases_fun should be refactored instead of
820 // called here.
821 static struct session_database *load_session_database(struct session *se, 
822                                                       char *id)
823 {
824     struct database *db = new_database(id, se->session_nmem);
825
826     session_init_databases_fun((void*) se, db);
827
828     // New sdb is head of se->databases list
829     return se->databases;
830 }
831
832 // Find an existing session database. If not found, load it
833 static struct session_database *find_session_database(struct session *se, 
834                                                       char *id)
835 {
836     struct session_database *sdb;
837
838     for (sdb = se->databases; sdb; sdb = sdb->next)
839         if (!strcmp(sdb->database->id, id))
840             return sdb;
841     return load_session_database(se, id);
842 }
843
844 // Apply a session override to a database
845 void session_apply_setting(struct session *se, char *dbname, char *setting,
846                            char *value)
847 {
848     struct session_database *sdb = find_session_database(se, dbname);
849     struct conf_service *service = se->service;
850     struct setting *new = nmem_malloc(se->session_nmem, sizeof(*new));
851     int offset = settings_create_offset(service, setting);
852
853     expand_settings_array(&sdb->settings, &sdb->num_settings, offset,
854                           se->session_nmem);
855     new->precedence = 0;
856     new->target = dbname;
857     new->name = setting;
858     new->value = value;
859     new->next = sdb->settings[offset];
860     sdb->settings[offset] = new;
861
862     se->settings_modified = 1;
863
864     // Force later recompute of settings-driven data structures
865     // (happens when a search starts and client connections are prepared)
866     switch (offset)
867     {
868     case PZ_XSLT:
869         if (sdb->map)
870         {
871             sdb->map = 0;
872         }
873         break;
874     }
875 }
876
877 void session_destroy(struct session *se)
878 {
879     struct session_database *sdb;
880     session_log(se, YLOG_DEBUG, "Destroying");
881     session_use(-1);
882     session_remove_cached_clients(se);
883
884     for (sdb = se->databases; sdb; sdb = sdb->next)
885         session_database_destroy(sdb);
886     normalize_cache_destroy(se->normalize_cache);
887     relevance_destroy(&se->relevance);
888     reclist_destroy(se->reclist);
889     nmem_destroy(se->nmem);
890     service_destroy(se->service);
891     yaz_mutex_destroy(&se->session_mutex);
892 }
893
894 size_t session_get_memory_status(struct session *session) {
895     size_t session_nmem;
896     if (session == 0)
897         return 0;
898     session_enter(session);
899     session_nmem = nmem_total(session->nmem);
900     session_leave(session);
901     return session_nmem;
902 }
903
904
905 struct session *new_session(NMEM nmem, struct conf_service *service,
906                             unsigned session_id)
907 {
908     int i;
909     struct session *session = nmem_malloc(nmem, sizeof(*session));
910
911     char tmp_str[50];
912
913     sprintf(tmp_str, "session#%u", session_id);
914
915     session->session_id = session_id;
916     session_log(session, YLOG_DEBUG, "New");
917     session->service = service;
918     session->relevance = 0;
919     session->total_records = 0;
920     session->number_of_warnings_unknown_elements = 0;
921     session->number_of_warnings_unknown_metadata = 0;
922     session->num_termlists = 0;
923     session->reclist = 0;
924     session->clients_active = 0;
925     session->clients_cached = 0;
926     session->settings_modified = 0;
927     session->session_nmem = nmem;
928     session->nmem = nmem_create();
929     session->databases = 0;
930     for (i = 0; i <= SESSION_WATCH_MAX; i++)
931     {
932         session->watchlist[i].data = 0;
933         session->watchlist[i].fun = 0;
934     }
935     session->normalize_cache = normalize_cache_create();
936     session->session_mutex = 0;
937     pazpar2_mutex_create(&session->session_mutex, tmp_str);
938     session_use(1);
939     return session;
940 }
941
942 const char * client_get_suggestions_xml(struct client *cl, WRBUF wrbuf);
943
944 static struct hitsbytarget *hitsbytarget_nb(struct session *se,
945                                             int *count, NMEM nmem)
946 {
947     struct hitsbytarget *res = 0;
948     struct client_list *l;
949     size_t sz = 0;
950
951     for (l = se->clients_active; l; l = l->next)
952         sz++;
953
954     res = nmem_malloc(nmem, sizeof(*res) * sz);
955     *count = 0;
956     for (l = se->clients_active; l; l = l->next)
957     {
958         struct client *cl = l->client;
959         WRBUF w = wrbuf_alloc();
960         const char *name = session_setting_oneval(client_get_database(cl),
961                                                   PZ_NAME);
962
963         res[*count].id = client_get_id(cl);
964         res[*count].name = *name ? name : "Unknown";
965         res[*count].hits = client_get_hits(cl);
966         res[*count].approximation = client_get_approximation(cl);
967         res[*count].records = client_get_num_records(cl);
968         res[*count].filtered = client_get_num_records_filtered(cl);
969         res[*count].diagnostic =
970             client_get_diagnostic(cl, &res[*count].addinfo);
971         res[*count].state = client_get_state_str(cl);
972         res[*count].connected  = client_get_connection(cl) ? 1 : 0;
973         session_settings_dump(se, client_get_database(cl), w);
974         res[*count].settings_xml = nmem_strdup(nmem, wrbuf_cstr(w));
975         wrbuf_rewind(w);
976         wrbuf_puts(w, "");
977         res[*count].suggestions_xml = nmem_strdup(nmem, client_get_suggestions_xml(cl, w));
978         wrbuf_destroy(w);
979         (*count)++;
980     }
981     return res;
982 }
983
984 struct hitsbytarget *get_hitsbytarget(struct session *se, int *count, NMEM nmem)
985 {
986     struct hitsbytarget *p;
987     session_enter(se);
988     p = hitsbytarget_nb(se, count, nmem);
989     session_leave(se);
990     return p;
991 }
992     
993 struct termlist_score **get_termlist_score(struct session *se,
994                                            const char *name, int *num)
995 {
996     int i;
997     struct termlist_score **tl = 0;
998
999     session_enter(se);
1000     for (i = 0; i < se->num_termlists; i++)
1001         if (!strcmp((const char *) se->termlists[i].name, name))
1002         {
1003             tl = termlist_highscore(se->termlists[i].termlist, num);
1004             break;
1005         }
1006     session_leave(se);
1007     return tl;
1008 }
1009
1010 // Compares two hitsbytarget nodes by hitcount
1011 static int cmp_ht(const void *p1, const void *p2)
1012 {
1013     const struct hitsbytarget *h1 = p1;
1014     const struct hitsbytarget *h2 = p2;
1015     return h2->hits - h1->hits;
1016 }
1017
1018 // Compares two hitsbytarget nodes by hitcount
1019 static int cmp_ht_approx(const void *p1, const void *p2)
1020 {
1021     const struct hitsbytarget *h1 = p1;
1022     const struct hitsbytarget *h2 = p2;
1023     return h2->approximation - h1->approximation;
1024 }
1025
1026 static int targets_termlist_nb(WRBUF wrbuf, struct session *se, int num,
1027                                NMEM nmem, int version)
1028 {
1029     struct hitsbytarget *ht;
1030     int count, i;
1031
1032     ht = hitsbytarget_nb(se, &count, nmem);
1033     if (version >= 2)
1034         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht_approx);
1035     else
1036         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht);
1037     for (i = 0; i < count && i < num && ht[i].hits > 0; i++)
1038     {
1039
1040         // do only print terms which have display names
1041     
1042         wrbuf_puts(wrbuf, "<term>\n");
1043
1044         wrbuf_puts(wrbuf, "<id>");
1045         wrbuf_xmlputs(wrbuf, ht[i].id);
1046         wrbuf_puts(wrbuf, "</id>\n");
1047         
1048         wrbuf_puts(wrbuf, "<name>");
1049         if (!ht[i].name || !ht[i].name[0])
1050             wrbuf_xmlputs(wrbuf, "NO TARGET NAME");
1051         else
1052             wrbuf_xmlputs(wrbuf, ht[i].name);
1053         wrbuf_puts(wrbuf, "</name>\n");
1054         
1055         wrbuf_printf(wrbuf, "<frequency>" ODR_INT_PRINTF "</frequency>\n",
1056                      ht[i].hits);
1057
1058         if (version >= 2) {
1059             // Should not print if we know it isn't a approximation.
1060             wrbuf_printf(wrbuf, "<approximation>" ODR_INT_PRINTF "</approximation>\n", ht[i].approximation);
1061             wrbuf_printf(wrbuf, "<records>%d</records>\n", ht[i].records - ht[i].filtered);
1062             wrbuf_printf(wrbuf, "<filtered>%d</filtered>\n", ht[i].filtered);
1063         }
1064
1065         wrbuf_puts(wrbuf, "<state>");
1066         wrbuf_xmlputs(wrbuf, ht[i].state);
1067         wrbuf_puts(wrbuf, "</state>\n");
1068         
1069         wrbuf_printf(wrbuf, "<diagnostic>%d</diagnostic>\n", 
1070                      ht[i].diagnostic);
1071         wrbuf_puts(wrbuf, "</term>\n");
1072     }
1073     return count;
1074 }
1075
1076 void perform_termlist(struct http_channel *c, struct session *se,
1077                       const char *name, int num, int version)
1078 {
1079     int i, j;
1080     NMEM nmem_tmp = nmem_create();
1081     char **names;
1082     int num_names = 0;
1083
1084     if (!name)
1085         name = "*";
1086
1087     nmem_strsplit(nmem_tmp, ",", name, &names, &num_names);
1088
1089     session_enter(se);
1090
1091     for (j = 0; j < num_names; j++)
1092     {
1093         const char *tname;
1094         int must_generate_empty = 1; /* bug 5350 */
1095
1096         for (i = 0; i < se->num_termlists; i++)
1097         {
1098             tname = se->termlists[i].name;
1099             if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1100             {
1101                 struct termlist_score **p = 0;
1102                 int len;
1103
1104                 wrbuf_puts(c->wrbuf, "<list name=\"");
1105                 wrbuf_xmlputs(c->wrbuf, tname);
1106                 wrbuf_puts(c->wrbuf, "\">\n");
1107                 must_generate_empty = 0;
1108
1109                 p = termlist_highscore(se->termlists[i].termlist, &len);
1110                 if (p)
1111                 {
1112                     int i;
1113                     for (i = 0; i < len && i < num; i++)
1114                     {
1115                         // prevent sending empty term elements
1116                         if (!p[i]->display_term || !p[i]->display_term[0])
1117                             continue;
1118                         
1119                         wrbuf_puts(c->wrbuf, "<term>");
1120                         wrbuf_puts(c->wrbuf, "<name>");
1121                         wrbuf_xmlputs(c->wrbuf, p[i]->display_term);
1122                         wrbuf_puts(c->wrbuf, "</name>");
1123                         
1124                         wrbuf_printf(c->wrbuf, 
1125                                      "<frequency>%d</frequency>", 
1126                                      p[i]->frequency);
1127                         wrbuf_puts(c->wrbuf, "</term>\n");
1128                     }
1129                 }
1130                 wrbuf_puts(c->wrbuf, "</list>\n");
1131             }
1132         }
1133         tname = "xtargets";
1134         if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1135         {
1136             wrbuf_puts(c->wrbuf, "<list name=\"");
1137             wrbuf_xmlputs(c->wrbuf, tname);
1138             wrbuf_puts(c->wrbuf, "\">\n");
1139
1140             targets_termlist_nb(c->wrbuf, se, num, c->nmem, version);
1141             wrbuf_puts(c->wrbuf, "</list>\n");
1142             must_generate_empty = 0;
1143         }
1144         if (must_generate_empty)
1145         {
1146             wrbuf_puts(c->wrbuf, "<list name=\"");
1147             wrbuf_xmlputs(c->wrbuf, names[j]);
1148             wrbuf_puts(c->wrbuf, "\"/>\n");
1149         }
1150     }
1151     session_leave(se);
1152     nmem_destroy(nmem_tmp);
1153 }
1154
1155 #ifdef MISSING_HEADERS
1156 void report_nmem_stats(void)
1157 {
1158     size_t in_use, is_free;
1159
1160     nmem_get_memory_in_use(&in_use);
1161     nmem_get_memory_free(&is_free);
1162
1163     yaz_log(YLOG_LOG, "nmem stat: use=%ld free=%ld", 
1164             (long) in_use, (long) is_free);
1165 }
1166 #endif
1167
1168 struct record_cluster *show_single_start(struct session *se, const char *id,
1169                                          struct record_cluster **prev_r,
1170                                          struct record_cluster **next_r)
1171 {
1172     struct record_cluster *r = 0;
1173
1174     session_enter(se);
1175     *prev_r = 0;
1176     *next_r = 0;
1177     if (se->reclist)
1178     {
1179         reclist_enter(se->reclist);
1180         while ((r = reclist_read_record(se->reclist)))
1181         {
1182             if (!strcmp(r->recid, id))
1183             {
1184                 *next_r = reclist_read_record(se->reclist);
1185                 break;
1186             }
1187             *prev_r = r;
1188         }
1189         reclist_leave(se->reclist);
1190     }
1191     if (!r)
1192         session_leave(se);
1193     return r;
1194 }
1195
1196 void show_single_stop(struct session *se, struct record_cluster *rec)
1197 {
1198     session_leave(se);
1199 }
1200
1201 struct record_cluster **show_range_start(struct session *se,
1202                                          struct reclist_sortparms *sp, 
1203                                          int start, int *num, int *total, Odr_int *sumhits, Odr_int *approx_hits)
1204 {
1205     struct record_cluster **recs;
1206     struct reclist_sortparms *spp;
1207     int i;
1208 #if USE_TIMING    
1209     yaz_timing_t t = yaz_timing_create();
1210 #endif
1211     session_enter(se);
1212     recs = nmem_malloc(se->nmem, *num * sizeof(struct record_cluster *));
1213     if (!se->relevance)
1214     {
1215         *num = 0;
1216         *total = 0;
1217         *sumhits = 0;        
1218         *approx_hits = 0;
1219         recs = 0;
1220     }
1221     else
1222     {
1223         struct client_list *l;
1224         
1225         for (spp = sp; spp; spp = spp->next)
1226             if (spp->type == Metadata_sortkey_relevance)
1227             {
1228                 relevance_prepare_read(se->relevance, se->reclist);
1229                 break;
1230             }
1231         reclist_sort(se->reclist, sp);
1232         
1233         reclist_enter(se->reclist);
1234         *total = reclist_get_num_records(se->reclist);
1235
1236         *sumhits = 0;
1237         *approx_hits = 0;
1238         for (l = se->clients_active; l; l = l->next) {
1239             *sumhits += client_get_hits(l->client);
1240             *approx_hits += client_get_approximation(l->client);
1241         }
1242         for (i = 0; i < start; i++)
1243             if (!reclist_read_record(se->reclist))
1244             {
1245                 *num = 0;
1246                 recs = 0;
1247                 break;
1248             }
1249         
1250         for (i = 0; i < *num; i++)
1251         {
1252             struct record_cluster *r = reclist_read_record(se->reclist);
1253             if (!r)
1254             {
1255                 *num = i;
1256                 break;
1257             }
1258             recs[i] = r;
1259         }
1260         reclist_leave(se->reclist);
1261     }
1262 #if USE_TIMING
1263     yaz_timing_stop(t);
1264     yaz_log(YLOG_LOG, "show %6.5f %3.2f %3.2f", 
1265             yaz_timing_get_real(t), yaz_timing_get_user(t),
1266             yaz_timing_get_sys(t));
1267     yaz_timing_destroy(&t);
1268 #endif
1269     return recs;
1270 }
1271
1272 void show_range_stop(struct session *se, struct record_cluster **recs)
1273 {
1274     session_leave(se);
1275 }
1276
1277 void statistics(struct session *se, struct statistics *stat)
1278 {
1279     struct client_list *l;
1280     int count = 0;
1281
1282     memset(stat, 0, sizeof(*stat));
1283     stat->num_hits = 0;
1284     for (l = se->clients_active; l; l = l->next)
1285     {
1286         struct client *cl = l->client;
1287         if (!client_get_connection(cl))
1288             stat->num_no_connection++;
1289         stat->num_hits += client_get_hits(cl);
1290         switch (client_get_state(cl))
1291         {
1292         case Client_Connecting: stat->num_connecting++; break;
1293         case Client_Working: stat->num_working++; break;
1294         case Client_Idle: stat->num_idle++; break;
1295         case Client_Failed: stat->num_failed++; break;
1296         case Client_Error: stat->num_error++; break;
1297         default: break;
1298         }
1299         count++;
1300     }
1301     stat->num_records = se->total_records;
1302
1303     stat->num_clients = count;
1304 }
1305
1306 static struct record_metadata *record_metadata_init(
1307     NMEM nmem, const char *value, enum conf_metadata_type type,
1308     struct _xmlAttr *attr)
1309 {
1310     struct record_metadata *rec_md = record_metadata_create(nmem);
1311     struct record_metadata_attr **attrp = &rec_md->attributes;
1312     
1313     for (; attr; attr = attr->next)
1314     {
1315         if (attr->children && attr->children->content)
1316         {
1317             if (strcmp((const char *) attr->name, "type"))
1318             {  /* skip the "type" attribute.. Its value is already part of
1319                   the element in output (md-%s) and so repeating it here
1320                   is redundant */
1321                 *attrp = nmem_malloc(nmem, sizeof(**attrp));
1322                 (*attrp)->name =
1323                     nmem_strdup(nmem, (const char *) attr->name);
1324                 (*attrp)->value =
1325                     nmem_strdup(nmem, (const char *) attr->children->content);
1326                 attrp = &(*attrp)->next;
1327             }
1328         }
1329     }
1330     *attrp = 0;
1331
1332     if (type == Metadata_type_generic)
1333     {
1334         char *p = nmem_strdup(nmem, value);
1335
1336         p = normalize7bit_generic(p, " ,/.:([");
1337         
1338         rec_md->data.text.disp = p;
1339         rec_md->data.text.sort = 0;
1340     }
1341     else if (type == Metadata_type_year || type == Metadata_type_date)
1342     {
1343         int first, last;
1344         int longdate = 0;
1345
1346         if (type == Metadata_type_date)
1347             longdate = 1;
1348         if (extract7bit_dates((char *) value, &first, &last, longdate) < 0)
1349             return 0;
1350
1351         rec_md->data.number.min = first;
1352         rec_md->data.number.max = last;
1353     }
1354     else
1355         return 0;
1356     return rec_md;
1357 }
1358
1359 static int get_mergekey_from_doc(xmlDoc *doc, xmlNode *root, const char *name,
1360                                  struct conf_service *service, WRBUF norm_wr)
1361 {
1362     xmlNode *n;
1363     int no_found = 0;
1364     for (n = root->children; n; n = n->next)
1365     {
1366         if (n->type != XML_ELEMENT_NODE)
1367             continue;
1368         if (!strcmp((const char *) n->name, "metadata"))
1369         {
1370             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1371             if (type == NULL) {
1372                 yaz_log(YLOG_FATAL, "Missing type attribute on metadata element. Skipping!");
1373             }
1374             else if (!strcmp(name, (const char *) type))
1375             {
1376                 xmlChar *value = xmlNodeListGetString(doc, n->children, 1);
1377                 if (value)
1378                 {
1379                     const char *norm_str;
1380                     pp2_charset_token_t prt =
1381                         pp2_charset_token_create(service->charsets, "mergekey");
1382                     
1383                     pp2_charset_token_first(prt, (const char *) value, 0);
1384                     if (wrbuf_len(norm_wr) > 0)
1385                         wrbuf_puts(norm_wr, " ");
1386                     wrbuf_puts(norm_wr, name);
1387                     while ((norm_str =
1388                             pp2_charset_token_next(prt)))
1389                     {
1390                         if (*norm_str)
1391                         {
1392                             wrbuf_puts(norm_wr, " ");
1393                             wrbuf_puts(norm_wr, norm_str);
1394                         }
1395                     }
1396                     xmlFree(value);
1397                     pp2_charset_token_destroy(prt);
1398                     no_found++;
1399                 }
1400             }
1401             xmlFree(type);
1402         }
1403     }
1404     return no_found;
1405 }
1406
1407 static const char *get_mergekey(xmlDoc *doc, struct client *cl, int record_no,
1408                                 struct conf_service *service, NMEM nmem)
1409 {
1410     char *mergekey_norm = 0;
1411     xmlNode *root = xmlDocGetRootElement(doc);
1412     WRBUF norm_wr = wrbuf_alloc();
1413
1414     /* consider mergekey from XSL first */
1415     xmlChar *mergekey = xmlGetProp(root, (xmlChar *) "mergekey");
1416     if (mergekey)
1417     {
1418         const char *norm_str;
1419         pp2_charset_token_t prt =
1420             pp2_charset_token_create(service->charsets, "mergekey");
1421
1422         pp2_charset_token_first(prt, (const char *) mergekey, 0);
1423         while ((norm_str = pp2_charset_token_next(prt)))
1424         {
1425             if (*norm_str)
1426             {
1427                 if (wrbuf_len(norm_wr))
1428                     wrbuf_puts(norm_wr, " ");
1429                 wrbuf_puts(norm_wr, norm_str);
1430             }
1431         }
1432         pp2_charset_token_destroy(prt);
1433         xmlFree(mergekey);
1434     }
1435     else
1436     {
1437         /* no mergekey defined in XSL. Look for mergekey metadata instead */
1438         int field_id;
1439         for (field_id = 0; field_id < service->num_metadata; field_id++)
1440         {
1441             struct conf_metadata *ser_md = &service->metadata[field_id];
1442             if (ser_md->mergekey != Metadata_mergekey_no)
1443             {
1444                 int r = get_mergekey_from_doc(doc, root, ser_md->name,
1445                                               service, norm_wr);
1446                 if (r == 0 && ser_md->mergekey == Metadata_mergekey_required)
1447                 {
1448                     /* no mergekey on this one and it is required.. 
1449                        Generate unique key instead */
1450                     wrbuf_rewind(norm_wr);
1451                     break;
1452                 }
1453             }
1454         }
1455     }
1456
1457     /* generate unique key if none is not generated already or is empty */
1458     if (wrbuf_len(norm_wr) == 0)
1459     {
1460         wrbuf_printf(norm_wr, "position: %s-%d",
1461                      client_get_id(cl), record_no);
1462     }
1463     else
1464     {
1465         const char *lead = "content: ";
1466         wrbuf_insert(norm_wr, 0, lead, strlen(lead));
1467     }
1468     if (wrbuf_len(norm_wr) > 0)
1469         mergekey_norm = nmem_strdup(nmem, wrbuf_cstr(norm_wr));
1470     wrbuf_destroy(norm_wr);
1471     return mergekey_norm;
1472 }
1473
1474 /** \brief see if metadata for pz:recordfilter exists 
1475     \param root xml root element of normalized record
1476     \param sdb session database for client
1477     \retval 0 if there is no metadata for pz:recordfilter
1478     \retval 1 if there is metadata for pz:recordfilter
1479
1480     If there is no pz:recordfilter defined, this function returns 1
1481     as well.
1482 */
1483     
1484 static int check_record_filter(xmlNode *root, struct session_database *sdb)
1485 {
1486     int match = 0;
1487     xmlNode *n;
1488     const char *s;
1489     s = session_setting_oneval(sdb, PZ_RECORDFILTER);
1490
1491     if (!s || !*s)
1492         return 1;
1493
1494     for (n = root->children; n; n = n->next)
1495     {
1496         if (n->type != XML_ELEMENT_NODE)
1497             continue;
1498         if (!strcmp((const char *) n->name, "metadata"))
1499         {
1500             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1501             if (type)
1502             {
1503                 size_t len;
1504                 int substring;
1505                 const char *eq;
1506
1507                 if ((eq = strchr(s, '=')))
1508                     substring = 0;
1509                 else if ((eq = strchr(s, '~')))
1510                     substring = 1;
1511                 if (eq)
1512                     len = eq - s;
1513                 else
1514                     len = strlen(s);
1515                 if (len == strlen((const char *)type) &&
1516                     !memcmp((const char *) type, s, len))
1517                 {
1518                     xmlChar *value = xmlNodeGetContent(n);
1519                     if (value && *value)
1520                     {
1521                         if (!eq ||
1522                             (substring && strstr((const char *) value, eq+1)) ||
1523                             (!substring && !strcmp((const char *) value, eq + 1)))
1524                             match = 1;
1525                     }
1526                     xmlFree(value);
1527                 }
1528                 xmlFree(type);
1529             }
1530         }
1531     }
1532     return match;
1533 }
1534
1535
1536 static int ingest_to_cluster(struct client *cl,
1537                              xmlDoc *xdoc,
1538                              xmlNode *root,
1539                              int record_no,
1540                              const char *mergekey_norm);
1541
1542 /** \brief ingest XML record
1543     \param cl client holds the result set for record
1544     \param rec record buffer (0 terminated)
1545     \param record_no record position (1, 2, ..)
1546     \param nmem working NMEM
1547     \retval 0 OK
1548     \retval -1 failure
1549     \retval -2 Filtered
1550 */
1551 int ingest_record(struct client *cl, const char *rec,
1552                   int record_no, NMEM nmem)
1553 {
1554     struct session *se = client_get_session(cl);
1555     int ret = 0;
1556     struct session_database *sdb = client_get_database(cl);
1557     struct conf_service *service = se->service;
1558     xmlDoc *xdoc = normalize_record(se, sdb, service, rec, nmem);
1559     xmlNode *root;
1560     const char *mergekey_norm;
1561     
1562     if (!xdoc)
1563         return -1;
1564     
1565     root = xmlDocGetRootElement(xdoc);
1566     
1567     if (!check_record_filter(root, sdb))
1568     {
1569         session_log(se, YLOG_LOG, "Filtered out record no %d from %s", record_no, sdb->database->id);
1570         xmlFreeDoc(xdoc);
1571         return -2;
1572     }
1573     
1574     mergekey_norm = get_mergekey(xdoc, cl, record_no, service, nmem);
1575     if (!mergekey_norm)
1576     {
1577         session_log(se, YLOG_WARN, "Got no mergekey");
1578         xmlFreeDoc(xdoc);
1579         return -1;
1580     }
1581     session_enter(se);
1582     if (client_get_session(cl) == se)
1583         ret = ingest_to_cluster(cl, xdoc, root, record_no, mergekey_norm);
1584     session_leave(se);
1585     
1586     xmlFreeDoc(xdoc);
1587     return ret;
1588 }
1589
1590 // Skip record on non-zero
1591 static int check_limit_local(struct client *cl,
1592                              struct record *record,
1593                              int record_no)
1594 {
1595     int skip_record = 0;
1596     struct session *se = client_get_session(cl);
1597     struct conf_service *service = se->service;
1598     NMEM nmem_tmp = nmem_create();
1599     struct session_database *sdb = client_get_database(cl);
1600     int l = 0;
1601     while (!skip_record)
1602     {
1603         struct conf_metadata *ser_md = 0;
1604         struct record_metadata *rec_md = 0;
1605         int md_field_id;
1606         char **values = 0;
1607         int i, num_v = 0;
1608         
1609         const char *name =
1610             client_get_facet_limit_local(cl, sdb, &l, nmem_tmp, &num_v,
1611                                          &values);
1612         if (!name)
1613             break;
1614         
1615         md_field_id = conf_service_metadata_field_id(service, name);
1616         if (md_field_id < 0)
1617         {
1618             skip_record = 1;
1619             break;
1620         }
1621         ser_md = &service->metadata[md_field_id];
1622         rec_md = record->metadata[md_field_id];
1623         yaz_log(YLOG_LOG, "check limit local %s", name);
1624         for (i = 0; i < num_v; )
1625         {
1626             if (rec_md)
1627             {
1628                 if (ser_md->type == Metadata_type_year 
1629                     || ser_md->type == Metadata_type_date)
1630                 {
1631                     int y = atoi(values[i]);
1632                     if (y >= rec_md->data.number.min 
1633                         && y <= rec_md->data.number.max)
1634                         break;
1635                 }
1636                 else
1637                 {
1638                     yaz_log(YLOG_LOG, "cmp: '%s' '%s'", rec_md->data.text.disp, values[i]);
1639                     if (!strcmp(rec_md->data.text.disp, values[i]))
1640                     {
1641                         // Value equals, should not be filtered.
1642                         break;
1643                     }
1644                 }
1645                 rec_md = rec_md->next;
1646             }
1647             else
1648             {
1649                 rec_md = record->metadata[md_field_id];
1650                 i++;
1651             }
1652         }
1653         // At end , not match
1654         if (i == num_v)
1655         {
1656             skip_record = 1;
1657             break;
1658         }
1659     }
1660     nmem_destroy(nmem_tmp);
1661     return skip_record;
1662 }
1663                              
1664 static int ingest_to_cluster(struct client *cl,
1665                              xmlDoc *xdoc,
1666                              xmlNode *root,
1667                              int record_no,
1668                              const char *mergekey_norm)
1669 {
1670     xmlNode *n;
1671     xmlChar *type = 0;
1672     xmlChar *value = 0;
1673     struct session *se = client_get_session(cl);
1674     struct conf_service *service = se->service;
1675     int term_factor = 1;
1676     struct record_cluster *cluster;
1677     struct session_database *sdb = client_get_database(cl);
1678     struct record *record = record_create(se->nmem, 
1679                                           service->num_metadata,
1680                                           service->num_sortkeys, cl,
1681                                           record_no);
1682
1683     for (n = root->children; n; n = n->next)
1684     {
1685         if (type)
1686             xmlFree(type);
1687         if (value)
1688             xmlFree(value);
1689         type = value = 0;
1690         
1691         if (n->type != XML_ELEMENT_NODE)
1692             continue;
1693         if (!strcmp((const char *) n->name, "metadata"))
1694         {
1695             struct conf_metadata *ser_md = 0;
1696             struct record_metadata **wheretoput = 0;
1697             struct record_metadata *rec_md = 0;
1698             int md_field_id = -1;
1699             
1700             type = xmlGetProp(n, (xmlChar *) "type");
1701             value = xmlNodeListGetString(xdoc, n->children, 1);
1702             
1703             if (!type || !value || !*value)
1704                 continue;
1705             
1706             md_field_id 
1707                 = conf_service_metadata_field_id(service, (const char *) type);
1708             if (md_field_id < 0)
1709             {
1710                 if (se->number_of_warnings_unknown_metadata == 0)
1711                 {
1712                     session_log(se, YLOG_WARN, 
1713                             "Ignoring unknown metadata element: %s", type);
1714                 }
1715                 se->number_of_warnings_unknown_metadata++;
1716                 continue;
1717             }
1718            
1719             ser_md = &service->metadata[md_field_id];
1720
1721             // non-merged metadata
1722             rec_md = record_metadata_init(se->nmem, (const char *) value,
1723                                           ser_md->type, n->properties);
1724             if (!rec_md)
1725             {
1726                 session_log(se, YLOG_WARN, "bad metadata data '%s' "
1727                             "for element '%s'", value, type);
1728                 continue;
1729             }
1730             wheretoput = &record->metadata[md_field_id];
1731             while (*wheretoput)
1732                 wheretoput = &(*wheretoput)->next;
1733             *wheretoput = rec_md;
1734         }
1735     }
1736
1737     if (check_limit_local(cl, record, record_no))
1738     {
1739         session_log(se, YLOG_LOG, "Facet filtered out record no %d from %s",
1740                     record_no, sdb->database->id);
1741         if (type)
1742             xmlFree(type);
1743         if (value)
1744             xmlFree(value);
1745         return -2;
1746     }
1747     cluster = reclist_insert(se->reclist, service, record,
1748                              mergekey_norm, &se->total_merged);
1749     if (!cluster)
1750         return -1;
1751
1752     {
1753         const char *use_term_factor_str =
1754             session_setting_oneval(sdb, PZ_TERMLIST_TERM_FACTOR);
1755         if (use_term_factor_str && use_term_factor_str[0] == '1')
1756         {
1757             int maxrecs = client_get_maxrecs(cl);
1758             int hits = (int) client_get_hits(cl);
1759             term_factor = MAX(hits, maxrecs) /  MAX(1, maxrecs);
1760             assert(term_factor >= 1);
1761             yaz_log(YLOG_DEBUG, "Using term factor: %d (%d / %d)", term_factor, MAX(hits, maxrecs), MAX(1, maxrecs));
1762         }
1763     }
1764
1765     if (global_parameters.dump_records)
1766         session_log(se, YLOG_LOG, "Cluster id %s from %s (#%d)", cluster->recid,
1767                     sdb->database->id, record_no);
1768
1769
1770     relevance_newrec(se->relevance, cluster);
1771     
1772     // now parsing XML record and adding data to cluster or record metadata
1773     for (n = root->children; n; n = n->next)
1774     {
1775         pp2_charset_token_t prt;
1776         if (type)
1777             xmlFree(type);
1778         if (value)
1779             xmlFree(value);
1780         type = value = 0;
1781         
1782         if (n->type != XML_ELEMENT_NODE)
1783             continue;
1784         if (!strcmp((const char *) n->name, "metadata"))
1785         {
1786             struct conf_metadata *ser_md = 0;
1787             struct conf_sortkey *ser_sk = 0;
1788             struct record_metadata **wheretoput = 0;
1789             struct record_metadata *rec_md = 0;
1790             int md_field_id = -1;
1791             int sk_field_id = -1;
1792             
1793             type = xmlGetProp(n, (xmlChar *) "type");
1794             value = xmlNodeListGetString(xdoc, n->children, 1);
1795             
1796             if (!type || !value || !*value)
1797                 continue;
1798             
1799             md_field_id 
1800                 = conf_service_metadata_field_id(service, (const char *) type);
1801             if (md_field_id < 0)
1802                 continue;
1803             
1804             ser_md = &service->metadata[md_field_id];
1805             
1806             if (ser_md->sortkey_offset >= 0)
1807             {
1808                 sk_field_id = ser_md->sortkey_offset;
1809                 ser_sk = &service->sortkeys[sk_field_id];
1810             }
1811
1812             // merged metadata
1813             rec_md = record_metadata_init(se->nmem, (const char *) value,
1814                                           ser_md->type, 0);
1815             if (!rec_md)
1816                 continue;
1817
1818             wheretoput = &cluster->metadata[md_field_id];
1819
1820             // and polulate with data:
1821             // assign cluster or record based on merge action
1822             if (ser_md->merge == Metadata_merge_unique)
1823             {
1824                 while (*wheretoput)
1825                 {
1826                     if (!strcmp((const char *) (*wheretoput)->data.text.disp, 
1827                                 rec_md->data.text.disp))
1828                         break;
1829                     wheretoput = &(*wheretoput)->next;
1830                 }
1831                 if (!*wheretoput)
1832                     *wheretoput = rec_md;
1833             }
1834             else if (ser_md->merge == Metadata_merge_longest)
1835             {
1836                 if (!*wheretoput 
1837                     || strlen(rec_md->data.text.disp) 
1838                     > strlen((*wheretoput)->data.text.disp))
1839                 {
1840                     *wheretoput = rec_md;
1841                     if (ser_sk)
1842                     {
1843                         const char *sort_str = 0;
1844                         int skip_article = 
1845                             ser_sk->type == Metadata_sortkey_skiparticle;
1846
1847                         if (!cluster->sortkeys[sk_field_id])
1848                             cluster->sortkeys[sk_field_id] = 
1849                                 nmem_malloc(se->nmem, 
1850                                             sizeof(union data_types));
1851                          
1852                         prt =
1853                             pp2_charset_token_create(service->charsets, "sort");
1854
1855                         pp2_charset_token_first(prt, rec_md->data.text.disp,
1856                                                 skip_article);
1857
1858                         pp2_charset_token_next(prt);
1859                          
1860                         sort_str = pp2_get_sort(prt);
1861                          
1862                         cluster->sortkeys[sk_field_id]->text.disp = 
1863                             rec_md->data.text.disp;
1864                         if (!sort_str)
1865                         {
1866                             sort_str = rec_md->data.text.disp;
1867                             session_log(se, YLOG_WARN, 
1868                                     "Could not make sortkey. Bug #1858");
1869                         }
1870                         cluster->sortkeys[sk_field_id]->text.sort = 
1871                             nmem_strdup(se->nmem, sort_str);
1872                         pp2_charset_token_destroy(prt);
1873                     }
1874                 }
1875             }
1876             else if (ser_md->merge == Metadata_merge_all)
1877             {
1878                 while (*wheretoput)
1879                     wheretoput = &(*wheretoput)->next;
1880                 *wheretoput = rec_md;
1881             }
1882             else if (ser_md->merge == Metadata_merge_range)
1883             {
1884                 if (!*wheretoput)
1885                 {
1886                     *wheretoput = rec_md;
1887                     if (ser_sk)
1888                         cluster->sortkeys[sk_field_id] 
1889                             = &rec_md->data;
1890                 }
1891                 else
1892                 {
1893                     int this_min = rec_md->data.number.min;
1894                     int this_max = rec_md->data.number.max;
1895                     if (this_min < (*wheretoput)->data.number.min)
1896                         (*wheretoput)->data.number.min = this_min;
1897                     if (this_max > (*wheretoput)->data.number.max)
1898                         (*wheretoput)->data.number.max = this_max;
1899                 }
1900             }
1901
1902
1903             // ranking of _all_ fields enabled ... 
1904             if (ser_md->rank)
1905                 relevance_countwords(se->relevance, cluster, 
1906                                      (char *) value, ser_md->rank,
1907                                      ser_md->name);
1908
1909             // construct facets ... unless the client already has reported them
1910             if (ser_md->termlist && !client_has_facet(cl, (char *) type))
1911             {
1912                 if (ser_md->type == Metadata_type_year)
1913                 {
1914                     char year[64];
1915                     sprintf(year, "%d", rec_md->data.number.max);
1916
1917                     add_facet(se, (char *) type, year, term_factor);
1918                     if (rec_md->data.number.max != rec_md->data.number.min)
1919                     {
1920                         sprintf(year, "%d", rec_md->data.number.min);
1921                         add_facet(se, (char *) type, year, term_factor);
1922                     }
1923                 }
1924                 else
1925                     add_facet(se, (char *) type, (char *) value, term_factor);
1926             }
1927
1928             // cleaning up
1929             xmlFree(type);
1930             xmlFree(value);
1931             type = value = 0;
1932         }
1933         else
1934         {
1935             if (se->number_of_warnings_unknown_elements == 0)
1936                 session_log(se, YLOG_WARN,
1937                         "Unexpected element in internal record: %s", n->name);
1938             se->number_of_warnings_unknown_elements++;
1939         }
1940     }
1941     if (type)
1942         xmlFree(type);
1943     if (value)
1944         xmlFree(value);
1945
1946     relevance_donerecord(se->relevance, cluster);
1947     se->total_records++;
1948
1949     return 0;
1950 }
1951
1952 void session_log(struct session *s, int level, const char *fmt, ...)
1953 {
1954     char buf[1024];
1955     va_list ap;
1956     va_start(ap, fmt);
1957
1958     yaz_vsnprintf(buf, sizeof(buf)-30, fmt, ap);
1959     yaz_log(level, "Session %u: %s", s->session_id, buf);
1960
1961     va_end(ap);
1962 }
1963
1964 /*
1965  * Local variables:
1966  * c-basic-offset: 4
1967  * c-file-style: "Stroustrup"
1968  * indent-tabs-mode: nil
1969  * End:
1970  * vim: shiftwidth=4 tabstop=8 expandtab
1971  */
1972