425d6ca740473bac2558042a9dccb70a0584f1f2
[pazpar2-moved-to-github.git] / src / session.c
1 /* This file is part of Pazpar2.
2    Copyright (C) 2006-2011 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 #define TERMLIST_HIGH_SCORE 25
80
81 #define MAX_CHUNK 15
82
83 #define MAX(a,b) ((a)>(b)?(a):(b))
84
85 // Note: Some things in this structure will eventually move to configuration
86 struct parameters global_parameters = 
87 {
88     0,   // dump_records
89     0,   // debug_mode
90     0,   // predictable sessions
91 };
92
93 struct client_list {
94     struct client *client;
95     struct client_list *next;
96 };
97
98 /* session counting (1) , disable client counting (0) */
99 static YAZ_MUTEX g_session_mutex = 0;
100 static int no_sessions = 0;
101 static int no_session_total = 0;
102
103 static int session_use(int delta)
104 {
105     int sessions;
106     if (!g_session_mutex)
107         yaz_mutex_create(&g_session_mutex);
108     yaz_mutex_enter(g_session_mutex);
109     no_sessions += delta;
110     if (delta > 0)
111         no_session_total += delta;
112     sessions = no_sessions;
113     yaz_mutex_leave(g_session_mutex);
114     yaz_log(YLOG_DEBUG, "%s sessions=%d", delta == 0 ? "" : (delta > 0 ? "INC" : "DEC"), no_sessions);
115     return sessions;
116 }
117
118 int sessions_count(void)
119 {
120     return session_use(0);
121 }
122
123 int session_count_total(void)
124 {
125     int total = 0;
126     if (!g_session_mutex)
127         return 0;
128     yaz_mutex_enter(g_session_mutex);
129     total = no_session_total;
130     yaz_mutex_leave(g_session_mutex);
131     return total;
132 }
133
134 static void log_xml_doc(xmlDoc *doc)
135 {
136     FILE *lf = yaz_log_file();
137     xmlChar *result = 0;
138     int len = 0;
139 #if LIBXML_VERSION >= 20600
140     xmlDocDumpFormatMemory(doc, &result, &len, 1);
141 #else
142     xmlDocDumpMemory(doc, &result, &len);
143 #endif
144     if (lf && len)
145     {
146         (void) fwrite(result, 1, len, lf);
147         fprintf(lf, "\n");
148     }
149     xmlFree(result);
150 }
151
152 static void session_enter(struct session *s)
153 {
154     yaz_mutex_enter(s->session_mutex);
155 }
156
157 static void session_leave(struct session *s)
158 {
159     yaz_mutex_leave(s->session_mutex);
160 }
161
162 static void session_normalize_facet(struct session *s, const char *type,
163                                     const char *value,
164                                     WRBUF display_wrbuf,
165                                     WRBUF facet_wrbuf)
166 {
167     struct conf_service *service = s->service;
168     pp2_charset_token_t prt;
169     const char *facet_component;
170     int i;
171     const char *icu_chain_id = 0;
172
173     for (i = 0; i < service->num_metadata; i++)
174         if (!strcmp((service->metadata + i)->name, type))
175             icu_chain_id = (service->metadata + i)->facetrule;
176     if (!icu_chain_id)
177         icu_chain_id = "facet";
178     prt = pp2_charset_token_create(service->charsets, icu_chain_id);
179     if (!prt)
180     {
181         yaz_log(YLOG_FATAL, "Unknown ICU chain '%s' for facet of type '%s'",
182                 icu_chain_id, type);
183         wrbuf_destroy(facet_wrbuf);
184         wrbuf_destroy(display_wrbuf);
185         return;
186     }
187     pp2_charset_token_first(prt, value, 0);
188     while ((facet_component = pp2_charset_token_next(prt)))
189     {
190         const char *display_component;
191         if (*facet_component)
192         {
193             if (wrbuf_len(facet_wrbuf))
194                 wrbuf_puts(facet_wrbuf, " ");
195             wrbuf_puts(facet_wrbuf, facet_component);
196         }
197         display_component = pp2_get_display(prt);
198         if (display_component)
199         {
200             if (wrbuf_len(display_wrbuf))
201                 wrbuf_puts(display_wrbuf, " ");
202             wrbuf_puts(display_wrbuf, display_component);
203         }
204     }
205     pp2_charset_token_destroy(prt);
206 }
207
208 void add_facet(struct session *s, const char *type, const char *value, int count)
209 {
210     WRBUF facet_wrbuf = wrbuf_alloc();
211     WRBUF display_wrbuf = wrbuf_alloc();
212
213     session_normalize_facet(s, type, value, display_wrbuf, facet_wrbuf);
214  
215     if (wrbuf_len(facet_wrbuf))
216     {
217         int i;
218         for (i = 0; i < s->num_termlists; i++)
219             if (!strcmp(s->termlists[i].name, type))
220                 break;
221         if (i == s->num_termlists)
222         {
223             if (i == SESSION_MAX_TERMLISTS)
224             {
225                 session_log(s, YLOG_FATAL, "Too many termlists");
226                 wrbuf_destroy(facet_wrbuf);
227                 wrbuf_destroy(display_wrbuf);
228                 return;
229             }
230             
231             s->termlists[i].name = nmem_strdup(s->nmem, type);
232             s->termlists[i].termlist 
233                 = termlist_create(s->nmem, TERMLIST_HIGH_SCORE);
234             s->num_termlists = i + 1;
235         }
236         
237 #if 0
238         session_log(s, YLOG_DEBUG, "Facets for %s: %s norm:%s (%d)", type, value, wrbuf_cstr(facet_wrbuf), count);
239 #endif
240         termlist_insert(s->termlists[i].termlist, wrbuf_cstr(display_wrbuf),
241                         wrbuf_cstr(facet_wrbuf), count);
242     }
243     wrbuf_destroy(facet_wrbuf);
244     wrbuf_destroy(display_wrbuf);
245 }
246
247 static xmlDoc *record_to_xml(struct session *se,
248                              struct session_database *sdb, const char *rec)
249 {
250     struct database *db = sdb->database;
251     xmlDoc *rdoc = 0;
252
253     rdoc = xmlParseMemory(rec, strlen(rec));
254
255     if (!rdoc)
256     {
257         session_log(se, YLOG_FATAL, "Non-wellformed XML received from %s",
258                     db->id);
259         return 0;
260     }
261
262     if (global_parameters.dump_records)
263     {
264         session_log(se, YLOG_LOG, "Un-normalized record from %s", db->id);
265         log_xml_doc(rdoc);
266     }
267
268     return rdoc;
269 }
270
271 #define MAX_XSLT_ARGS 16
272
273 // Add static values from session database settings if applicable
274 static void insert_settings_parameters(struct session_database *sdb,
275                                        struct conf_service *service,
276                                        char **parms,
277                                        NMEM nmem)
278 {
279     int i;
280     int nparms = 0;
281     int offset = 0;
282
283     for (i = 0; i < service->num_metadata; i++)
284     {
285         struct conf_metadata *md = &service->metadata[i];
286         int setting;
287
288         if (md->setting == Metadata_setting_parameter &&
289             (setting = settings_lookup_offset(service, md->name)) >= 0)
290         {
291             const char *val = session_setting_oneval(sdb, setting);
292             if (val && nparms < MAX_XSLT_ARGS)
293             {
294                 char *buf;
295                 int len = strlen(val);
296                 buf = nmem_malloc(nmem, len + 3);
297                 buf[0] = '\'';
298                 strcpy(buf + 1, val);
299                 buf[len+1] = '\'';
300                 buf[len+2] = '\0';
301                 parms[offset++] = md->name;
302                 parms[offset++] = buf;
303                 nparms++;
304             }
305         }
306     }
307     parms[offset] = 0;
308 }
309
310 // Add static values from session database settings if applicable
311 static void insert_settings_values(struct session_database *sdb, xmlDoc *doc,
312     struct conf_service *service)
313 {
314     int i;
315
316     for (i = 0; i < service->num_metadata; i++)
317     {
318         struct conf_metadata *md = &service->metadata[i];
319         int offset;
320
321         if (md->setting == Metadata_setting_postproc &&
322             (offset = settings_lookup_offset(service, md->name)) >= 0)
323         {
324             const char *val = session_setting_oneval(sdb, offset);
325             if (val)
326             {
327                 xmlNode *r = xmlDocGetRootElement(doc);
328                 xmlNode *n = xmlNewTextChild(r, 0, (xmlChar *) "metadata",
329                                              (xmlChar *) val);
330                 xmlSetProp(n, (xmlChar *) "type", (xmlChar *) md->name);
331             }
332         }
333     }
334 }
335
336 static xmlDoc *normalize_record(struct session *se,
337                                 struct session_database *sdb,
338                                 struct conf_service *service,
339                                 const char *rec, NMEM nmem)
340 {
341     xmlDoc *rdoc = record_to_xml(se, sdb, rec);
342
343     if (rdoc)
344     {
345         char *parms[MAX_XSLT_ARGS*2+1];
346         
347         insert_settings_parameters(sdb, service, parms, nmem);
348         
349         if (normalize_record_transform(sdb->map, &rdoc, (const char **)parms))
350         {
351             session_log(se, YLOG_WARN, "Normalize failed from %s",
352                         sdb->database->id);
353         }
354         else
355         {
356             insert_settings_values(sdb, rdoc, service);
357             
358             if (global_parameters.dump_records)
359             {
360                 session_log(se, YLOG_LOG, "Normalized record from %s", 
361                             sdb->database->id);
362                 log_xml_doc(rdoc);
363             }
364         }
365     }
366     return rdoc;
367 }
368
369 void session_settings_dump(struct session *se,
370                            struct session_database *db,
371                            WRBUF w)
372 {
373     if (db->settings)
374     {
375         int i, num = db->num_settings;
376         for (i = 0; i < num; i++)
377         {
378             struct setting *s = db->settings[i];
379             for (;s ; s = s->next)
380             {
381                 wrbuf_puts(w, "<set name=\"");
382                 wrbuf_xmlputs(w, s->name);
383                 wrbuf_puts(w, "\" value=\"");
384                 wrbuf_xmlputs(w, s->value);
385                 wrbuf_puts(w, "\"/>");
386             }
387             if (db->settings[i])
388                 wrbuf_puts(w, "\n");
389         }
390     }
391 }
392
393 // Retrieve first defined value for 'name' for given database.
394 // Will be extended to take into account user associated with session
395 const char *session_setting_oneval(struct session_database *db, int offset)
396 {
397     if (offset >= db->num_settings || !db->settings[offset])
398         return "";
399     return db->settings[offset]->value;
400 }
401
402 // Prepare XSLT stylesheets for record normalization
403 // Structures are allocated on the session_wide nmem to avoid having
404 // to recompute this for every search. This would lead
405 // to leaking if a single session was to repeatedly change the PZ_XSLT
406 // setting. However, this is not a realistic use scenario.
407 static int prepare_map(struct session *se, struct session_database *sdb)
408 {
409     const char *s;
410
411     if (sdb->settings && sdb->settings[PZ_XSLT] && !sdb->map &&
412         (s = session_setting_oneval(sdb, PZ_XSLT)))        
413     {
414         char auto_stylesheet[256];
415
416         if (!strcmp(s, "auto"))
417         {
418             const char *request_syntax = session_setting_oneval(
419                 sdb, PZ_REQUESTSYNTAX);
420             if (request_syntax)
421             {
422                 char *cp;
423                 yaz_snprintf(auto_stylesheet, sizeof(auto_stylesheet),
424                              "%s.xsl", request_syntax);
425                 for (cp = auto_stylesheet; *cp; cp++)
426                 {
427                     /* deliberately only consider ASCII */
428                     if (*cp > 32 && *cp < 127)
429                         *cp = tolower(*cp);
430                 }
431                 s = auto_stylesheet;
432             }
433             else
434             {
435                 session_log(se, YLOG_WARN,
436                             "No pz:requestsyntax for auto stylesheet");
437             }
438         }
439         sdb->map = normalize_cache_get(se->normalize_cache,
440                                        se->service, s);
441         if (!sdb->map)
442             return -1;
443     }
444     return 0;
445 }
446
447 // called if watch should be removed because http_channel is to be destroyed
448 static void session_watch_cancel(void *data, struct http_channel *c,
449                                  void *data2)
450 {
451     struct session_watchentry *ent = data;
452
453     ent->fun = 0;
454     ent->data = 0;
455     ent->obs = 0;
456 }
457
458 // set watch. Returns 0=OK, -1 if watch is already set
459 int session_set_watch(struct session *s, int what, 
460                       session_watchfun fun, void *data,
461                       struct http_channel *chan)
462 {
463     int ret;
464     session_enter(s);
465     if (s->watchlist[what].fun)
466         ret = -1;
467     else
468     {
469         
470         s->watchlist[what].fun = fun;
471         s->watchlist[what].data = data;
472         s->watchlist[what].obs = http_add_observer(chan, &s->watchlist[what],
473                                                    session_watch_cancel);
474         ret = 0;
475     }
476     session_leave(s);
477     return ret;
478 }
479
480 void session_alert_watch(struct session *s, int what)
481 {
482     assert(s);
483     session_enter(s);
484     if (s->watchlist[what].fun)
485     {
486         /* our watch is no longer associated with http_channel */
487         void *data;
488         session_watchfun fun;
489
490         http_remove_observer(s->watchlist[what].obs);
491         fun  = s->watchlist[what].fun;
492         data = s->watchlist[what].data;
493
494         /* reset watch before fun is invoked - in case fun wants to set
495            it again */
496         s->watchlist[what].fun = 0;
497         s->watchlist[what].data = 0;
498         s->watchlist[what].obs = 0;
499
500         session_leave(s);
501         session_log(s, YLOG_DEBUG,
502                     "Alert Watch: %d calling function: %p", what, fun);
503         fun(data);
504     }
505     else
506         session_leave(s);
507 }
508
509 //callback for grep_databases
510 static void select_targets_callback(struct session *se,
511                                     struct session_database *db)
512 {
513     struct client *cl;
514     struct client_list *l;
515
516     for (l = se->clients_cached; l; l = l->next)
517         if (client_get_database(l->client) == db)
518             break;
519
520     if (l)
521         cl = l->client;
522     else
523     {
524         cl = client_create(db->database->id);
525         client_set_database(cl, db);
526         client_set_session(cl, se);
527
528         l = xmalloc(sizeof(*l));
529         l->client = cl;
530         l->next = se->clients_cached;
531         se->clients_cached = l;
532     }
533
534     l = xmalloc(sizeof(*l));
535     l->client = cl;
536     l->next = se->clients_active;
537     se->clients_active = l;
538 }
539
540 static void session_reset_active_clients(struct session *se,
541                                          struct client_list *new_list)
542 {
543     struct client_list *l;
544
545     session_enter(se);
546     l = se->clients_active;
547     se->clients_active = new_list;
548     session_leave(se);
549
550     while (l)
551     {
552         struct client_list *l_next = l->next;
553         xfree(l);
554         l = l_next;
555     }
556 }
557
558 static void session_remove_cached_clients(struct session *se)
559 {
560     struct client_list *l;
561
562     session_reset_active_clients(se, 0);
563
564     session_enter(se);
565     l = se->clients_cached;
566     se->clients_cached = 0;
567     session_leave(se);
568
569     while (l)
570     {
571         struct client_list *l_next = l->next;
572         client_lock(l->client);
573         client_set_session(l->client, 0);
574         client_set_database(l->client, 0);
575         client_unlock(l->client);
576         client_destroy(l->client);
577         xfree(l);
578         l = l_next;
579     }
580 }
581
582 // Associates a set of clients with a session;
583 // Note: Session-databases represent databases with per-session 
584 // setting overrides
585 static int select_targets(struct session *se, const char *filter)
586 {
587     return session_grep_databases(se, filter, select_targets_callback);
588 }
589
590 int session_active_clients(struct session *s)
591 {
592     struct client_list *l;
593     int res = 0;
594
595     for (l = s->clients_active; l; l = l->next)
596         if (client_is_active(l->client))
597             res++;
598
599     return res;
600 }
601
602 int session_is_preferred_clients_ready(struct session *s)
603 {
604     struct client_list *l;
605     int res = 0;
606
607     for (l = s->clients_active; l; l = l->next)
608         if (client_is_active_preferred(l->client))
609             res++;
610     session_log(s, YLOG_DEBUG, "Has %d active preferred clients.", res);
611     return res == 0;
612 }
613
614 void session_sort(struct session *se, const char *field, int increasing)
615 {
616     struct session_sorted_results *sr;
617     struct client_list *l;
618
619     session_enter(se);
620
621     /* see if we already have sorted for this critieria */
622     for (sr = se->sorted_results; sr; sr = sr->next)
623     {
624         if (!strcmp(field, sr->field) && increasing == sr->increasing)
625             break;
626     }
627     if (sr)
628     {
629         yaz_log(YLOG_LOG, "search_sort: field=%s increasing=%d already fetched",
630                 field, increasing);
631         session_leave(se);
632         return;
633     }
634     yaz_log(YLOG_LOG, "search_sort: field=%s increasing=%d must fetch",
635             field, increasing);
636     sr = nmem_malloc(se->nmem, sizeof(*sr));
637     sr->field = nmem_strdup(se->nmem, field);
638     sr->increasing = increasing;
639     sr->next = se->sorted_results;
640     se->sorted_results = sr;
641     
642     for (l = se->clients_active; l; l = l->next)
643     {
644         struct client *cl = l->client;
645         struct timeval tval;
646         if (client_prep_connection(cl, se->service->z3950_operation_timeout,
647                                    se->service->z3950_session_timeout,
648                                    se->service->server->iochan_man,
649                                    &tval))
650             client_start_search(cl);
651     }
652     session_leave(se);
653 }
654
655 enum pazpar2_error_code session_search(struct session *se,
656                                        const char *query,
657                                        const char *startrecs,
658                                        const char *maxrecs,
659                                        const char *filter,
660                                        const char *limit,
661                                        const char **addinfo,
662                                        const char *sort_field, int increasing)
663 {
664     int live_channels = 0;
665     int no_working = 0;
666     int no_failed_query = 0;
667     int no_failed_limit = 0;
668     struct client_list *l, *l0;
669     struct timeval tval;
670     facet_limits_t facet_limits;
671
672     session_log(se, YLOG_DEBUG, "Search");
673
674     *addinfo = 0;
675
676     if (se->settings_modified)
677         session_remove_cached_clients(se);
678     else
679         session_reset_active_clients(se, 0);
680     
681     session_enter(se);
682     reclist_destroy(se->reclist);
683     se->reclist = 0;
684     se->settings_modified = 0;
685     relevance_destroy(&se->relevance);
686     nmem_reset(se->nmem);
687     se->total_records = se->total_merged = 0;
688     se->num_termlists = 0;
689
690     /* reset list of sorted results and clear to relevance search */
691     se->sorted_results = nmem_malloc(se->nmem, sizeof(*se->sorted_results));
692     se->sorted_results->field = nmem_strdup(se->nmem, sort_field);
693     se->sorted_results->increasing = increasing;
694     se->sorted_results->next = 0;
695     
696     live_channels = select_targets(se, filter);
697     if (!live_channels)
698     {
699         session_leave(se);
700         return PAZPAR2_NO_TARGETS;
701     }
702     se->reclist = reclist_create(se->nmem);
703
704     yaz_gettimeofday(&tval);
705     
706     tval.tv_sec += 5;
707
708     facet_limits = facet_limits_create(limit);
709     if (!facet_limits)
710     {
711         *addinfo = "limit";
712         session_leave(se);
713         return PAZPAR2_MALFORMED_PARAMETER_VALUE;
714     }
715
716     l0 = se->clients_active;
717     se->clients_active = 0;
718     session_leave(se);
719
720     for (l = l0; l; l = l->next)
721     {
722         int parse_ret;
723         struct client *cl = l->client;
724
725         if (prepare_map(se, client_get_database(cl)) < 0)
726             continue;
727
728         parse_ret = client_parse_query(cl, query, facet_limits, startrecs,
729             maxrecs);
730         if (parse_ret == -1)
731             no_failed_query++;
732         else if (parse_ret == -2)
733             no_failed_limit++;
734         else
735         {
736             int r =
737                 client_prep_connection(cl, se->service->z3950_operation_timeout,
738                                        se->service->z3950_session_timeout,
739                                        se->service->server->iochan_man,
740                                        &tval);
741             if (parse_ret == 1 && r == 2)
742             {
743                 session_log(se, YLOG_LOG, "client REUSE %s", client_get_id(cl));
744                 client_reingest(cl);
745             }
746             else
747             {
748                 session_log(se, YLOG_LOG, "client NEW %s", client_get_id(cl));
749                 client_start_search(cl);
750             }
751             no_working++;
752         }
753     }
754     facet_limits_destroy(facet_limits);
755     session_reset_active_clients(se, l0);
756
757     if (no_working == 0)
758     {
759         if (no_failed_query > 0)
760         {
761             *addinfo = "query";
762             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
763         }
764         else if (no_failed_limit > 0)
765         {
766             *addinfo = "limit";
767             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
768         }
769         else
770             return PAZPAR2_NO_TARGETS;
771     }
772     yaz_log(YLOG_LOG, "session_start_search done");
773     return PAZPAR2_NO_ERROR;
774 }
775
776 // Creates a new session_database object for a database
777 static void session_init_databases_fun(void *context, struct database *db)
778 {
779     struct session *se = (struct session *) context;
780     struct session_database *new = nmem_malloc(se->session_nmem, sizeof(*new));
781     int i;
782
783     new->database = db;
784     
785     new->map = 0;
786     assert(db->settings);
787     new->settings = nmem_malloc(se->session_nmem,
788                                 sizeof(struct settings *) * db->num_settings);
789     new->num_settings = db->num_settings;
790     for (i = 0; i < db->num_settings; i++)
791     {
792         struct setting *setting = db->settings[i];
793         new->settings[i] = setting;
794     }
795     new->next = se->databases;
796     se->databases = new;
797 }
798
799 // Doesn't free memory associated with sdb -- nmem takes care of that
800 static void session_database_destroy(struct session_database *sdb)
801 {
802     sdb->map = 0;
803 }
804
805 // Initialize session_database list -- this represents this session's view
806 // of the database list -- subject to modification by the settings ws command
807 void session_init_databases(struct session *se)
808 {
809     se->databases = 0;
810     predef_grep_databases(se, se->service, session_init_databases_fun);
811 }
812
813 // Probably session_init_databases_fun should be refactored instead of
814 // called here.
815 static struct session_database *load_session_database(struct session *se, 
816                                                       char *id)
817 {
818     struct database *db = new_database(id, se->session_nmem);
819
820     session_init_databases_fun((void*) se, db);
821
822     // New sdb is head of se->databases list
823     return se->databases;
824 }
825
826 // Find an existing session database. If not found, load it
827 static struct session_database *find_session_database(struct session *se, 
828                                                       char *id)
829 {
830     struct session_database *sdb;
831
832     for (sdb = se->databases; sdb; sdb = sdb->next)
833         if (!strcmp(sdb->database->id, id))
834             return sdb;
835     return load_session_database(se, id);
836 }
837
838 // Apply a session override to a database
839 void session_apply_setting(struct session *se, char *dbname, char *setting,
840                            char *value)
841 {
842     struct session_database *sdb = find_session_database(se, dbname);
843     struct conf_service *service = se->service;
844     struct setting *new = nmem_malloc(se->session_nmem, sizeof(*new));
845     int offset = settings_create_offset(service, setting);
846
847     expand_settings_array(&sdb->settings, &sdb->num_settings, offset,
848                           se->session_nmem);
849     new->precedence = 0;
850     new->target = dbname;
851     new->name = setting;
852     new->value = value;
853     new->next = sdb->settings[offset];
854     sdb->settings[offset] = new;
855
856     se->settings_modified = 1;
857
858     // Force later recompute of settings-driven data structures
859     // (happens when a search starts and client connections are prepared)
860     switch (offset)
861     {
862     case PZ_XSLT:
863         if (sdb->map)
864         {
865             sdb->map = 0;
866         }
867         break;
868     }
869 }
870
871 void session_destroy(struct session *se)
872 {
873     struct session_database *sdb;
874     session_log(se, YLOG_DEBUG, "Destroying");
875     session_use(-1);
876     session_remove_cached_clients(se);
877
878     for (sdb = se->databases; sdb; sdb = sdb->next)
879         session_database_destroy(sdb);
880     normalize_cache_destroy(se->normalize_cache);
881     relevance_destroy(&se->relevance);
882     reclist_destroy(se->reclist);
883     nmem_destroy(se->nmem);
884     service_destroy(se->service);
885     yaz_mutex_destroy(&se->session_mutex);
886 }
887
888 size_t session_get_memory_status(struct session *session) {
889     size_t session_nmem;
890     if (session == 0)
891         return 0;
892     session_enter(session);
893     session_nmem = nmem_total(session->nmem);
894     session_leave(session);
895     return session_nmem;
896 }
897
898
899 struct session *new_session(NMEM nmem, struct conf_service *service,
900                             unsigned session_id)
901 {
902     int i;
903     struct session *session = nmem_malloc(nmem, sizeof(*session));
904
905     char tmp_str[50];
906
907     sprintf(tmp_str, "session#%u", session_id);
908
909     session->session_id = session_id;
910     session_log(session, YLOG_DEBUG, "New");
911     session->service = service;
912     session->relevance = 0;
913     session->total_records = 0;
914     session->number_of_warnings_unknown_elements = 0;
915     session->number_of_warnings_unknown_metadata = 0;
916     session->num_termlists = 0;
917     session->reclist = 0;
918     session->clients_active = 0;
919     session->clients_cached = 0;
920     session->settings_modified = 0;
921     session->session_nmem = nmem;
922     session->nmem = nmem_create();
923     session->databases = 0;
924     for (i = 0; i <= SESSION_WATCH_MAX; i++)
925     {
926         session->watchlist[i].data = 0;
927         session->watchlist[i].fun = 0;
928     }
929     session->normalize_cache = normalize_cache_create();
930     session->session_mutex = 0;
931     pazpar2_mutex_create(&session->session_mutex, tmp_str);
932     session_use(1);
933     return session;
934 }
935
936 const char * client_get_suggestions_xml(struct client *cl, WRBUF wrbuf);
937
938 static struct hitsbytarget *hitsbytarget_nb(struct session *se,
939                                             int *count, NMEM nmem)
940 {
941     struct hitsbytarget *res = 0;
942     struct client_list *l;
943     size_t sz = 0;
944
945     for (l = se->clients_active; l; l = l->next)
946         sz++;
947
948     res = nmem_malloc(nmem, sizeof(*res) * sz);
949     *count = 0;
950     for (l = se->clients_active; l; l = l->next)
951     {
952         struct client *cl = l->client;
953         WRBUF w = wrbuf_alloc();
954         const char *name = session_setting_oneval(client_get_database(cl),
955                                                   PZ_NAME);
956
957         res[*count].id = client_get_id(cl);
958         res[*count].name = *name ? name : "Unknown";
959         res[*count].hits = client_get_hits(cl);
960         res[*count].records = client_get_num_records(cl);
961         res[*count].diagnostic =
962             client_get_diagnostic(cl, &res[*count].addinfo);
963         res[*count].state = client_get_state_str(cl);
964         res[*count].connected  = client_get_connection(cl) ? 1 : 0;
965         session_settings_dump(se, client_get_database(cl), w);
966         res[*count].settings_xml = nmem_strdup(nmem, wrbuf_cstr(w));
967         wrbuf_rewind(w);
968         wrbuf_puts(w, "");
969         res[*count].suggestions_xml = nmem_strdup(nmem, client_get_suggestions_xml(cl, w));
970         wrbuf_destroy(w);
971         (*count)++;
972     }
973     return res;
974 }
975
976 struct hitsbytarget *get_hitsbytarget(struct session *se, int *count, NMEM nmem)
977 {
978     struct hitsbytarget *p;
979     session_enter(se);
980     p = hitsbytarget_nb(se, count, nmem);
981     session_leave(se);
982     return p;
983 }
984     
985 struct termlist_score **get_termlist_score(struct session *se,
986                                            const char *name, int *num)
987 {
988     int i;
989     struct termlist_score **tl = 0;
990
991     session_enter(se);
992     for (i = 0; i < se->num_termlists; i++)
993         if (!strcmp((const char *) se->termlists[i].name, name))
994         {
995             tl = termlist_highscore(se->termlists[i].termlist, num);
996             break;
997         }
998     session_leave(se);
999     return tl;
1000 }
1001
1002 // Compares two hitsbytarget nodes by hitcount
1003 static int cmp_ht(const void *p1, const void *p2)
1004 {
1005     const struct hitsbytarget *h1 = p1;
1006     const struct hitsbytarget *h2 = p2;
1007     return h2->hits - h1->hits;
1008 }
1009
1010 static int targets_termlist_nb(WRBUF wrbuf, struct session *se, int num,
1011                                NMEM nmem)
1012 {
1013     struct hitsbytarget *ht;
1014     int count, i;
1015
1016     ht = hitsbytarget_nb(se, &count, nmem);
1017     qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht);
1018     for (i = 0; i < count && i < num && ht[i].hits > 0; i++)
1019     {
1020
1021         // do only print terms which have display names
1022     
1023         wrbuf_puts(wrbuf, "<term>\n");
1024
1025         wrbuf_puts(wrbuf, "<id>");
1026         wrbuf_xmlputs(wrbuf, ht[i].id);
1027         wrbuf_puts(wrbuf, "</id>\n");
1028         
1029         wrbuf_puts(wrbuf, "<name>");
1030         if (!ht[i].name || !ht[i].name[0])
1031             wrbuf_xmlputs(wrbuf, "NO TARGET NAME");
1032         else
1033             wrbuf_xmlputs(wrbuf, ht[i].name);
1034         wrbuf_puts(wrbuf, "</name>\n");
1035         
1036         wrbuf_printf(wrbuf, "<frequency>" ODR_INT_PRINTF "</frequency>\n",
1037                      ht[i].hits);
1038         
1039         wrbuf_puts(wrbuf, "<state>");
1040         wrbuf_xmlputs(wrbuf, ht[i].state);
1041         wrbuf_puts(wrbuf, "</state>\n");
1042         
1043         wrbuf_printf(wrbuf, "<diagnostic>%d</diagnostic>\n", 
1044                      ht[i].diagnostic);
1045         wrbuf_puts(wrbuf, "</term>\n");
1046     }
1047     return count;
1048 }
1049
1050 void perform_termlist(struct http_channel *c, struct session *se,
1051                       const char *name, int num)
1052 {
1053     int i, j;
1054     NMEM nmem_tmp = nmem_create();
1055     char **names;
1056     int num_names = 0;
1057
1058     if (!name)
1059         name = "*";
1060
1061     nmem_strsplit(nmem_tmp, ",", name, &names, &num_names);
1062
1063     session_enter(se);
1064
1065     for (j = 0; j < num_names; j++)
1066     {
1067         const char *tname;
1068         
1069         for (i = 0; i < se->num_termlists; i++)
1070         {
1071             tname = se->termlists[i].name;
1072             if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1073             {
1074                 struct termlist_score **p = 0;
1075                 int len;
1076
1077                 wrbuf_puts(c->wrbuf, "<list name=\"");
1078                 wrbuf_xmlputs(c->wrbuf, tname);
1079                 wrbuf_puts(c->wrbuf, "\">\n");
1080
1081                 p = termlist_highscore(se->termlists[i].termlist, &len);
1082                 if (p)
1083                 {
1084                     int i;
1085                     for (i = 0; i < len && i < num; i++)
1086                     {
1087                         // prevent sending empty term elements
1088                         if (!p[i]->display_term || !p[i]->display_term[0])
1089                             continue;
1090                         
1091                         wrbuf_puts(c->wrbuf, "<term>");
1092                         wrbuf_puts(c->wrbuf, "<name>");
1093                         wrbuf_xmlputs(c->wrbuf, p[i]->display_term);
1094                         wrbuf_puts(c->wrbuf, "</name>");
1095                         
1096                         wrbuf_printf(c->wrbuf, 
1097                                      "<frequency>%d</frequency>", 
1098                                      p[i]->frequency);
1099                         wrbuf_puts(c->wrbuf, "</term>\n");
1100                     }
1101                 }
1102                 wrbuf_puts(c->wrbuf, "</list>\n");
1103             }
1104         }
1105         tname = "xtargets";
1106         if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1107         {
1108             wrbuf_puts(c->wrbuf, "<list name=\"");
1109             wrbuf_xmlputs(c->wrbuf, tname);
1110             wrbuf_puts(c->wrbuf, "\">\n");
1111
1112             targets_termlist_nb(c->wrbuf, se, num, c->nmem);
1113             wrbuf_puts(c->wrbuf, "</list>\n");
1114         }
1115     }
1116     session_leave(se);
1117     nmem_destroy(nmem_tmp);
1118 }
1119
1120 #ifdef MISSING_HEADERS
1121 void report_nmem_stats(void)
1122 {
1123     size_t in_use, is_free;
1124
1125     nmem_get_memory_in_use(&in_use);
1126     nmem_get_memory_free(&is_free);
1127
1128     yaz_log(YLOG_LOG, "nmem stat: use=%ld free=%ld", 
1129             (long) in_use, (long) is_free);
1130 }
1131 #endif
1132
1133 struct record_cluster *show_single_start(struct session *se, const char *id,
1134                                          struct record_cluster **prev_r,
1135                                          struct record_cluster **next_r)
1136 {
1137     struct record_cluster *r = 0;
1138
1139     session_enter(se);
1140     *prev_r = 0;
1141     *next_r = 0;
1142     if (se->reclist)
1143     {
1144         reclist_enter(se->reclist);
1145         while ((r = reclist_read_record(se->reclist)))
1146         {
1147             if (!strcmp(r->recid, id))
1148             {
1149                 *next_r = reclist_read_record(se->reclist);
1150                 break;
1151             }
1152             *prev_r = r;
1153         }
1154         reclist_leave(se->reclist);
1155     }
1156     if (!r)
1157         session_leave(se);
1158     return r;
1159 }
1160
1161 void show_single_stop(struct session *se, struct record_cluster *rec)
1162 {
1163     session_leave(se);
1164 }
1165
1166 struct record_cluster **show_range_start(struct session *se,
1167                                          struct reclist_sortparms *sp, 
1168                                          int start, int *num, int *total, Odr_int *sumhits)
1169 {
1170     struct record_cluster **recs;
1171     struct reclist_sortparms *spp;
1172     int i;
1173 #if USE_TIMING    
1174     yaz_timing_t t = yaz_timing_create();
1175 #endif
1176     session_enter(se);
1177     recs = nmem_malloc(se->nmem, *num * sizeof(struct record_cluster *));
1178     if (!se->relevance)
1179     {
1180         *num = 0;
1181         *total = 0;
1182         *sumhits = 0;
1183         recs = 0;
1184     }
1185     else
1186     {
1187         struct client_list *l;
1188         
1189         for (spp = sp; spp; spp = spp->next)
1190             if (spp->type == Metadata_sortkey_relevance)
1191             {
1192                 relevance_prepare_read(se->relevance, se->reclist);
1193                 break;
1194             }
1195         reclist_sort(se->reclist, sp);
1196         
1197         reclist_enter(se->reclist);
1198         *total = reclist_get_num_records(se->reclist);
1199
1200         *sumhits = 0;
1201         for (l = se->clients_active; l; l = l->next)
1202             *sumhits += client_get_hits(l->client);
1203         
1204         for (i = 0; i < start; i++)
1205             if (!reclist_read_record(se->reclist))
1206             {
1207                 *num = 0;
1208                 recs = 0;
1209                 break;
1210             }
1211         
1212         for (i = 0; i < *num; i++)
1213         {
1214             struct record_cluster *r = reclist_read_record(se->reclist);
1215             if (!r)
1216             {
1217                 *num = i;
1218                 break;
1219             }
1220             recs[i] = r;
1221         }
1222         reclist_leave(se->reclist);
1223     }
1224 #if USE_TIMING
1225     yaz_timing_stop(t);
1226     yaz_log(YLOG_LOG, "show %6.5f %3.2f %3.2f", 
1227             yaz_timing_get_real(t), yaz_timing_get_user(t),
1228             yaz_timing_get_sys(t));
1229     yaz_timing_destroy(&t);
1230 #endif
1231     return recs;
1232 }
1233
1234 void show_range_stop(struct session *se, struct record_cluster **recs)
1235 {
1236     session_leave(se);
1237 }
1238
1239 void statistics(struct session *se, struct statistics *stat)
1240 {
1241     struct client_list *l;
1242     int count = 0;
1243
1244     memset(stat, 0, sizeof(*stat));
1245     stat->num_hits = 0;
1246     for (l = se->clients_active; l; l = l->next)
1247     {
1248         struct client *cl = l->client;
1249         if (!client_get_connection(cl))
1250             stat->num_no_connection++;
1251         stat->num_hits += client_get_hits(cl);
1252         switch (client_get_state(cl))
1253         {
1254         case Client_Connecting: stat->num_connecting++; break;
1255         case Client_Working: stat->num_working++; break;
1256         case Client_Idle: stat->num_idle++; break;
1257         case Client_Failed: stat->num_failed++; break;
1258         case Client_Error: stat->num_error++; break;
1259         default: break;
1260         }
1261         count++;
1262     }
1263     stat->num_records = se->total_records;
1264
1265     stat->num_clients = count;
1266 }
1267
1268 static struct record_metadata *record_metadata_init(
1269     NMEM nmem, const char *value, enum conf_metadata_type type,
1270     struct _xmlAttr *attr)
1271 {
1272     struct record_metadata *rec_md = record_metadata_create(nmem);
1273     struct record_metadata_attr **attrp = &rec_md->attributes;
1274     
1275     for (; attr; attr = attr->next)
1276     {
1277         if (attr->children && attr->children->content)
1278         {
1279             if (strcmp((const char *) attr->name, "type"))
1280             {  /* skip the "type" attribute.. Its value is already part of
1281                   the element in output (md-%s) and so repeating it here
1282                   is redundant */
1283                 *attrp = nmem_malloc(nmem, sizeof(**attrp));
1284                 (*attrp)->name =
1285                     nmem_strdup(nmem, (const char *) attr->name);
1286                 (*attrp)->value =
1287                     nmem_strdup(nmem, (const char *) attr->children->content);
1288                 attrp = &(*attrp)->next;
1289             }
1290         }
1291     }
1292     *attrp = 0;
1293
1294     if (type == Metadata_type_generic)
1295     {
1296         char *p = nmem_strdup(nmem, value);
1297
1298         p = normalize7bit_generic(p, " ,/.:([");
1299         
1300         rec_md->data.text.disp = p;
1301         rec_md->data.text.sort = 0;
1302     }
1303     else if (type == Metadata_type_year || type == Metadata_type_date)
1304     {
1305         int first, last;
1306         int longdate = 0;
1307
1308         if (type == Metadata_type_date)
1309             longdate = 1;
1310         if (extract7bit_dates((char *) value, &first, &last, longdate) < 0)
1311             return 0;
1312
1313         rec_md->data.number.min = first;
1314         rec_md->data.number.max = last;
1315     }
1316     else
1317         return 0;
1318     return rec_md;
1319 }
1320
1321 static int get_mergekey_from_doc(xmlDoc *doc, xmlNode *root, const char *name,
1322                                  struct conf_service *service, WRBUF norm_wr)
1323 {
1324     xmlNode *n;
1325     int no_found = 0;
1326     for (n = root->children; n; n = n->next)
1327     {
1328         if (n->type != XML_ELEMENT_NODE)
1329             continue;
1330         if (!strcmp((const char *) n->name, "metadata"))
1331         {
1332             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1333             if (type == NULL) {
1334                 yaz_log(YLOG_FATAL, "Missing type attribute on metadata element. Skipping!");
1335             }
1336             else if (!strcmp(name, (const char *) type))
1337             {
1338                 xmlChar *value = xmlNodeListGetString(doc, n->children, 1);
1339                 if (value)
1340                 {
1341                     const char *norm_str;
1342                     pp2_charset_token_t prt =
1343                         pp2_charset_token_create(service->charsets, "mergekey");
1344                     
1345                     pp2_charset_token_first(prt, (const char *) value, 0);
1346                     if (wrbuf_len(norm_wr) > 0)
1347                         wrbuf_puts(norm_wr, " ");
1348                     wrbuf_puts(norm_wr, name);
1349                     while ((norm_str =
1350                             pp2_charset_token_next(prt)))
1351                     {
1352                         if (*norm_str)
1353                         {
1354                             wrbuf_puts(norm_wr, " ");
1355                             wrbuf_puts(norm_wr, norm_str);
1356                         }
1357                     }
1358                     xmlFree(value);
1359                     pp2_charset_token_destroy(prt);
1360                     no_found++;
1361                 }
1362             }
1363             xmlFree(type);
1364         }
1365     }
1366     return no_found;
1367 }
1368
1369 static const char *get_mergekey(xmlDoc *doc, struct client *cl, int record_no,
1370                                 struct conf_service *service, NMEM nmem)
1371 {
1372     char *mergekey_norm = 0;
1373     xmlNode *root = xmlDocGetRootElement(doc);
1374     WRBUF norm_wr = wrbuf_alloc();
1375
1376     /* consider mergekey from XSL first */
1377     xmlChar *mergekey = xmlGetProp(root, (xmlChar *) "mergekey");
1378     if (mergekey)
1379     {
1380         const char *norm_str;
1381         pp2_charset_token_t prt =
1382             pp2_charset_token_create(service->charsets, "mergekey");
1383
1384         pp2_charset_token_first(prt, (const char *) mergekey, 0);
1385         while ((norm_str = pp2_charset_token_next(prt)))
1386         {
1387             if (*norm_str)
1388             {
1389                 if (wrbuf_len(norm_wr))
1390                     wrbuf_puts(norm_wr, " ");
1391                 wrbuf_puts(norm_wr, norm_str);
1392             }
1393         }
1394         pp2_charset_token_destroy(prt);
1395         xmlFree(mergekey);
1396     }
1397     else
1398     {
1399         /* no mergekey defined in XSL. Look for mergekey metadata instead */
1400         int field_id;
1401         for (field_id = 0; field_id < service->num_metadata; field_id++)
1402         {
1403             struct conf_metadata *ser_md = &service->metadata[field_id];
1404             if (ser_md->mergekey != Metadata_mergekey_no)
1405             {
1406                 int r = get_mergekey_from_doc(doc, root, ser_md->name,
1407                                               service, norm_wr);
1408                 if (r == 0 && ser_md->mergekey == Metadata_mergekey_required)
1409                 {
1410                     /* no mergekey on this one and it is required.. 
1411                        Generate unique key instead */
1412                     wrbuf_rewind(norm_wr);
1413                     break;
1414                 }
1415             }
1416         }
1417     }
1418
1419     /* generate unique key if none is not generated already or is empty */
1420     if (wrbuf_len(norm_wr) == 0)
1421     {
1422         wrbuf_printf(norm_wr, "position: %s-%d",
1423                      client_get_id(cl), record_no);
1424     }
1425     else
1426     {
1427         const char *lead = "content: ";
1428         wrbuf_insert(norm_wr, 0, lead, strlen(lead));
1429     }
1430     if (wrbuf_len(norm_wr) > 0)
1431         mergekey_norm = nmem_strdup(nmem, wrbuf_cstr(norm_wr));
1432     wrbuf_destroy(norm_wr);
1433     return mergekey_norm;
1434 }
1435
1436 /** \brief see if metadata for pz:recordfilter exists 
1437     \param root xml root element of normalized record
1438     \param sdb session database for client
1439     \retval 0 if there is no metadata for pz:recordfilter
1440     \retval 1 if there is metadata for pz:recordfilter
1441
1442     If there is no pz:recordfilter defined, this function returns 1
1443     as well.
1444 */
1445     
1446 static int check_record_filter(xmlNode *root, struct session_database *sdb)
1447 {
1448     int match = 0;
1449     xmlNode *n;
1450     const char *s;
1451     s = session_setting_oneval(sdb, PZ_RECORDFILTER);
1452
1453     if (!s || !*s)
1454         return 1;
1455
1456     for (n = root->children; n; n = n->next)
1457     {
1458         if (n->type != XML_ELEMENT_NODE)
1459             continue;
1460         if (!strcmp((const char *) n->name, "metadata"))
1461         {
1462             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1463             if (type)
1464             {
1465                 size_t len;
1466                 int substring;
1467                 const char *eq;
1468
1469                 if ((eq = strchr(s, '=')))
1470                     substring = 0;
1471                 else if ((eq = strchr(s, '~')))
1472                     substring = 1;
1473                 if (eq)
1474                     len = eq - s;
1475                 else
1476                     len = strlen(s);
1477                 if (len == strlen((const char *)type) &&
1478                     !memcmp((const char *) type, s, len))
1479                 {
1480                     xmlChar *value = xmlNodeGetContent(n);
1481                     if (value && *value)
1482                     {
1483                         if (!eq ||
1484                             (substring && strstr((const char *) value, eq+1)) ||
1485                             (!substring && !strcmp((const char *) value, eq + 1)))
1486                             match = 1;
1487                     }
1488                     xmlFree(value);
1489                 }
1490                 xmlFree(type);
1491             }
1492         }
1493     }
1494     return match;
1495 }
1496
1497
1498 static int ingest_to_cluster(struct client *cl,
1499                              xmlDoc *xdoc,
1500                              xmlNode *root,
1501                              int record_no,
1502                              const char *mergekey_norm);
1503
1504 /** \brief ingest XML record
1505     \param cl client holds the result set for record
1506     \param rec record buffer (0 terminated)
1507     \param record_no record position (1, 2, ..)
1508     \param nmem working NMEM
1509     \retval 0 OK
1510     \retval -1 failure
1511     \retval -2 Filtered
1512 */
1513 int ingest_record(struct client *cl, const char *rec,
1514                   int record_no, NMEM nmem)
1515 {
1516     struct session *se = client_get_session(cl);
1517     int ret = 0;
1518     struct session_database *sdb = client_get_database(cl);
1519     struct conf_service *service = se->service;
1520     xmlDoc *xdoc = normalize_record(se, sdb, service, rec, nmem);
1521     xmlNode *root;
1522     const char *mergekey_norm;
1523     
1524     if (!xdoc)
1525         return -1;
1526     
1527     root = xmlDocGetRootElement(xdoc);
1528     
1529     if (!check_record_filter(root, sdb))
1530     {
1531         session_log(se, YLOG_LOG, "Filtered out record no %d from %s",
1532                     record_no, sdb->database->id);
1533         xmlFreeDoc(xdoc);
1534         return -2;
1535     }
1536     
1537     mergekey_norm = get_mergekey(xdoc, cl, record_no, service, nmem);
1538     if (!mergekey_norm)
1539     {
1540         session_log(se, YLOG_WARN, "Got no mergekey");
1541         xmlFreeDoc(xdoc);
1542         return -1;
1543     }
1544     session_enter(se);
1545     if (client_get_session(cl) == se)
1546         ret = ingest_to_cluster(cl, xdoc, root, record_no, mergekey_norm);
1547     session_leave(se);
1548     
1549     xmlFreeDoc(xdoc);
1550     return ret;
1551 }
1552
1553 static int check_limit_local(struct client *cl,
1554                              struct record *record,
1555                              int record_no)
1556 {
1557     int skip_record = 0;
1558     struct session *se = client_get_session(cl);
1559     struct conf_service *service = se->service;
1560     NMEM nmem_tmp = nmem_create();
1561     struct session_database *sdb = client_get_database(cl);
1562     int l = 0;
1563     while (!skip_record)
1564     {
1565         struct conf_metadata *ser_md = 0;
1566         struct record_metadata *rec_md = 0;
1567         int md_field_id;
1568         char **values = 0;
1569         int i, num_v = 0;
1570         
1571         const char *name =
1572             client_get_facet_limit_local(cl, sdb, &l, nmem_tmp, &num_v,
1573                                          &values);
1574         if (!name)
1575             break;
1576         
1577         md_field_id = conf_service_metadata_field_id(service, name);
1578         if (md_field_id < 0)
1579         {
1580             skip_record = 1;
1581             break;
1582         }
1583         ser_md = &service->metadata[md_field_id];
1584         rec_md = record->metadata[md_field_id];
1585         yaz_log(YLOG_LOG, "check limit local %s", name);
1586         for (i = 0; i < num_v; )
1587         {
1588             if (rec_md)
1589             {
1590                 if (ser_md->type == Metadata_type_year 
1591                     || ser_md->type == Metadata_type_date)
1592                 {
1593                     int y = atoi(values[i]);
1594                     if (y >= rec_md->data.number.min 
1595                         && y <= rec_md->data.number.max)
1596                         break;
1597                 }
1598                 else
1599                 {
1600                     yaz_log(YLOG_LOG, "cmp: '%s' '%s'",
1601                             rec_md->data.text.disp, values[i]);
1602                     if (!strcmp(rec_md->data.text.disp, values[i]))
1603                     {
1604                         break;
1605                     }
1606                 }
1607                 rec_md = rec_md->next;
1608             }
1609             else
1610             {
1611                 rec_md = record->metadata[md_field_id];
1612                 i++;
1613             }
1614         }
1615         if (i == num_v)
1616         {
1617             skip_record = 1;
1618             break;
1619         }
1620     }
1621     nmem_destroy(nmem_tmp);
1622     return skip_record;
1623 }
1624                              
1625 static int ingest_to_cluster(struct client *cl,
1626                              xmlDoc *xdoc,
1627                              xmlNode *root,
1628                              int record_no,
1629                              const char *mergekey_norm)
1630 {
1631     xmlNode *n;
1632     xmlChar *type = 0;
1633     xmlChar *value = 0;
1634     struct session *se = client_get_session(cl);
1635     struct conf_service *service = se->service;
1636     int term_factor = 1;
1637     struct record_cluster *cluster;
1638     struct session_database *sdb = client_get_database(cl);
1639     struct record *record = record_create(se->nmem, 
1640                                           service->num_metadata,
1641                                           service->num_sortkeys, cl,
1642                                           record_no);
1643
1644     for (n = root->children; n; n = n->next)
1645     {
1646         if (type)
1647             xmlFree(type);
1648         if (value)
1649             xmlFree(value);
1650         type = value = 0;
1651         
1652         if (n->type != XML_ELEMENT_NODE)
1653             continue;
1654         if (!strcmp((const char *) n->name, "metadata"))
1655         {
1656             struct conf_metadata *ser_md = 0;
1657             struct record_metadata **wheretoput = 0;
1658             struct record_metadata *rec_md = 0;
1659             int md_field_id = -1;
1660             
1661             type = xmlGetProp(n, (xmlChar *) "type");
1662             value = xmlNodeListGetString(xdoc, n->children, 1);
1663             
1664             if (!type || !value || !*value)
1665                 continue;
1666             
1667             md_field_id 
1668                 = conf_service_metadata_field_id(service, (const char *) type);
1669             if (md_field_id < 0)
1670             {
1671                 if (se->number_of_warnings_unknown_metadata == 0)
1672                 {
1673                     session_log(se, YLOG_WARN, 
1674                             "Ignoring unknown metadata element: %s", type);
1675                 }
1676                 se->number_of_warnings_unknown_metadata++;
1677                 continue;
1678             }
1679            
1680             ser_md = &service->metadata[md_field_id];
1681
1682             // non-merged metadata
1683             rec_md = record_metadata_init(se->nmem, (const char *) value,
1684                                           ser_md->type, n->properties);
1685             if (!rec_md)
1686             {
1687                 session_log(se, YLOG_WARN, "bad metadata data '%s' "
1688                             "for element '%s'", value, type);
1689                 continue;
1690             }
1691             wheretoput = &record->metadata[md_field_id];
1692             while (*wheretoput)
1693                 wheretoput = &(*wheretoput)->next;
1694             *wheretoput = rec_md;
1695         }
1696     }
1697
1698     if (check_limit_local(cl, record, record_no))
1699     {
1700         session_log(se, YLOG_LOG, "Facet filtered out record no %d from %s",
1701                     record_no, sdb->database->id);
1702         if (type)
1703             xmlFree(type);
1704         if (value)
1705             xmlFree(value);
1706         return -2;
1707     }
1708     cluster = reclist_insert(se->reclist, service, record,
1709                              mergekey_norm, &se->total_merged);
1710     if (!cluster)
1711         return -1;
1712
1713     {
1714         const char *use_term_factor_str =
1715             session_setting_oneval(sdb, PZ_TERMLIST_TERM_FACTOR);
1716         if (use_term_factor_str && use_term_factor_str[0] == '1')
1717         {
1718             int maxrecs = client_get_maxrecs(cl);
1719             int hits = (int) client_get_hits(cl);
1720             term_factor = MAX(hits, maxrecs) /  MAX(1, maxrecs);
1721             assert(term_factor >= 1);
1722             yaz_log(YLOG_DEBUG, "Using term factor: %d (%d / %d)", term_factor, MAX(hits, maxrecs), MAX(1, maxrecs));
1723         }
1724     }
1725
1726     if (global_parameters.dump_records)
1727         session_log(se, YLOG_LOG, "Cluster id %s from %s (#%d)", cluster->recid,
1728                     sdb->database->id, record_no);
1729
1730
1731     relevance_newrec(se->relevance, cluster);
1732     
1733     // now parsing XML record and adding data to cluster or record metadata
1734     for (n = root->children; n; n = n->next)
1735     {
1736         pp2_charset_token_t prt;
1737         if (type)
1738             xmlFree(type);
1739         if (value)
1740             xmlFree(value);
1741         type = value = 0;
1742         
1743         if (n->type != XML_ELEMENT_NODE)
1744             continue;
1745         if (!strcmp((const char *) n->name, "metadata"))
1746         {
1747             struct conf_metadata *ser_md = 0;
1748             struct conf_sortkey *ser_sk = 0;
1749             struct record_metadata **wheretoput = 0;
1750             struct record_metadata *rec_md = 0;
1751             int md_field_id = -1;
1752             int sk_field_id = -1;
1753             
1754             type = xmlGetProp(n, (xmlChar *) "type");
1755             value = xmlNodeListGetString(xdoc, n->children, 1);
1756             
1757             if (!type || !value || !*value)
1758                 continue;
1759             
1760             md_field_id 
1761                 = conf_service_metadata_field_id(service, (const char *) type);
1762             if (md_field_id < 0)
1763                 continue;
1764             
1765             ser_md = &service->metadata[md_field_id];
1766             
1767             if (ser_md->sortkey_offset >= 0)
1768             {
1769                 sk_field_id = ser_md->sortkey_offset;
1770                 ser_sk = &service->sortkeys[sk_field_id];
1771             }
1772
1773             // merged metadata
1774             rec_md = record_metadata_init(se->nmem, (const char *) value,
1775                                           ser_md->type, 0);
1776             if (!rec_md)
1777                 continue;
1778
1779             wheretoput = &cluster->metadata[md_field_id];
1780
1781             // and polulate with data:
1782             // assign cluster or record based on merge action
1783             if (ser_md->merge == Metadata_merge_unique)
1784             {
1785                 while (*wheretoput)
1786                 {
1787                     if (!strcmp((const char *) (*wheretoput)->data.text.disp, 
1788                                 rec_md->data.text.disp))
1789                         break;
1790                     wheretoput = &(*wheretoput)->next;
1791                 }
1792                 if (!*wheretoput)
1793                     *wheretoput = rec_md;
1794             }
1795             else if (ser_md->merge == Metadata_merge_longest)
1796             {
1797                 if (!*wheretoput 
1798                     || strlen(rec_md->data.text.disp) 
1799                     > strlen((*wheretoput)->data.text.disp))
1800                 {
1801                     *wheretoput = rec_md;
1802                     if (ser_sk)
1803                     {
1804                         const char *sort_str = 0;
1805                         int skip_article = 
1806                             ser_sk->type == Metadata_sortkey_skiparticle;
1807
1808                         if (!cluster->sortkeys[sk_field_id])
1809                             cluster->sortkeys[sk_field_id] = 
1810                                 nmem_malloc(se->nmem, 
1811                                             sizeof(union data_types));
1812                          
1813                         prt =
1814                             pp2_charset_token_create(service->charsets, "sort");
1815
1816                         pp2_charset_token_first(prt, rec_md->data.text.disp,
1817                                                 skip_article);
1818
1819                         pp2_charset_token_next(prt);
1820                          
1821                         sort_str = pp2_get_sort(prt);
1822                          
1823                         cluster->sortkeys[sk_field_id]->text.disp = 
1824                             rec_md->data.text.disp;
1825                         if (!sort_str)
1826                         {
1827                             sort_str = rec_md->data.text.disp;
1828                             session_log(se, YLOG_WARN, 
1829                                     "Could not make sortkey. Bug #1858");
1830                         }
1831                         cluster->sortkeys[sk_field_id]->text.sort = 
1832                             nmem_strdup(se->nmem, sort_str);
1833                         pp2_charset_token_destroy(prt);
1834                     }
1835                 }
1836             }
1837             else if (ser_md->merge == Metadata_merge_all)
1838             {
1839                 while (*wheretoput)
1840                     wheretoput = &(*wheretoput)->next;
1841                 *wheretoput = rec_md;
1842             }
1843             else if (ser_md->merge == Metadata_merge_range)
1844             {
1845                 if (!*wheretoput)
1846                 {
1847                     *wheretoput = rec_md;
1848                     if (ser_sk)
1849                         cluster->sortkeys[sk_field_id] 
1850                             = &rec_md->data;
1851                 }
1852                 else
1853                 {
1854                     int this_min = rec_md->data.number.min;
1855                     int this_max = rec_md->data.number.max;
1856                     if (this_min < (*wheretoput)->data.number.min)
1857                         (*wheretoput)->data.number.min = this_min;
1858                     if (this_max > (*wheretoput)->data.number.max)
1859                         (*wheretoput)->data.number.max = this_max;
1860                 }
1861             }
1862
1863
1864             // ranking of _all_ fields enabled ... 
1865             if (ser_md->rank)
1866                 relevance_countwords(se->relevance, cluster, 
1867                                      (char *) value, ser_md->rank,
1868                                      ser_md->name);
1869
1870             // construct facets ... unless the client already has reported them
1871             if (ser_md->termlist && !client_has_facet(cl, (char *) type))
1872             {
1873                 if (ser_md->type == Metadata_type_year)
1874                 {
1875                     char year[64];
1876                     sprintf(year, "%d", rec_md->data.number.max);
1877
1878                     add_facet(se, (char *) type, year, term_factor);
1879                     if (rec_md->data.number.max != rec_md->data.number.min)
1880                     {
1881                         sprintf(year, "%d", rec_md->data.number.min);
1882                         add_facet(se, (char *) type, year, term_factor);
1883                     }
1884                 }
1885                 else
1886                     add_facet(se, (char *) type, (char *) value, term_factor);
1887             }
1888
1889             // cleaning up
1890             xmlFree(type);
1891             xmlFree(value);
1892             type = value = 0;
1893         }
1894         else
1895         {
1896             if (se->number_of_warnings_unknown_elements == 0)
1897                 session_log(se, YLOG_WARN,
1898                         "Unexpected element in internal record: %s", n->name);
1899             se->number_of_warnings_unknown_elements++;
1900         }
1901     }
1902     if (type)
1903         xmlFree(type);
1904     if (value)
1905         xmlFree(value);
1906
1907     relevance_donerecord(se->relevance, cluster);
1908     se->total_records++;
1909
1910     return 0;
1911 }
1912
1913 void session_log(struct session *s, int level, const char *fmt, ...)
1914 {
1915     char buf[1024];
1916     va_list ap;
1917     va_start(ap, fmt);
1918
1919     yaz_vsnprintf(buf, sizeof(buf)-30, fmt, ap);
1920     yaz_log(level, "Session %u: %s", s->session_id, buf);
1921
1922     va_end(ap);
1923 }
1924
1925 /*
1926  * Local variables:
1927  * c-basic-offset: 4
1928  * c-file-style: "Stroustrup"
1929  * indent-tabs-mode: nil
1930  * End:
1931  * vim: shiftwidth=4 tabstop=8 expandtab
1932  */
1933