3e00cc10fa0d9603b76c9ab0bff1a0a4f057a851
[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 #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
527         l = xmalloc(sizeof(*l));
528         l->client = cl;
529         l->next = se->clients_cached;
530         se->clients_cached = l;
531     }
532     /* set session always. If may be 0 if client is not active */
533     client_set_session(cl, se);
534
535     l = xmalloc(sizeof(*l));
536     l->client = cl;
537     l->next = se->clients_active;
538     se->clients_active = l;
539 }
540
541 static void session_reset_active_clients(struct session *se,
542                                          struct client_list *new_list)
543 {
544     struct client_list *l;
545
546     session_enter(se);
547     l = se->clients_active;
548     se->clients_active = new_list;
549     session_leave(se);
550
551     while (l)
552     {
553         struct client_list *l_next = l->next;
554
555         client_lock(l->client); 
556         client_set_session(l->client, 0); /* mark client inactive */
557         client_unlock(l->client);
558
559         xfree(l);
560         l = l_next;
561     }
562 }
563
564 static void session_remove_cached_clients(struct session *se)
565 {
566     struct client_list *l;
567
568     session_reset_active_clients(se, 0);
569
570     session_enter(se);
571     l = se->clients_cached;
572     se->clients_cached = 0;
573     session_leave(se);
574
575     while (l)
576     {
577         struct client_list *l_next = l->next;
578         client_lock(l->client);
579         client_set_session(l->client, 0);
580         client_set_database(l->client, 0);
581         client_unlock(l->client);
582         client_destroy(l->client);
583         xfree(l);
584         l = l_next;
585     }
586 }
587
588 // Associates a set of clients with a session;
589 // Note: Session-databases represent databases with per-session 
590 // setting overrides
591 static int select_targets(struct session *se, const char *filter)
592 {
593     return session_grep_databases(se, filter, select_targets_callback);
594 }
595
596 int session_active_clients(struct session *s)
597 {
598     struct client_list *l;
599     int res = 0;
600
601     for (l = s->clients_active; l; l = l->next)
602         if (client_is_active(l->client))
603             res++;
604
605     return res;
606 }
607
608 int session_is_preferred_clients_ready(struct session *s)
609 {
610     struct client_list *l;
611     int res = 0;
612
613     for (l = s->clients_active; l; l = l->next)
614         if (client_is_active_preferred(l->client))
615             res++;
616     session_log(s, YLOG_DEBUG, "Has %d active preferred clients.", res);
617     return res == 0;
618 }
619
620 void session_sort(struct session *se, const char *field, int increasing)
621 {
622     struct session_sorted_results *sr;
623     struct client_list *l;
624
625     session_enter(se);
626
627     /* see if we already have sorted for this critieria */
628     for (sr = se->sorted_results; sr; sr = sr->next)
629     {
630         if (!strcmp(field, sr->field) && increasing == sr->increasing)
631             break;
632     }
633     if (sr)
634     {
635         yaz_log(YLOG_LOG, "search_sort: field=%s increasing=%d already fetched",
636                 field, increasing);
637         session_leave(se);
638         return;
639     }
640     yaz_log(YLOG_LOG, "search_sort: field=%s increasing=%d must fetch",
641             field, increasing);
642     sr = nmem_malloc(se->nmem, sizeof(*sr));
643     sr->field = nmem_strdup(se->nmem, field);
644     sr->increasing = increasing;
645     sr->next = se->sorted_results;
646     se->sorted_results = sr;
647     
648     for (l = se->clients_active; l; l = l->next)
649     {
650         struct client *cl = l->client;
651         if (client_get_state(cl) == Client_Connecting ||
652             client_get_state(cl) == Client_Idle ||
653             client_get_state(cl) == Client_Working)
654             client_start_search(cl);
655     }
656     session_leave(se);
657 }
658
659 enum pazpar2_error_code session_search(struct session *se,
660                                        const char *query,
661                                        const char *startrecs,
662                                        const char *maxrecs,
663                                        const char *filter,
664                                        const char *limit,
665                                        const char **addinfo,
666                                        const char *sort_field, int increasing)
667 {
668     int live_channels = 0;
669     int no_working = 0;
670     int no_failed_query = 0;
671     int no_failed_limit = 0;
672     struct client_list *l, *l0;
673     struct timeval tval;
674     facet_limits_t facet_limits;
675
676     session_log(se, YLOG_DEBUG, "Search");
677
678     *addinfo = 0;
679
680     if (se->settings_modified)
681         session_remove_cached_clients(se);
682     else
683         session_reset_active_clients(se, 0);
684     
685     session_enter(se);
686     reclist_destroy(se->reclist);
687     se->reclist = 0;
688     se->settings_modified = 0;
689     relevance_destroy(&se->relevance);
690     nmem_reset(se->nmem);
691     se->total_records = se->total_merged = 0;
692     se->num_termlists = 0;
693
694     /* reset list of sorted results and clear to relevance search */
695     se->sorted_results = nmem_malloc(se->nmem, sizeof(*se->sorted_results));
696     se->sorted_results->field = nmem_strdup(se->nmem, sort_field);
697     se->sorted_results->increasing = increasing;
698     se->sorted_results->next = 0;
699     
700     live_channels = select_targets(se, filter);
701     if (!live_channels)
702     {
703         session_leave(se);
704         return PAZPAR2_NO_TARGETS;
705     }
706     se->reclist = reclist_create(se->nmem);
707
708     yaz_gettimeofday(&tval);
709     
710     tval.tv_sec += 5;
711
712     facet_limits = facet_limits_create(limit);
713     if (!facet_limits)
714     {
715         *addinfo = "limit";
716         session_leave(se);
717         return PAZPAR2_MALFORMED_PARAMETER_VALUE;
718     }
719
720     l0 = se->clients_active;
721     se->clients_active = 0;
722     session_leave(se);
723
724     for (l = l0; l; l = l->next)
725     {
726         int parse_ret;
727         struct client *cl = l->client;
728
729         if (prepare_map(se, client_get_database(cl)) < 0)
730             continue;
731
732         parse_ret = client_parse_query(cl, query, facet_limits, startrecs,
733             maxrecs);
734         if (parse_ret == -1)
735             no_failed_query++;
736         else if (parse_ret == -2)
737             no_failed_limit++;
738         else
739         {
740             int r =
741                 client_prep_connection(cl, se->service->z3950_operation_timeout,
742                                        se->service->z3950_session_timeout,
743                                        se->service->server->iochan_man,
744                                        &tval, 0);
745             if (parse_ret == 1 && r == 2)
746             {
747                 session_log(se, YLOG_LOG, "client REUSE %s", client_get_id(cl));
748                 client_reingest(cl);
749             }
750             else
751             {
752                 session_log(se, YLOG_LOG, "client NEW %s", client_get_id(cl));
753                 client_start_search(cl);
754             }
755             no_working++;
756         }
757     }
758     facet_limits_destroy(facet_limits);
759     session_reset_active_clients(se, l0);
760
761     if (no_working == 0)
762     {
763         if (no_failed_query > 0)
764         {
765             *addinfo = "query";
766             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
767         }
768         else if (no_failed_limit > 0)
769         {
770             *addinfo = "limit";
771             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
772         }
773         else
774             return PAZPAR2_NO_TARGETS;
775     }
776     yaz_log(YLOG_LOG, "session_start_search done");
777     return PAZPAR2_NO_ERROR;
778 }
779
780 // Creates a new session_database object for a database
781 static void session_init_databases_fun(void *context, struct database *db)
782 {
783     struct session *se = (struct session *) context;
784     struct session_database *new = nmem_malloc(se->session_nmem, sizeof(*new));
785     int i;
786
787     new->database = db;
788     
789     new->map = 0;
790     assert(db->settings);
791     new->settings = nmem_malloc(se->session_nmem,
792                                 sizeof(struct settings *) * db->num_settings);
793     new->num_settings = db->num_settings;
794     for (i = 0; i < db->num_settings; i++)
795     {
796         struct setting *setting = db->settings[i];
797         new->settings[i] = setting;
798     }
799     new->next = se->databases;
800     se->databases = new;
801 }
802
803 // Doesn't free memory associated with sdb -- nmem takes care of that
804 static void session_database_destroy(struct session_database *sdb)
805 {
806     sdb->map = 0;
807 }
808
809 // Initialize session_database list -- this represents this session's view
810 // of the database list -- subject to modification by the settings ws command
811 void session_init_databases(struct session *se)
812 {
813     se->databases = 0;
814     predef_grep_databases(se, se->service, session_init_databases_fun);
815 }
816
817 // Probably session_init_databases_fun should be refactored instead of
818 // called here.
819 static struct session_database *load_session_database(struct session *se, 
820                                                       char *id)
821 {
822     struct database *db = new_database(id, se->session_nmem);
823
824     session_init_databases_fun((void*) se, db);
825
826     // New sdb is head of se->databases list
827     return se->databases;
828 }
829
830 // Find an existing session database. If not found, load it
831 static struct session_database *find_session_database(struct session *se, 
832                                                       char *id)
833 {
834     struct session_database *sdb;
835
836     for (sdb = se->databases; sdb; sdb = sdb->next)
837         if (!strcmp(sdb->database->id, id))
838             return sdb;
839     return load_session_database(se, id);
840 }
841
842 // Apply a session override to a database
843 void session_apply_setting(struct session *se, char *dbname, char *setting,
844                            char *value)
845 {
846     struct session_database *sdb = find_session_database(se, dbname);
847     struct conf_service *service = se->service;
848     struct setting *new = nmem_malloc(se->session_nmem, sizeof(*new));
849     int offset = settings_create_offset(service, setting);
850
851     expand_settings_array(&sdb->settings, &sdb->num_settings, offset,
852                           se->session_nmem);
853     new->precedence = 0;
854     new->target = dbname;
855     new->name = setting;
856     new->value = value;
857     new->next = sdb->settings[offset];
858     sdb->settings[offset] = new;
859
860     se->settings_modified = 1;
861
862     // Force later recompute of settings-driven data structures
863     // (happens when a search starts and client connections are prepared)
864     switch (offset)
865     {
866     case PZ_XSLT:
867         if (sdb->map)
868         {
869             sdb->map = 0;
870         }
871         break;
872     }
873 }
874
875 void session_destroy(struct session *se)
876 {
877     struct session_database *sdb;
878     session_log(se, YLOG_DEBUG, "Destroying");
879     session_use(-1);
880     session_remove_cached_clients(se);
881
882     for (sdb = se->databases; sdb; sdb = sdb->next)
883         session_database_destroy(sdb);
884     normalize_cache_destroy(se->normalize_cache);
885     relevance_destroy(&se->relevance);
886     reclist_destroy(se->reclist);
887     nmem_destroy(se->nmem);
888     service_destroy(se->service);
889     yaz_mutex_destroy(&se->session_mutex);
890 }
891
892 size_t session_get_memory_status(struct session *session) {
893     size_t session_nmem;
894     if (session == 0)
895         return 0;
896     session_enter(session);
897     session_nmem = nmem_total(session->nmem);
898     session_leave(session);
899     return session_nmem;
900 }
901
902
903 struct session *new_session(NMEM nmem, struct conf_service *service,
904                             unsigned session_id)
905 {
906     int i;
907     struct session *session = nmem_malloc(nmem, sizeof(*session));
908
909     char tmp_str[50];
910
911     sprintf(tmp_str, "session#%u", session_id);
912
913     session->session_id = session_id;
914     session_log(session, YLOG_DEBUG, "New");
915     session->service = service;
916     session->relevance = 0;
917     session->total_records = 0;
918     session->number_of_warnings_unknown_elements = 0;
919     session->number_of_warnings_unknown_metadata = 0;
920     session->num_termlists = 0;
921     session->reclist = 0;
922     session->clients_active = 0;
923     session->clients_cached = 0;
924     session->settings_modified = 0;
925     session->session_nmem = nmem;
926     session->nmem = nmem_create();
927     session->databases = 0;
928     for (i = 0; i <= SESSION_WATCH_MAX; i++)
929     {
930         session->watchlist[i].data = 0;
931         session->watchlist[i].fun = 0;
932     }
933     session->normalize_cache = normalize_cache_create();
934     session->session_mutex = 0;
935     pazpar2_mutex_create(&session->session_mutex, tmp_str);
936     session_use(1);
937     return session;
938 }
939
940 const char * client_get_suggestions_xml(struct client *cl, WRBUF wrbuf);
941
942 static struct hitsbytarget *hitsbytarget_nb(struct session *se,
943                                             int *count, NMEM nmem)
944 {
945     struct hitsbytarget *res = 0;
946     struct client_list *l;
947     size_t sz = 0;
948
949     for (l = se->clients_active; l; l = l->next)
950         sz++;
951
952     res = nmem_malloc(nmem, sizeof(*res) * sz);
953     *count = 0;
954     for (l = se->clients_active; l; l = l->next)
955     {
956         struct client *cl = l->client;
957         WRBUF w = wrbuf_alloc();
958         const char *name = session_setting_oneval(client_get_database(cl),
959                                                   PZ_NAME);
960
961         res[*count].id = client_get_id(cl);
962         res[*count].name = *name ? name : "Unknown";
963         res[*count].hits = client_get_hits(cl);
964         res[*count].records = client_get_num_records(cl);
965         res[*count].diagnostic =
966             client_get_diagnostic(cl, &res[*count].addinfo);
967         res[*count].state = client_get_state_str(cl);
968         res[*count].connected  = client_get_connection(cl) ? 1 : 0;
969         session_settings_dump(se, client_get_database(cl), w);
970         res[*count].settings_xml = nmem_strdup(nmem, wrbuf_cstr(w));
971         wrbuf_rewind(w);
972         wrbuf_puts(w, "");
973         res[*count].suggestions_xml = nmem_strdup(nmem, client_get_suggestions_xml(cl, w));
974         wrbuf_destroy(w);
975         (*count)++;
976     }
977     return res;
978 }
979
980 struct hitsbytarget *get_hitsbytarget(struct session *se, int *count, NMEM nmem)
981 {
982     struct hitsbytarget *p;
983     session_enter(se);
984     p = hitsbytarget_nb(se, count, nmem);
985     session_leave(se);
986     return p;
987 }
988     
989 struct termlist_score **get_termlist_score(struct session *se,
990                                            const char *name, int *num)
991 {
992     int i;
993     struct termlist_score **tl = 0;
994
995     session_enter(se);
996     for (i = 0; i < se->num_termlists; i++)
997         if (!strcmp((const char *) se->termlists[i].name, name))
998         {
999             tl = termlist_highscore(se->termlists[i].termlist, num);
1000             break;
1001         }
1002     session_leave(se);
1003     return tl;
1004 }
1005
1006 // Compares two hitsbytarget nodes by hitcount
1007 static int cmp_ht(const void *p1, const void *p2)
1008 {
1009     const struct hitsbytarget *h1 = p1;
1010     const struct hitsbytarget *h2 = p2;
1011     return h2->hits - h1->hits;
1012 }
1013
1014 static int targets_termlist_nb(WRBUF wrbuf, struct session *se, int num,
1015                                NMEM nmem)
1016 {
1017     struct hitsbytarget *ht;
1018     int count, i;
1019
1020     ht = hitsbytarget_nb(se, &count, nmem);
1021     qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht);
1022     for (i = 0; i < count && i < num && ht[i].hits > 0; i++)
1023     {
1024
1025         // do only print terms which have display names
1026     
1027         wrbuf_puts(wrbuf, "<term>\n");
1028
1029         wrbuf_puts(wrbuf, "<id>");
1030         wrbuf_xmlputs(wrbuf, ht[i].id);
1031         wrbuf_puts(wrbuf, "</id>\n");
1032         
1033         wrbuf_puts(wrbuf, "<name>");
1034         if (!ht[i].name || !ht[i].name[0])
1035             wrbuf_xmlputs(wrbuf, "NO TARGET NAME");
1036         else
1037             wrbuf_xmlputs(wrbuf, ht[i].name);
1038         wrbuf_puts(wrbuf, "</name>\n");
1039         
1040         wrbuf_printf(wrbuf, "<frequency>" ODR_INT_PRINTF "</frequency>\n",
1041                      ht[i].hits);
1042         
1043         wrbuf_puts(wrbuf, "<state>");
1044         wrbuf_xmlputs(wrbuf, ht[i].state);
1045         wrbuf_puts(wrbuf, "</state>\n");
1046         
1047         wrbuf_printf(wrbuf, "<diagnostic>%d</diagnostic>\n", 
1048                      ht[i].diagnostic);
1049         wrbuf_puts(wrbuf, "</term>\n");
1050     }
1051     return count;
1052 }
1053
1054 void perform_termlist(struct http_channel *c, struct session *se,
1055                       const char *name, int num)
1056 {
1057     int i, j;
1058     NMEM nmem_tmp = nmem_create();
1059     char **names;
1060     int num_names = 0;
1061
1062     if (!name)
1063         name = "*";
1064
1065     nmem_strsplit(nmem_tmp, ",", name, &names, &num_names);
1066
1067     session_enter(se);
1068
1069     for (j = 0; j < num_names; j++)
1070     {
1071         const char *tname;
1072         
1073         for (i = 0; i < se->num_termlists; i++)
1074         {
1075             tname = se->termlists[i].name;
1076             if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1077             {
1078                 struct termlist_score **p = 0;
1079                 int len;
1080
1081                 wrbuf_puts(c->wrbuf, "<list name=\"");
1082                 wrbuf_xmlputs(c->wrbuf, tname);
1083                 wrbuf_puts(c->wrbuf, "\">\n");
1084
1085                 p = termlist_highscore(se->termlists[i].termlist, &len);
1086                 if (p)
1087                 {
1088                     int i;
1089                     for (i = 0; i < len && i < num; i++)
1090                     {
1091                         // prevent sending empty term elements
1092                         if (!p[i]->display_term || !p[i]->display_term[0])
1093                             continue;
1094                         
1095                         wrbuf_puts(c->wrbuf, "<term>");
1096                         wrbuf_puts(c->wrbuf, "<name>");
1097                         wrbuf_xmlputs(c->wrbuf, p[i]->display_term);
1098                         wrbuf_puts(c->wrbuf, "</name>");
1099                         
1100                         wrbuf_printf(c->wrbuf, 
1101                                      "<frequency>%d</frequency>", 
1102                                      p[i]->frequency);
1103                         wrbuf_puts(c->wrbuf, "</term>\n");
1104                     }
1105                 }
1106                 wrbuf_puts(c->wrbuf, "</list>\n");
1107             }
1108         }
1109         tname = "xtargets";
1110         if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1111         {
1112             wrbuf_puts(c->wrbuf, "<list name=\"");
1113             wrbuf_xmlputs(c->wrbuf, tname);
1114             wrbuf_puts(c->wrbuf, "\">\n");
1115
1116             targets_termlist_nb(c->wrbuf, se, num, c->nmem);
1117             wrbuf_puts(c->wrbuf, "</list>\n");
1118         }
1119     }
1120     session_leave(se);
1121     nmem_destroy(nmem_tmp);
1122 }
1123
1124 #ifdef MISSING_HEADERS
1125 void report_nmem_stats(void)
1126 {
1127     size_t in_use, is_free;
1128
1129     nmem_get_memory_in_use(&in_use);
1130     nmem_get_memory_free(&is_free);
1131
1132     yaz_log(YLOG_LOG, "nmem stat: use=%ld free=%ld", 
1133             (long) in_use, (long) is_free);
1134 }
1135 #endif
1136
1137 struct record_cluster *show_single_start(struct session *se, const char *id,
1138                                          struct record_cluster **prev_r,
1139                                          struct record_cluster **next_r)
1140 {
1141     struct record_cluster *r = 0;
1142
1143     session_enter(se);
1144     *prev_r = 0;
1145     *next_r = 0;
1146     if (se->reclist)
1147     {
1148         reclist_enter(se->reclist);
1149         while ((r = reclist_read_record(se->reclist)))
1150         {
1151             if (!strcmp(r->recid, id))
1152             {
1153                 *next_r = reclist_read_record(se->reclist);
1154                 break;
1155             }
1156             *prev_r = r;
1157         }
1158         reclist_leave(se->reclist);
1159     }
1160     if (!r)
1161         session_leave(se);
1162     return r;
1163 }
1164
1165 void show_single_stop(struct session *se, struct record_cluster *rec)
1166 {
1167     session_leave(se);
1168 }
1169
1170 struct record_cluster **show_range_start(struct session *se,
1171                                          struct reclist_sortparms *sp, 
1172                                          int start, int *num, int *total, Odr_int *sumhits)
1173 {
1174     struct record_cluster **recs;
1175     struct reclist_sortparms *spp;
1176     int i;
1177 #if USE_TIMING    
1178     yaz_timing_t t = yaz_timing_create();
1179 #endif
1180     session_enter(se);
1181     recs = nmem_malloc(se->nmem, *num * sizeof(struct record_cluster *));
1182     if (!se->relevance)
1183     {
1184         *num = 0;
1185         *total = 0;
1186         *sumhits = 0;
1187         recs = 0;
1188     }
1189     else
1190     {
1191         struct client_list *l;
1192         
1193         for (spp = sp; spp; spp = spp->next)
1194             if (spp->type == Metadata_sortkey_relevance)
1195             {
1196                 relevance_prepare_read(se->relevance, se->reclist);
1197                 break;
1198             }
1199         reclist_sort(se->reclist, sp);
1200         
1201         reclist_enter(se->reclist);
1202         *total = reclist_get_num_records(se->reclist);
1203
1204         *sumhits = 0;
1205         for (l = se->clients_active; l; l = l->next)
1206             *sumhits += client_get_hits(l->client);
1207         
1208         for (i = 0; i < start; i++)
1209             if (!reclist_read_record(se->reclist))
1210             {
1211                 *num = 0;
1212                 recs = 0;
1213                 break;
1214             }
1215         
1216         for (i = 0; i < *num; i++)
1217         {
1218             struct record_cluster *r = reclist_read_record(se->reclist);
1219             if (!r)
1220             {
1221                 *num = i;
1222                 break;
1223             }
1224             recs[i] = r;
1225         }
1226         reclist_leave(se->reclist);
1227     }
1228 #if USE_TIMING
1229     yaz_timing_stop(t);
1230     yaz_log(YLOG_LOG, "show %6.5f %3.2f %3.2f", 
1231             yaz_timing_get_real(t), yaz_timing_get_user(t),
1232             yaz_timing_get_sys(t));
1233     yaz_timing_destroy(&t);
1234 #endif
1235     return recs;
1236 }
1237
1238 void show_range_stop(struct session *se, struct record_cluster **recs)
1239 {
1240     session_leave(se);
1241 }
1242
1243 void statistics(struct session *se, struct statistics *stat)
1244 {
1245     struct client_list *l;
1246     int count = 0;
1247
1248     memset(stat, 0, sizeof(*stat));
1249     stat->num_hits = 0;
1250     for (l = se->clients_active; l; l = l->next)
1251     {
1252         struct client *cl = l->client;
1253         if (!client_get_connection(cl))
1254             stat->num_no_connection++;
1255         stat->num_hits += client_get_hits(cl);
1256         switch (client_get_state(cl))
1257         {
1258         case Client_Connecting: stat->num_connecting++; break;
1259         case Client_Working: stat->num_working++; break;
1260         case Client_Idle: stat->num_idle++; break;
1261         case Client_Failed: stat->num_failed++; break;
1262         case Client_Error: stat->num_error++; break;
1263         default: break;
1264         }
1265         count++;
1266     }
1267     stat->num_records = se->total_records;
1268
1269     stat->num_clients = count;
1270 }
1271
1272 static struct record_metadata *record_metadata_init(
1273     NMEM nmem, const char *value, enum conf_metadata_type type,
1274     struct _xmlAttr *attr)
1275 {
1276     struct record_metadata *rec_md = record_metadata_create(nmem);
1277     struct record_metadata_attr **attrp = &rec_md->attributes;
1278     
1279     for (; attr; attr = attr->next)
1280     {
1281         if (attr->children && attr->children->content)
1282         {
1283             if (strcmp((const char *) attr->name, "type"))
1284             {  /* skip the "type" attribute.. Its value is already part of
1285                   the element in output (md-%s) and so repeating it here
1286                   is redundant */
1287                 *attrp = nmem_malloc(nmem, sizeof(**attrp));
1288                 (*attrp)->name =
1289                     nmem_strdup(nmem, (const char *) attr->name);
1290                 (*attrp)->value =
1291                     nmem_strdup(nmem, (const char *) attr->children->content);
1292                 attrp = &(*attrp)->next;
1293             }
1294         }
1295     }
1296     *attrp = 0;
1297
1298     if (type == Metadata_type_generic)
1299     {
1300         char *p = nmem_strdup(nmem, value);
1301
1302         p = normalize7bit_generic(p, " ,/.:([");
1303         
1304         rec_md->data.text.disp = p;
1305         rec_md->data.text.sort = 0;
1306     }
1307     else if (type == Metadata_type_year || type == Metadata_type_date)
1308     {
1309         int first, last;
1310         int longdate = 0;
1311
1312         if (type == Metadata_type_date)
1313             longdate = 1;
1314         if (extract7bit_dates((char *) value, &first, &last, longdate) < 0)
1315             return 0;
1316
1317         rec_md->data.number.min = first;
1318         rec_md->data.number.max = last;
1319     }
1320     else
1321         return 0;
1322     return rec_md;
1323 }
1324
1325 static int get_mergekey_from_doc(xmlDoc *doc, xmlNode *root, const char *name,
1326                                  struct conf_service *service, WRBUF norm_wr)
1327 {
1328     xmlNode *n;
1329     int no_found = 0;
1330     for (n = root->children; n; n = n->next)
1331     {
1332         if (n->type != XML_ELEMENT_NODE)
1333             continue;
1334         if (!strcmp((const char *) n->name, "metadata"))
1335         {
1336             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1337             if (type == NULL) {
1338                 yaz_log(YLOG_FATAL, "Missing type attribute on metadata element. Skipping!");
1339             }
1340             else if (!strcmp(name, (const char *) type))
1341             {
1342                 xmlChar *value = xmlNodeListGetString(doc, n->children, 1);
1343                 if (value)
1344                 {
1345                     const char *norm_str;
1346                     pp2_charset_token_t prt =
1347                         pp2_charset_token_create(service->charsets, "mergekey");
1348                     
1349                     pp2_charset_token_first(prt, (const char *) value, 0);
1350                     if (wrbuf_len(norm_wr) > 0)
1351                         wrbuf_puts(norm_wr, " ");
1352                     wrbuf_puts(norm_wr, name);
1353                     while ((norm_str =
1354                             pp2_charset_token_next(prt)))
1355                     {
1356                         if (*norm_str)
1357                         {
1358                             wrbuf_puts(norm_wr, " ");
1359                             wrbuf_puts(norm_wr, norm_str);
1360                         }
1361                     }
1362                     xmlFree(value);
1363                     pp2_charset_token_destroy(prt);
1364                     no_found++;
1365                 }
1366             }
1367             xmlFree(type);
1368         }
1369     }
1370     return no_found;
1371 }
1372
1373 static const char *get_mergekey(xmlDoc *doc, struct client *cl, int record_no,
1374                                 struct conf_service *service, NMEM nmem)
1375 {
1376     char *mergekey_norm = 0;
1377     xmlNode *root = xmlDocGetRootElement(doc);
1378     WRBUF norm_wr = wrbuf_alloc();
1379
1380     /* consider mergekey from XSL first */
1381     xmlChar *mergekey = xmlGetProp(root, (xmlChar *) "mergekey");
1382     if (mergekey)
1383     {
1384         const char *norm_str;
1385         pp2_charset_token_t prt =
1386             pp2_charset_token_create(service->charsets, "mergekey");
1387
1388         pp2_charset_token_first(prt, (const char *) mergekey, 0);
1389         while ((norm_str = pp2_charset_token_next(prt)))
1390         {
1391             if (*norm_str)
1392             {
1393                 if (wrbuf_len(norm_wr))
1394                     wrbuf_puts(norm_wr, " ");
1395                 wrbuf_puts(norm_wr, norm_str);
1396             }
1397         }
1398         pp2_charset_token_destroy(prt);
1399         xmlFree(mergekey);
1400     }
1401     else
1402     {
1403         /* no mergekey defined in XSL. Look for mergekey metadata instead */
1404         int field_id;
1405         for (field_id = 0; field_id < service->num_metadata; field_id++)
1406         {
1407             struct conf_metadata *ser_md = &service->metadata[field_id];
1408             if (ser_md->mergekey != Metadata_mergekey_no)
1409             {
1410                 int r = get_mergekey_from_doc(doc, root, ser_md->name,
1411                                               service, norm_wr);
1412                 if (r == 0 && ser_md->mergekey == Metadata_mergekey_required)
1413                 {
1414                     /* no mergekey on this one and it is required.. 
1415                        Generate unique key instead */
1416                     wrbuf_rewind(norm_wr);
1417                     break;
1418                 }
1419             }
1420         }
1421     }
1422
1423     /* generate unique key if none is not generated already or is empty */
1424     if (wrbuf_len(norm_wr) == 0)
1425     {
1426         wrbuf_printf(norm_wr, "position: %s-%d",
1427                      client_get_id(cl), record_no);
1428     }
1429     else
1430     {
1431         const char *lead = "content: ";
1432         wrbuf_insert(norm_wr, 0, lead, strlen(lead));
1433     }
1434     if (wrbuf_len(norm_wr) > 0)
1435         mergekey_norm = nmem_strdup(nmem, wrbuf_cstr(norm_wr));
1436     wrbuf_destroy(norm_wr);
1437     return mergekey_norm;
1438 }
1439
1440 /** \brief see if metadata for pz:recordfilter exists 
1441     \param root xml root element of normalized record
1442     \param sdb session database for client
1443     \retval 0 if there is no metadata for pz:recordfilter
1444     \retval 1 if there is metadata for pz:recordfilter
1445
1446     If there is no pz:recordfilter defined, this function returns 1
1447     as well.
1448 */
1449     
1450 static int check_record_filter(xmlNode *root, struct session_database *sdb)
1451 {
1452     int match = 0;
1453     xmlNode *n;
1454     const char *s;
1455     s = session_setting_oneval(sdb, PZ_RECORDFILTER);
1456
1457     if (!s || !*s)
1458         return 1;
1459
1460     for (n = root->children; n; n = n->next)
1461     {
1462         if (n->type != XML_ELEMENT_NODE)
1463             continue;
1464         if (!strcmp((const char *) n->name, "metadata"))
1465         {
1466             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1467             if (type)
1468             {
1469                 size_t len;
1470                 int substring;
1471                 const char *eq;
1472
1473                 if ((eq = strchr(s, '=')))
1474                     substring = 0;
1475                 else if ((eq = strchr(s, '~')))
1476                     substring = 1;
1477                 if (eq)
1478                     len = eq - s;
1479                 else
1480                     len = strlen(s);
1481                 if (len == strlen((const char *)type) &&
1482                     !memcmp((const char *) type, s, len))
1483                 {
1484                     xmlChar *value = xmlNodeGetContent(n);
1485                     if (value && *value)
1486                     {
1487                         if (!eq ||
1488                             (substring && strstr((const char *) value, eq+1)) ||
1489                             (!substring && !strcmp((const char *) value, eq + 1)))
1490                             match = 1;
1491                     }
1492                     xmlFree(value);
1493                 }
1494                 xmlFree(type);
1495             }
1496         }
1497     }
1498     return match;
1499 }
1500
1501
1502 static int ingest_to_cluster(struct client *cl,
1503                              xmlDoc *xdoc,
1504                              xmlNode *root,
1505                              int record_no,
1506                              const char *mergekey_norm);
1507
1508 /** \brief ingest XML record
1509     \param cl client holds the result set for record
1510     \param rec record buffer (0 terminated)
1511     \param record_no record position (1, 2, ..)
1512     \param nmem working NMEM
1513     \retval 0 OK
1514     \retval -1 failure
1515     \retval -2 Filtered
1516 */
1517 int ingest_record(struct client *cl, const char *rec,
1518                   int record_no, NMEM nmem)
1519 {
1520     struct session *se = client_get_session(cl);
1521     int ret = 0;
1522     struct session_database *sdb = client_get_database(cl);
1523     struct conf_service *service = se->service;
1524     xmlDoc *xdoc = normalize_record(se, sdb, service, rec, nmem);
1525     xmlNode *root;
1526     const char *mergekey_norm;
1527     
1528     if (!xdoc)
1529         return -1;
1530     
1531     root = xmlDocGetRootElement(xdoc);
1532     
1533     if (!check_record_filter(root, sdb))
1534     {
1535         session_log(se, YLOG_LOG, "Filtered out record no %d from %s",
1536                     record_no, sdb->database->id);
1537         xmlFreeDoc(xdoc);
1538         return -2;
1539     }
1540     
1541     mergekey_norm = get_mergekey(xdoc, cl, record_no, service, nmem);
1542     if (!mergekey_norm)
1543     {
1544         session_log(se, YLOG_WARN, "Got no mergekey");
1545         xmlFreeDoc(xdoc);
1546         return -1;
1547     }
1548     session_enter(se);
1549     if (client_get_session(cl) == se)
1550         ret = ingest_to_cluster(cl, xdoc, root, record_no, mergekey_norm);
1551     session_leave(se);
1552     
1553     xmlFreeDoc(xdoc);
1554     return ret;
1555 }
1556
1557 static int check_limit_local(struct client *cl,
1558                              struct record *record,
1559                              int record_no)
1560 {
1561     int skip_record = 0;
1562     struct session *se = client_get_session(cl);
1563     struct conf_service *service = se->service;
1564     NMEM nmem_tmp = nmem_create();
1565     struct session_database *sdb = client_get_database(cl);
1566     int l = 0;
1567     while (!skip_record)
1568     {
1569         struct conf_metadata *ser_md = 0;
1570         struct record_metadata *rec_md = 0;
1571         int md_field_id;
1572         char **values = 0;
1573         int i, num_v = 0;
1574         
1575         const char *name =
1576             client_get_facet_limit_local(cl, sdb, &l, nmem_tmp, &num_v,
1577                                          &values);
1578         if (!name)
1579             break;
1580         
1581         md_field_id = conf_service_metadata_field_id(service, name);
1582         if (md_field_id < 0)
1583         {
1584             skip_record = 1;
1585             break;
1586         }
1587         ser_md = &service->metadata[md_field_id];
1588         rec_md = record->metadata[md_field_id];
1589         yaz_log(YLOG_LOG, "check limit local %s", name);
1590         for (i = 0; i < num_v; )
1591         {
1592             if (rec_md)
1593             {
1594                 if (ser_md->type == Metadata_type_year 
1595                     || ser_md->type == Metadata_type_date)
1596                 {
1597                     int y = atoi(values[i]);
1598                     if (y >= rec_md->data.number.min 
1599                         && y <= rec_md->data.number.max)
1600                         break;
1601                 }
1602                 else
1603                 {
1604                     yaz_log(YLOG_LOG, "cmp: '%s' '%s'",
1605                             rec_md->data.text.disp, values[i]);
1606                     if (!strcmp(rec_md->data.text.disp, values[i]))
1607                     {
1608                         break;
1609                     }
1610                 }
1611                 rec_md = rec_md->next;
1612             }
1613             else
1614             {
1615                 rec_md = record->metadata[md_field_id];
1616                 i++;
1617             }
1618         }
1619         if (i == num_v)
1620         {
1621             skip_record = 1;
1622             break;
1623         }
1624     }
1625     nmem_destroy(nmem_tmp);
1626     return skip_record;
1627 }
1628                              
1629 static int ingest_to_cluster(struct client *cl,
1630                              xmlDoc *xdoc,
1631                              xmlNode *root,
1632                              int record_no,
1633                              const char *mergekey_norm)
1634 {
1635     xmlNode *n;
1636     xmlChar *type = 0;
1637     xmlChar *value = 0;
1638     struct session *se = client_get_session(cl);
1639     struct conf_service *service = se->service;
1640     int term_factor = 1;
1641     struct record_cluster *cluster;
1642     struct session_database *sdb = client_get_database(cl);
1643     struct record *record = record_create(se->nmem, 
1644                                           service->num_metadata,
1645                                           service->num_sortkeys, cl,
1646                                           record_no);
1647
1648     for (n = root->children; n; n = n->next)
1649     {
1650         if (type)
1651             xmlFree(type);
1652         if (value)
1653             xmlFree(value);
1654         type = value = 0;
1655         
1656         if (n->type != XML_ELEMENT_NODE)
1657             continue;
1658         if (!strcmp((const char *) n->name, "metadata"))
1659         {
1660             struct conf_metadata *ser_md = 0;
1661             struct record_metadata **wheretoput = 0;
1662             struct record_metadata *rec_md = 0;
1663             int md_field_id = -1;
1664             
1665             type = xmlGetProp(n, (xmlChar *) "type");
1666             value = xmlNodeListGetString(xdoc, n->children, 1);
1667             
1668             if (!type || !value || !*value)
1669                 continue;
1670             
1671             md_field_id 
1672                 = conf_service_metadata_field_id(service, (const char *) type);
1673             if (md_field_id < 0)
1674             {
1675                 if (se->number_of_warnings_unknown_metadata == 0)
1676                 {
1677                     session_log(se, YLOG_WARN, 
1678                             "Ignoring unknown metadata element: %s", type);
1679                 }
1680                 se->number_of_warnings_unknown_metadata++;
1681                 continue;
1682             }
1683            
1684             ser_md = &service->metadata[md_field_id];
1685
1686             // non-merged metadata
1687             rec_md = record_metadata_init(se->nmem, (const char *) value,
1688                                           ser_md->type, n->properties);
1689             if (!rec_md)
1690             {
1691                 session_log(se, YLOG_WARN, "bad metadata data '%s' "
1692                             "for element '%s'", value, type);
1693                 continue;
1694             }
1695             wheretoput = &record->metadata[md_field_id];
1696             while (*wheretoput)
1697                 wheretoput = &(*wheretoput)->next;
1698             *wheretoput = rec_md;
1699         }
1700     }
1701
1702     if (check_limit_local(cl, record, record_no))
1703     {
1704         session_log(se, YLOG_LOG, "Facet filtered out record no %d from %s",
1705                     record_no, sdb->database->id);
1706         if (type)
1707             xmlFree(type);
1708         if (value)
1709             xmlFree(value);
1710         return -2;
1711     }
1712     cluster = reclist_insert(se->reclist, service, record,
1713                              mergekey_norm, &se->total_merged);
1714     if (!cluster)
1715         return -1;
1716
1717     {
1718         const char *use_term_factor_str =
1719             session_setting_oneval(sdb, PZ_TERMLIST_TERM_FACTOR);
1720         if (use_term_factor_str && use_term_factor_str[0] == '1')
1721         {
1722             int maxrecs = client_get_maxrecs(cl);
1723             int hits = (int) client_get_hits(cl);
1724             term_factor = MAX(hits, maxrecs) /  MAX(1, maxrecs);
1725             assert(term_factor >= 1);
1726             yaz_log(YLOG_DEBUG, "Using term factor: %d (%d / %d)", term_factor, MAX(hits, maxrecs), MAX(1, maxrecs));
1727         }
1728     }
1729
1730     if (global_parameters.dump_records)
1731         session_log(se, YLOG_LOG, "Cluster id %s from %s (#%d)", cluster->recid,
1732                     sdb->database->id, record_no);
1733
1734
1735     relevance_newrec(se->relevance, cluster);
1736     
1737     // now parsing XML record and adding data to cluster or record metadata
1738     for (n = root->children; n; n = n->next)
1739     {
1740         pp2_charset_token_t prt;
1741         if (type)
1742             xmlFree(type);
1743         if (value)
1744             xmlFree(value);
1745         type = value = 0;
1746         
1747         if (n->type != XML_ELEMENT_NODE)
1748             continue;
1749         if (!strcmp((const char *) n->name, "metadata"))
1750         {
1751             struct conf_metadata *ser_md = 0;
1752             struct conf_sortkey *ser_sk = 0;
1753             struct record_metadata **wheretoput = 0;
1754             struct record_metadata *rec_md = 0;
1755             int md_field_id = -1;
1756             int sk_field_id = -1;
1757             
1758             type = xmlGetProp(n, (xmlChar *) "type");
1759             value = xmlNodeListGetString(xdoc, n->children, 1);
1760             
1761             if (!type || !value || !*value)
1762                 continue;
1763             
1764             md_field_id 
1765                 = conf_service_metadata_field_id(service, (const char *) type);
1766             if (md_field_id < 0)
1767                 continue;
1768             
1769             ser_md = &service->metadata[md_field_id];
1770             
1771             if (ser_md->sortkey_offset >= 0)
1772             {
1773                 sk_field_id = ser_md->sortkey_offset;
1774                 ser_sk = &service->sortkeys[sk_field_id];
1775             }
1776
1777             // merged metadata
1778             rec_md = record_metadata_init(se->nmem, (const char *) value,
1779                                           ser_md->type, 0);
1780             if (!rec_md)
1781                 continue;
1782
1783             wheretoput = &cluster->metadata[md_field_id];
1784
1785             // and polulate with data:
1786             // assign cluster or record based on merge action
1787             if (ser_md->merge == Metadata_merge_unique)
1788             {
1789                 while (*wheretoput)
1790                 {
1791                     if (!strcmp((const char *) (*wheretoput)->data.text.disp, 
1792                                 rec_md->data.text.disp))
1793                         break;
1794                     wheretoput = &(*wheretoput)->next;
1795                 }
1796                 if (!*wheretoput)
1797                     *wheretoput = rec_md;
1798             }
1799             else if (ser_md->merge == Metadata_merge_longest)
1800             {
1801                 if (!*wheretoput 
1802                     || strlen(rec_md->data.text.disp) 
1803                     > strlen((*wheretoput)->data.text.disp))
1804                 {
1805                     *wheretoput = rec_md;
1806                     if (ser_sk)
1807                     {
1808                         const char *sort_str = 0;
1809                         int skip_article = 
1810                             ser_sk->type == Metadata_sortkey_skiparticle;
1811
1812                         if (!cluster->sortkeys[sk_field_id])
1813                             cluster->sortkeys[sk_field_id] = 
1814                                 nmem_malloc(se->nmem, 
1815                                             sizeof(union data_types));
1816                          
1817                         prt =
1818                             pp2_charset_token_create(service->charsets, "sort");
1819
1820                         pp2_charset_token_first(prt, rec_md->data.text.disp,
1821                                                 skip_article);
1822
1823                         pp2_charset_token_next(prt);
1824                          
1825                         sort_str = pp2_get_sort(prt);
1826                          
1827                         cluster->sortkeys[sk_field_id]->text.disp = 
1828                             rec_md->data.text.disp;
1829                         if (!sort_str)
1830                         {
1831                             sort_str = rec_md->data.text.disp;
1832                             session_log(se, YLOG_WARN, 
1833                                     "Could not make sortkey. Bug #1858");
1834                         }
1835                         cluster->sortkeys[sk_field_id]->text.sort = 
1836                             nmem_strdup(se->nmem, sort_str);
1837                         pp2_charset_token_destroy(prt);
1838                     }
1839                 }
1840             }
1841             else if (ser_md->merge == Metadata_merge_all)
1842             {
1843                 while (*wheretoput)
1844                     wheretoput = &(*wheretoput)->next;
1845                 *wheretoput = rec_md;
1846             }
1847             else if (ser_md->merge == Metadata_merge_range)
1848             {
1849                 if (!*wheretoput)
1850                 {
1851                     *wheretoput = rec_md;
1852                     if (ser_sk)
1853                         cluster->sortkeys[sk_field_id] 
1854                             = &rec_md->data;
1855                 }
1856                 else
1857                 {
1858                     int this_min = rec_md->data.number.min;
1859                     int this_max = rec_md->data.number.max;
1860                     if (this_min < (*wheretoput)->data.number.min)
1861                         (*wheretoput)->data.number.min = this_min;
1862                     if (this_max > (*wheretoput)->data.number.max)
1863                         (*wheretoput)->data.number.max = this_max;
1864                 }
1865             }
1866
1867
1868             // ranking of _all_ fields enabled ... 
1869             if (ser_md->rank)
1870                 relevance_countwords(se->relevance, cluster, 
1871                                      (char *) value, ser_md->rank,
1872                                      ser_md->name);
1873
1874             // construct facets ... unless the client already has reported them
1875             if (ser_md->termlist && !client_has_facet(cl, (char *) type))
1876             {
1877                 if (ser_md->type == Metadata_type_year)
1878                 {
1879                     char year[64];
1880                     sprintf(year, "%d", rec_md->data.number.max);
1881
1882                     add_facet(se, (char *) type, year, term_factor);
1883                     if (rec_md->data.number.max != rec_md->data.number.min)
1884                     {
1885                         sprintf(year, "%d", rec_md->data.number.min);
1886                         add_facet(se, (char *) type, year, term_factor);
1887                     }
1888                 }
1889                 else
1890                     add_facet(se, (char *) type, (char *) value, term_factor);
1891             }
1892
1893             // cleaning up
1894             xmlFree(type);
1895             xmlFree(value);
1896             type = value = 0;
1897         }
1898         else
1899         {
1900             if (se->number_of_warnings_unknown_elements == 0)
1901                 session_log(se, YLOG_WARN,
1902                         "Unexpected element in internal record: %s", n->name);
1903             se->number_of_warnings_unknown_elements++;
1904         }
1905     }
1906     if (type)
1907         xmlFree(type);
1908     if (value)
1909         xmlFree(value);
1910
1911     relevance_donerecord(se->relevance, cluster);
1912     se->total_records++;
1913
1914     return 0;
1915 }
1916
1917 void session_log(struct session *s, int level, const char *fmt, ...)
1918 {
1919     char buf[1024];
1920     va_list ap;
1921     va_start(ap, fmt);
1922
1923     yaz_vsnprintf(buf, sizeof(buf)-30, fmt, ap);
1924     yaz_log(level, "Session %u: %s", s->session_id, buf);
1925
1926     va_end(ap);
1927 }
1928
1929 /*
1930  * Local variables:
1931  * c-basic-offset: 4
1932  * c-file-style: "Stroustrup"
1933  * indent-tabs-mode: nil
1934  * End:
1935  * vim: shiftwidth=4 tabstop=8 expandtab
1936  */
1937