0d463d6856644a5b5bdc4de2674252e302d7a26e
[pazpar2-moved-to-github.git] / src / session.c
1 /* This file is part of Pazpar2.
2    Copyright (C) 2006-2012 Index Data
3
4 Pazpar2 is free software; you can redistribute it and/or modify it under
5 the terms of the GNU General Public License as published by the Free
6 Software Foundation; either version 2, or (at your option) any later
7 version.
8
9 Pazpar2 is distributed in the hope that it will be useful, but WITHOUT ANY
10 WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
12 for more details.
13
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17
18 */
19
20 /** \file session.c
21     \brief high-level logic; mostly user sessions and settings
22 */
23
24 #if HAVE_CONFIG_H
25 #include <config.h>
26 #endif
27
28 #include <time.h>
29 #include <stdlib.h>
30 #include <stdio.h>
31 #include <string.h>
32 #if HAVE_SYS_TIME_H
33 #include <sys/time.h>
34 #endif
35 #if HAVE_UNISTD_H
36 #include <unistd.h>
37 #endif
38 #ifdef WIN32
39 #include <windows.h>
40 #endif
41 #include <signal.h>
42 #include <ctype.h>
43 #include <assert.h>
44 #include <math.h>
45
46 #include <yaz/marcdisp.h>
47 #include <yaz/comstack.h>
48 #include <yaz/tcpip.h>
49 #include <yaz/proto.h>
50 #include <yaz/readconf.h>
51 #include <yaz/pquery.h>
52 #include <yaz/otherinfo.h>
53 #include <yaz/yaz-util.h>
54 #include <yaz/nmem.h>
55 #include <yaz/query-charset.h>
56 #include <yaz/querytowrbuf.h>
57 #include <yaz/oid_db.h>
58 #include <yaz/snprintf.h>
59 #include <yaz/gettimeofday.h>
60
61 #define USE_TIMING 0
62 #if USE_TIMING
63 #include <yaz/timing.h>
64 #endif
65
66 #include "ppmutex.h"
67 #include "parameters.h"
68 #include "session.h"
69 #include "eventl.h"
70 #include "http.h"
71 #include "termlists.h"
72 #include "reclists.h"
73 #include "relevance.h"
74 #include "database.h"
75 #include "client.h"
76 #include "settings.h"
77 #include "normalize7bit.h"
78
79 #include <libxml/tree.h>
80
81 #define 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 = termlist_create(s->nmem);
233             s->num_termlists = i + 1;
234         }
235         
236 #if 0
237         session_log(s, YLOG_LOG, "Facets for %s: %s norm:%s (%d)", type, value, wrbuf_cstr(facet_wrbuf), count);
238 #endif
239         termlist_insert(s->termlists[i].termlist, wrbuf_cstr(display_wrbuf),
240                         wrbuf_cstr(facet_wrbuf), count);
241     }
242     wrbuf_destroy(facet_wrbuf);
243     wrbuf_destroy(display_wrbuf);
244 }
245
246 static xmlDoc *record_to_xml(struct session *se,
247                              struct session_database *sdb, const char *rec)
248 {
249     struct database *db = sdb->database;
250     xmlDoc *rdoc = 0;
251
252     rdoc = xmlParseMemory(rec, strlen(rec));
253
254     if (!rdoc)
255     {
256         session_log(se, YLOG_FATAL, "Non-wellformed XML received from %s",
257                     db->id);
258         return 0;
259     }
260
261     if (global_parameters.dump_records)
262     {
263         session_log(se, YLOG_LOG, "Un-normalized record from %s", db->id);
264         log_xml_doc(rdoc);
265     }
266
267     return rdoc;
268 }
269
270 #define MAX_XSLT_ARGS 16
271
272 // Add static values from session database settings if applicable
273 static void insert_settings_parameters(struct session_database *sdb,
274                                        struct conf_service *service,
275                                        char **parms,
276                                        NMEM nmem)
277 {
278     int i;
279     int nparms = 0;
280     int offset = 0;
281
282     for (i = 0; i < service->num_metadata; i++)
283     {
284         struct conf_metadata *md = &service->metadata[i];
285         int setting;
286
287         if (md->setting == Metadata_setting_parameter &&
288             (setting = settings_lookup_offset(service, md->name)) >= 0)
289         {
290             const char *val = session_setting_oneval(sdb, setting);
291             if (val && nparms < MAX_XSLT_ARGS)
292             {
293                 char *buf;
294                 int len = strlen(val);
295                 buf = nmem_malloc(nmem, len + 3);
296                 buf[0] = '\'';
297                 strcpy(buf + 1, val);
298                 buf[len+1] = '\'';
299                 buf[len+2] = '\0';
300                 parms[offset++] = md->name;
301                 parms[offset++] = buf;
302                 nparms++;
303             }
304         }
305     }
306     parms[offset] = 0;
307 }
308
309 // Add static values from session database settings if applicable
310 static void insert_settings_values(struct session_database *sdb, xmlDoc *doc,
311     struct conf_service *service)
312 {
313     int i;
314
315     for (i = 0; i < service->num_metadata; i++)
316     {
317         struct conf_metadata *md = &service->metadata[i];
318         int offset;
319
320         if (md->setting == Metadata_setting_postproc &&
321             (offset = settings_lookup_offset(service, md->name)) >= 0)
322         {
323             const char *val = session_setting_oneval(sdb, offset);
324             if (val)
325             {
326                 xmlNode *r = xmlDocGetRootElement(doc);
327                 xmlNode *n = xmlNewTextChild(r, 0, (xmlChar *) "metadata",
328                                              (xmlChar *) val);
329                 xmlSetProp(n, (xmlChar *) "type", (xmlChar *) md->name);
330             }
331         }
332     }
333 }
334
335 static xmlDoc *normalize_record(struct session *se,
336                                 struct session_database *sdb,
337                                 struct conf_service *service,
338                                 const char *rec, NMEM nmem)
339 {
340     xmlDoc *rdoc = record_to_xml(se, sdb, rec);
341
342     if (rdoc)
343     {
344         char *parms[MAX_XSLT_ARGS*2+1];
345         
346         insert_settings_parameters(sdb, service, parms, nmem);
347         
348         if (normalize_record_transform(sdb->map, &rdoc, (const char **)parms))
349         {
350             session_log(se, YLOG_WARN, "Normalize failed from %s",
351                         sdb->database->id);
352         }
353         else
354         {
355             insert_settings_values(sdb, rdoc, service);
356             
357             if (global_parameters.dump_records)
358             {
359                 session_log(se, YLOG_LOG, "Normalized record from %s", 
360                             sdb->database->id);
361                 log_xml_doc(rdoc);
362             }
363         }
364     }
365     return rdoc;
366 }
367
368 void session_settings_dump(struct session *se,
369                            struct session_database *db,
370                            WRBUF w)
371 {
372     if (db->settings)
373     {
374         int i, num = db->num_settings;
375         for (i = 0; i < num; i++)
376         {
377             struct setting *s = db->settings[i];
378             for (;s ; s = s->next)
379             {
380                 wrbuf_puts(w, "<set name=\"");
381                 wrbuf_xmlputs(w, s->name);
382                 wrbuf_puts(w, "\" value=\"");
383                 wrbuf_xmlputs(w, s->value);
384                 wrbuf_puts(w, "\"/>");
385             }
386             if (db->settings[i])
387                 wrbuf_puts(w, "\n");
388         }
389     }
390 }
391
392 // Retrieve first defined value for 'name' for given database.
393 // Will be extended to take into account user associated with session
394 const char *session_setting_oneval(struct session_database *db, int offset)
395 {
396     if (offset >= db->num_settings || !db->settings[offset])
397         return "";
398     return db->settings[offset]->value;
399 }
400
401 // Prepare XSLT stylesheets for record normalization
402 // Structures are allocated on the session_wide nmem to avoid having
403 // to recompute this for every search. This would lead
404 // to leaking if a single session was to repeatedly change the PZ_XSLT
405 // setting. However, this is not a realistic use scenario.
406 static int prepare_map(struct session *se, struct session_database *sdb)
407 {
408     if (sdb->settings && !sdb->map)
409     {
410         const char *s;
411
412         if (sdb->settings[PZ_XSLT] &&
413             (s = session_setting_oneval(sdb, PZ_XSLT)))        
414         {
415             char auto_stylesheet[256];
416             
417             if (!strcmp(s, "auto"))
418             {
419                 const char *request_syntax = session_setting_oneval(
420                     sdb, PZ_REQUESTSYNTAX);
421                 if (request_syntax)
422                 {
423                     char *cp;
424                     yaz_snprintf(auto_stylesheet, sizeof(auto_stylesheet),
425                                  "%s.xsl", request_syntax);
426                     for (cp = auto_stylesheet; *cp; cp++)
427                     {
428                         /* deliberately only consider ASCII */
429                         if (*cp > 32 && *cp < 127)
430                             *cp = tolower(*cp);
431                     }
432                     s = auto_stylesheet;
433                 }
434                 else
435                 {
436                     session_log(se, YLOG_WARN,
437                                 "No pz:requestsyntax for auto stylesheet");
438                 }
439             }
440             sdb->map = normalize_cache_get(se->normalize_cache,
441                                            se->service, s);
442             if (!sdb->map)
443                 return -1;
444         }
445     }
446     return 0;
447 }
448
449 // called if watch should be removed because http_channel is to be destroyed
450 static void session_watch_cancel(void *data, struct http_channel *c,
451                                  void *data2)
452 {
453     struct session_watchentry *ent = data;
454
455     ent->fun = 0;
456     ent->data = 0;
457     ent->obs = 0;
458 }
459
460 // set watch. Returns 0=OK, -1 if watch is already set
461 int session_set_watch(struct session *s, int what, 
462                       session_watchfun fun, void *data,
463                       struct http_channel *chan)
464 {
465     int ret;
466     session_enter(s);
467     if (s->watchlist[what].fun)
468         ret = -1;
469     else
470     {
471         
472         s->watchlist[what].fun = fun;
473         s->watchlist[what].data = data;
474         s->watchlist[what].obs = http_add_observer(chan, &s->watchlist[what],
475                                                    session_watch_cancel);
476         ret = 0;
477     }
478     session_leave(s);
479     return ret;
480 }
481
482 void session_alert_watch(struct session *s, int what)
483 {
484     assert(s);
485     session_enter(s);
486     if (s->watchlist[what].fun)
487     {
488         /* our watch is no longer associated with http_channel */
489         void *data;
490         session_watchfun fun;
491
492         http_remove_observer(s->watchlist[what].obs);
493         fun  = s->watchlist[what].fun;
494         data = s->watchlist[what].data;
495
496         /* reset watch before fun is invoked - in case fun wants to set
497            it again */
498         s->watchlist[what].fun = 0;
499         s->watchlist[what].data = 0;
500         s->watchlist[what].obs = 0;
501
502         session_leave(s);
503         session_log(s, YLOG_DEBUG,
504                     "Alert Watch: %d calling function: %p", what, fun);
505         fun(data);
506     }
507     else
508         session_leave(s);
509 }
510
511 //callback for grep_databases
512 static void select_targets_callback(struct session *se,
513                                     struct session_database *db)
514 {
515     struct client *cl;
516     struct client_list *l;
517
518     for (l = se->clients_cached; l; l = l->next)
519         if (client_get_database(l->client) == db)
520             break;
521
522     if (l)
523         cl = l->client;
524     else
525     {
526         cl = client_create(db->database->id);
527         client_set_database(cl, db);
528
529         l = xmalloc(sizeof(*l));
530         l->client = cl;
531         l->next = se->clients_cached;
532         se->clients_cached = l;
533     }
534     /* set session always. If may be 0 if client is not active */
535     client_set_session(cl, se);
536
537     l = xmalloc(sizeof(*l));
538     l->client = cl;
539     l->next = se->clients_active;
540     se->clients_active = l;
541 }
542
543 static void session_reset_active_clients(struct session *se,
544                                          struct client_list *new_list)
545 {
546     struct client_list *l;
547
548     session_enter(se);
549     l = se->clients_active;
550     se->clients_active = new_list;
551     session_leave(se);
552
553     while (l)
554     {
555         struct client_list *l_next = l->next;
556
557         client_lock(l->client); 
558         client_set_session(l->client, 0); /* mark client inactive */
559         client_unlock(l->client);
560
561         xfree(l);
562         l = l_next;
563     }
564 }
565
566 static void session_remove_cached_clients(struct session *se)
567 {
568     struct client_list *l;
569
570     session_reset_active_clients(se, 0);
571
572     session_enter(se);
573     l = se->clients_cached;
574     se->clients_cached = 0;
575     session_leave(se);
576
577     while (l)
578     {
579         struct client_list *l_next = l->next;
580         client_lock(l->client);
581         client_set_session(l->client, 0);
582         client_set_database(l->client, 0);
583         client_unlock(l->client);
584         client_destroy(l->client);
585         xfree(l);
586         l = l_next;
587     }
588 }
589
590 // Associates a set of clients with a session;
591 // Note: Session-databases represent databases with per-session 
592 // setting overrides
593 static int select_targets(struct session *se, const char *filter)
594 {
595     return session_grep_databases(se, filter, select_targets_callback);
596 }
597
598 int session_active_clients(struct session *s)
599 {
600     struct client_list *l;
601     int res = 0;
602
603     for (l = s->clients_active; l; l = l->next)
604         if (client_is_active(l->client))
605             res++;
606
607     return res;
608 }
609
610 int session_is_preferred_clients_ready(struct session *s)
611 {
612     struct client_list *l;
613     int res = 0;
614
615     for (l = s->clients_active; l; l = l->next)
616         if (client_is_active_preferred(l->client))
617             res++;
618     session_log(s, YLOG_DEBUG, "Has %d active preferred clients.", res);
619     return res == 0;
620 }
621
622 static void session_clear_set(struct session *se,
623                               const char *sort_field, int increasing)
624 {
625     reclist_destroy(se->reclist);
626     se->reclist = 0;
627     if (nmem_total(se->nmem))
628         session_log(se, YLOG_DEBUG, "NMEN operation usage %zd",
629                     nmem_total(se->nmem));
630     nmem_reset(se->nmem);
631     se->total_records = se->total_merged = 0;
632     se->num_termlists = 0;
633     
634     /* reset list of sorted results and clear to relevance search */
635     se->sorted_results = nmem_malloc(se->nmem, sizeof(*se->sorted_results));
636     se->sorted_results->field = nmem_strdup(se->nmem, sort_field);
637     se->sorted_results->increasing = increasing;
638     se->sorted_results->next = 0;
639     
640     se->reclist = reclist_create(se->nmem);
641 }
642
643 void session_sort(struct session *se, const char *field, int increasing, int clear_set)
644 {
645     struct session_sorted_results *sr;
646     struct client_list *l;
647
648     session_enter(se);
649
650     yaz_log(YLOG_LOG, "session_sort field=%s", field);
651     // TODO In order for this to work, clear_set may only be true on first call. Every following (poll) may not. 
652     // I do not think we can decide this from the outside of the session. 
653     // The logic should be when we change to/away from a native sort order, 
654     // it should cleared on the first call
655     if (clear_set)
656     {
657         session_clear_set(se, field, increasing);
658     }
659     else
660     {
661         /* see if we already have sorted for this critieria */
662         for (sr = se->sorted_results; sr; sr = sr->next)
663         {
664             if (!strcmp(field, sr->field) && increasing == sr->increasing)
665                 break;
666         }
667         if (sr)
668         {
669             session_log(se, YLOG_DEBUG, "search_sort: field=%s increasing=%d already fetched",
670                         field, increasing);
671             session_leave(se);
672             return;
673         }
674         session_log(se, YLOG_DEBUG, "search_sort: field=%s increasing=%d must fetch",
675                     field, increasing);
676         sr = nmem_malloc(se->nmem, sizeof(*sr));
677         sr->field = nmem_strdup(se->nmem, field);
678         sr->increasing = increasing;
679         sr->next = se->sorted_results;
680         se->sorted_results = sr;
681     }
682         
683     for (l = se->clients_active; l; l = l->next)
684     {
685         struct client *cl = l->client;
686         if (client_get_state(cl) == Client_Connecting ||
687             client_get_state(cl) == Client_Idle ||
688             client_get_state(cl) == Client_Working)
689             client_start_search(cl);
690     }
691     session_leave(se);
692 }
693
694 enum pazpar2_error_code session_search(struct session *se,
695                                        const char *query,
696                                        const char *startrecs,
697                                        const char *maxrecs,
698                                        const char *filter,
699                                        const char *limit,
700                                        const char **addinfo,
701                                        const char *sort_field, int increasing)
702 {
703     int live_channels = 0;
704     int no_working = 0;
705     int no_failed_query = 0;
706     int no_failed_limit = 0;
707     struct client_list *l, *l0;
708     struct timeval tval;
709     facet_limits_t facet_limits;
710
711     session_log(se, YLOG_DEBUG, "Search");
712
713     *addinfo = 0;
714
715     if (se->settings_modified)
716         session_remove_cached_clients(se);
717     else
718         session_reset_active_clients(se, 0);
719     
720     session_enter(se);
721     se->settings_modified = 0;
722     session_clear_set(se, sort_field, increasing);
723     relevance_destroy(&se->relevance);
724
725     live_channels = select_targets(se, filter);
726     if (!live_channels)
727     {
728         session_leave(se);
729         return PAZPAR2_NO_TARGETS;
730     }
731
732     yaz_gettimeofday(&tval);
733     
734     tval.tv_sec += 5;
735
736     facet_limits = facet_limits_create(limit);
737     if (!facet_limits)
738     {
739         *addinfo = "limit";
740         session_leave(se);
741         return PAZPAR2_MALFORMED_PARAMETER_VALUE;
742     }
743
744     l0 = se->clients_active;
745     se->clients_active = 0;
746     session_leave(se);
747
748     for (l = l0; l; l = l->next)
749     {
750         int parse_ret;
751         struct client *cl = l->client;
752
753         if (prepare_map(se, client_get_database(cl)) < 0)
754             continue;
755
756         parse_ret = client_parse_query(cl, query, facet_limits, startrecs,
757                                        maxrecs, se->service->ccl_bibset);
758         if (parse_ret == -1)
759             no_failed_query++;
760         else if (parse_ret == -2)
761             no_failed_limit++;
762         else
763         {
764             int r =
765                 client_prep_connection(cl, se->service->z3950_operation_timeout,
766                                        se->service->z3950_session_timeout,
767                                        se->service->server->iochan_man,
768                                        &tval);
769             if (parse_ret == 1 && r == 2)
770             {
771                 session_log(se, YLOG_LOG, "client %s REUSE result", client_get_id(cl));
772                 client_reingest(cl);
773             }
774             else if (r)
775             {
776                 session_log(se, YLOG_LOG, "client %s NEW search", client_get_id(cl));
777                 client_start_search(cl);
778             }
779             no_working++;
780         }
781     }
782     facet_limits_destroy(facet_limits);
783     session_reset_active_clients(se, l0);
784
785     if (no_working == 0)
786     {
787         if (no_failed_query > 0)
788         {
789             *addinfo = "query";
790             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
791         }
792         else if (no_failed_limit > 0)
793         {
794             *addinfo = "limit";
795             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
796         }
797         else
798             return PAZPAR2_NO_TARGETS;
799     }
800     session_log(se, YLOG_LOG, "session_start_search done");
801     return PAZPAR2_NO_ERROR;
802 }
803
804 // Creates a new session_database object for a database
805 static void session_init_databases_fun(void *context, struct database *db)
806 {
807     struct session *se = (struct session *) context;
808     struct session_database *new = nmem_malloc(se->session_nmem, sizeof(*new));
809     int i;
810
811     new->database = db;
812     
813     new->map = 0;
814     assert(db->settings);
815     new->settings = nmem_malloc(se->session_nmem,
816                                 sizeof(struct settings *) * db->num_settings);
817     new->num_settings = db->num_settings;
818     for (i = 0; i < db->num_settings; i++)
819     {
820         struct setting *setting = db->settings[i];
821         new->settings[i] = setting;
822     }
823     new->next = se->databases;
824     se->databases = new;
825 }
826
827 // Doesn't free memory associated with sdb -- nmem takes care of that
828 static void session_database_destroy(struct session_database *sdb)
829 {
830     sdb->map = 0;
831 }
832
833 // Initialize session_database list -- this represents this session's view
834 // of the database list -- subject to modification by the settings ws command
835 void session_init_databases(struct session *se)
836 {
837     se->databases = 0;
838     predef_grep_databases(se, se->service, session_init_databases_fun);
839 }
840
841 // Probably session_init_databases_fun should be refactored instead of
842 // called here.
843 static struct session_database *load_session_database(struct session *se, 
844                                                       char *id)
845 {
846     struct database *db = new_database_inherit_settings(id, se->session_nmem, se->service->settings);
847     session_init_databases_fun((void*) se, db);
848
849     // New sdb is head of se->databases list
850     return se->databases;
851 }
852
853 // Find an existing session database. If not found, load it
854 static struct session_database *find_session_database(struct session *se, 
855                                                       char *id)
856 {
857     struct session_database *sdb;
858
859     for (sdb = se->databases; sdb; sdb = sdb->next)
860         if (!strcmp(sdb->database->id, id))
861             return sdb;
862     return load_session_database(se, id);
863 }
864
865 // Apply a session override to a database
866 void session_apply_setting(struct session *se, char *dbname, char *setting,
867                            char *value)
868 {
869     struct session_database *sdb = find_session_database(se, dbname);
870     struct conf_service *service = se->service;
871     struct setting *new = nmem_malloc(se->session_nmem, sizeof(*new));
872     int offset = settings_create_offset(service, setting);
873
874     expand_settings_array(&sdb->settings, &sdb->num_settings, offset,
875                           se->session_nmem);
876     new->precedence = 0;
877     new->target = dbname;
878     new->name = setting;
879     new->value = value;
880     new->next = sdb->settings[offset];
881     sdb->settings[offset] = new;
882
883     se->settings_modified = 1;
884
885     // Force later recompute of settings-driven data structures
886     // (happens when a search starts and client connections are prepared)
887     switch (offset)
888     {
889     case PZ_XSLT:
890         if (sdb->map)
891         {
892             sdb->map = 0;
893         }
894         break;
895     }
896 }
897
898 void session_destroy(struct session *se)
899 {
900     struct session_database *sdb;
901     session_log(se, YLOG_DEBUG, "Destroying");
902     session_use(-1);
903     session_remove_cached_clients(se);
904
905     for (sdb = se->databases; sdb; sdb = sdb->next)
906         session_database_destroy(sdb);
907     normalize_cache_destroy(se->normalize_cache);
908     relevance_destroy(&se->relevance);
909     reclist_destroy(se->reclist);
910     if (nmem_total(se->nmem))
911         session_log(se, YLOG_DEBUG, "NMEN operation usage %zd", nmem_total(se->nmem));
912     if (nmem_total(se->session_nmem))
913         session_log(se, YLOG_DEBUG, "NMEN session usage %zd", nmem_total(se->session_nmem));
914     nmem_destroy(se->nmem);
915     service_destroy(se->service);
916     yaz_mutex_destroy(&se->session_mutex);
917 }
918
919 size_t session_get_memory_status(struct session *session) {
920     size_t session_nmem;
921     if (session == 0)
922         return 0;
923     session_enter(session);
924     session_nmem = nmem_total(session->nmem);
925     session_leave(session);
926     return session_nmem;
927 }
928
929
930 struct session *new_session(NMEM nmem, struct conf_service *service,
931                             unsigned session_id)
932 {
933     int i;
934     struct session *session = nmem_malloc(nmem, sizeof(*session));
935
936     char tmp_str[50];
937
938     sprintf(tmp_str, "session#%u", session_id);
939
940     session->session_id = session_id;
941     session_log(session, YLOG_DEBUG, "New");
942     session->service = service;
943     session->relevance = 0;
944     session->total_records = 0;
945     session->number_of_warnings_unknown_elements = 0;
946     session->number_of_warnings_unknown_metadata = 0;
947     session->num_termlists = 0;
948     session->reclist = 0;
949     session->clients_active = 0;
950     session->clients_cached = 0;
951     session->settings_modified = 0;
952     session->session_nmem = nmem;
953     session->nmem = nmem_create();
954     session->databases = 0;
955     for (i = 0; i <= SESSION_WATCH_MAX; i++)
956     {
957         session->watchlist[i].data = 0;
958         session->watchlist[i].fun = 0;
959     }
960     session->normalize_cache = normalize_cache_create();
961     session->session_mutex = 0;
962     pazpar2_mutex_create(&session->session_mutex, tmp_str);
963     session_use(1);
964     return session;
965 }
966
967 const char * client_get_suggestions_xml(struct client *cl, WRBUF wrbuf);
968
969 static struct hitsbytarget *hitsbytarget_nb(struct session *se,
970                                             int *count, NMEM nmem)
971 {
972     struct hitsbytarget *res = 0;
973     struct client_list *l;
974     size_t sz = 0;
975
976     for (l = se->clients_active; l; l = l->next)
977         sz++;
978
979     res = nmem_malloc(nmem, sizeof(*res) * sz);
980     *count = 0;
981     for (l = se->clients_active; l; l = l->next)
982     {
983         struct client *cl = l->client;
984         WRBUF w = wrbuf_alloc();
985         const char *name = session_setting_oneval(client_get_database(cl),
986                                                   PZ_NAME);
987
988         res[*count].id = client_get_id(cl);
989         res[*count].name = *name ? name : "Unknown";
990         res[*count].hits = client_get_hits(cl);
991         res[*count].approximation = client_get_approximation(cl);
992         res[*count].records = client_get_num_records(cl);
993         res[*count].filtered = client_get_num_records_filtered(cl);
994         res[*count].diagnostic =
995             client_get_diagnostic(cl, &res[*count].addinfo);
996         res[*count].state = client_get_state_str(cl);
997         res[*count].connected  = client_get_connection(cl) ? 1 : 0;
998         session_settings_dump(se, client_get_database(cl), w);
999         res[*count].settings_xml = nmem_strdup(nmem, wrbuf_cstr(w));
1000         wrbuf_rewind(w);
1001         wrbuf_puts(w, "");
1002         res[*count].suggestions_xml = nmem_strdup(nmem, client_get_suggestions_xml(cl, w));
1003         wrbuf_destroy(w);
1004         (*count)++;
1005     }
1006     return res;
1007 }
1008
1009 struct hitsbytarget *get_hitsbytarget(struct session *se, int *count, NMEM nmem)
1010 {
1011     struct hitsbytarget *p;
1012     session_enter(se);
1013     p = hitsbytarget_nb(se, count, nmem);
1014     session_leave(se);
1015     return p;
1016 }
1017
1018 // Compares two hitsbytarget nodes by hitcount
1019 static int cmp_ht(const void *p1, const void *p2)
1020 {
1021     const struct hitsbytarget *h1 = p1;
1022     const struct hitsbytarget *h2 = p2;
1023     return h2->hits - h1->hits;
1024 }
1025
1026 // Compares two hitsbytarget nodes by hitcount
1027 static int cmp_ht_approx(const void *p1, const void *p2)
1028 {
1029     const struct hitsbytarget *h1 = p1;
1030     const struct hitsbytarget *h2 = p2;
1031     return h2->approximation - h1->approximation;
1032 }
1033
1034 static int targets_termlist_nb(WRBUF wrbuf, struct session *se, int num,
1035                                NMEM nmem, int version)
1036 {
1037     struct hitsbytarget *ht;
1038     int count, i;
1039
1040     ht = hitsbytarget_nb(se, &count, nmem);
1041     if (version >= 2)
1042         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht_approx);
1043     else
1044         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht);
1045     for (i = 0; i < count && i < num && ht[i].hits > 0; i++)
1046     {
1047
1048         // do only print terms which have display names
1049     
1050         wrbuf_puts(wrbuf, "<term>\n");
1051
1052         wrbuf_puts(wrbuf, "<id>");
1053         wrbuf_xmlputs(wrbuf, ht[i].id);
1054         wrbuf_puts(wrbuf, "</id>\n");
1055         
1056         wrbuf_puts(wrbuf, "<name>");
1057         if (!ht[i].name || !ht[i].name[0])
1058             wrbuf_xmlputs(wrbuf, "NO TARGET NAME");
1059         else
1060             wrbuf_xmlputs(wrbuf, ht[i].name);
1061         wrbuf_puts(wrbuf, "</name>\n");
1062         
1063         wrbuf_printf(wrbuf, "<frequency>" ODR_INT_PRINTF "</frequency>\n",
1064                      ht[i].hits);
1065
1066         if (version >= 2) {
1067             // Should not print if we know it isn't a approximation.
1068             wrbuf_printf(wrbuf, "<approximation>" ODR_INT_PRINTF "</approximation>\n", ht[i].approximation);
1069             wrbuf_printf(wrbuf, "<records>%d</records>\n", ht[i].records - ht[i].filtered);
1070             wrbuf_printf(wrbuf, "<filtered>%d</filtered>\n", ht[i].filtered);
1071         }
1072
1073         wrbuf_puts(wrbuf, "<state>");
1074         wrbuf_xmlputs(wrbuf, ht[i].state);
1075         wrbuf_puts(wrbuf, "</state>\n");
1076         
1077         wrbuf_printf(wrbuf, "<diagnostic>%d</diagnostic>\n", 
1078                      ht[i].diagnostic);
1079         wrbuf_puts(wrbuf, "</term>\n");
1080     }
1081     return count;
1082 }
1083
1084 void perform_termlist(struct http_channel *c, struct session *se,
1085                       const char *name, int num, int version)
1086 {
1087     int i, j;
1088     NMEM nmem_tmp = nmem_create();
1089     char **names;
1090     int num_names = 0;
1091
1092     if (!name)
1093         name = "*";
1094
1095     nmem_strsplit(nmem_tmp, ",", name, &names, &num_names);
1096
1097     session_enter(se);
1098
1099     for (j = 0; j < num_names; j++)
1100     {
1101         const char *tname;
1102         int must_generate_empty = 1; /* bug 5350 */
1103
1104         for (i = 0; i < se->num_termlists; i++)
1105         {
1106             tname = se->termlists[i].name;
1107             if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1108             {
1109                 struct termlist_score **p = 0;
1110                 int len;
1111
1112                 wrbuf_puts(c->wrbuf, "<list name=\"");
1113                 wrbuf_xmlputs(c->wrbuf, tname);
1114                 wrbuf_puts(c->wrbuf, "\">\n");
1115                 must_generate_empty = 0;
1116
1117                 p = termlist_highscore(se->termlists[i].termlist, &len,
1118                                        nmem_tmp);
1119                 if (p)
1120                 {
1121                     int i;
1122                     for (i = 0; i < len && i < num; i++)
1123                     {
1124                         // prevent sending empty term elements
1125                         if (!p[i]->display_term || !p[i]->display_term[0])
1126                             continue;
1127                         
1128                         wrbuf_puts(c->wrbuf, "<term>");
1129                         wrbuf_puts(c->wrbuf, "<name>");
1130                         wrbuf_xmlputs(c->wrbuf, p[i]->display_term);
1131                         wrbuf_puts(c->wrbuf, "</name>");
1132                         
1133                         wrbuf_printf(c->wrbuf, 
1134                                      "<frequency>%d</frequency>", 
1135                                      p[i]->frequency);
1136                         wrbuf_puts(c->wrbuf, "</term>\n");
1137                     }
1138                 }
1139                 wrbuf_puts(c->wrbuf, "</list>\n");
1140             }
1141         }
1142         tname = "xtargets";
1143         if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1144         {
1145             wrbuf_puts(c->wrbuf, "<list name=\"");
1146             wrbuf_xmlputs(c->wrbuf, tname);
1147             wrbuf_puts(c->wrbuf, "\">\n");
1148
1149             targets_termlist_nb(c->wrbuf, se, num, c->nmem, version);
1150             wrbuf_puts(c->wrbuf, "</list>\n");
1151             must_generate_empty = 0;
1152         }
1153         if (must_generate_empty)
1154         {
1155             wrbuf_puts(c->wrbuf, "<list name=\"");
1156             wrbuf_xmlputs(c->wrbuf, names[j]);
1157             wrbuf_puts(c->wrbuf, "\"/>\n");
1158         }
1159     }
1160     session_leave(se);
1161     nmem_destroy(nmem_tmp);
1162 }
1163
1164 #ifdef MISSING_HEADERS
1165 void report_nmem_stats(void)
1166 {
1167     size_t in_use, is_free;
1168
1169     nmem_get_memory_in_use(&in_use);
1170     nmem_get_memory_free(&is_free);
1171
1172     yaz_log(YLOG_LOG, "nmem stat: use=%ld free=%ld", 
1173             (long) in_use, (long) is_free);
1174 }
1175 #endif
1176
1177 struct record_cluster *show_single_start(struct session *se, const char *id,
1178                                          struct record_cluster **prev_r,
1179                                          struct record_cluster **next_r)
1180 {
1181     struct record_cluster *r = 0;
1182
1183     session_enter(se);
1184     *prev_r = 0;
1185     *next_r = 0;
1186     if (se->reclist)
1187     {
1188         reclist_enter(se->reclist);
1189         while ((r = reclist_read_record(se->reclist)))
1190         {
1191             if (!strcmp(r->recid, id))
1192             {
1193                 *next_r = reclist_read_record(se->reclist);
1194                 break;
1195             }
1196             *prev_r = r;
1197         }
1198         reclist_leave(se->reclist);
1199     }
1200     if (!r)
1201         session_leave(se);
1202     return r;
1203 }
1204
1205 void show_single_stop(struct session *se, struct record_cluster *rec)
1206 {
1207     session_leave(se);
1208 }
1209
1210 struct record_cluster **show_range_start(struct session *se,
1211                                          struct reclist_sortparms *sp, 
1212                                          int start, int *num, int *total, Odr_int *sumhits, Odr_int *approx_hits)
1213 {
1214     struct record_cluster **recs;
1215     struct reclist_sortparms *spp;
1216     int i;
1217 #if USE_TIMING    
1218     yaz_timing_t t = yaz_timing_create();
1219 #endif
1220     session_enter(se);
1221     recs = nmem_malloc(se->nmem, *num * sizeof(struct record_cluster *));
1222     if (!se->relevance)
1223     {
1224         *num = 0;
1225         *total = 0;
1226         *sumhits = 0;        
1227         *approx_hits = 0;
1228         recs = 0;
1229     }
1230     else
1231     {
1232         struct client_list *l;
1233         
1234         for (spp = sp; spp; spp = spp->next)
1235             if (spp->type == Metadata_sortkey_relevance)
1236             {
1237                 relevance_prepare_read(se->relevance, se->reclist);
1238                 break;
1239             }
1240         reclist_sort(se->reclist, sp);
1241         
1242         reclist_enter(se->reclist);
1243         *total = reclist_get_num_records(se->reclist);
1244
1245         *sumhits = 0;
1246         *approx_hits = 0;
1247         for (l = se->clients_active; l; l = l->next) {
1248             *sumhits += client_get_hits(l->client);
1249             *approx_hits += client_get_approximation(l->client);
1250         }
1251         for (i = 0; i < start; i++)
1252             if (!reclist_read_record(se->reclist))
1253             {
1254                 *num = 0;
1255                 recs = 0;
1256                 break;
1257             }
1258         
1259         for (i = 0; i < *num; i++)
1260         {
1261             struct record_cluster *r = reclist_read_record(se->reclist);
1262             if (!r)
1263             {
1264                 *num = i;
1265                 break;
1266             }
1267             recs[i] = r;
1268         }
1269         reclist_leave(se->reclist);
1270     }
1271 #if USE_TIMING
1272     yaz_timing_stop(t);
1273     yaz_log(YLOG_LOG, "show %6.5f %3.2f %3.2f", 
1274             yaz_timing_get_real(t), yaz_timing_get_user(t),
1275             yaz_timing_get_sys(t));
1276     yaz_timing_destroy(&t);
1277 #endif
1278     return recs;
1279 }
1280
1281 void show_range_stop(struct session *se, struct record_cluster **recs)
1282 {
1283     session_leave(se);
1284 }
1285
1286 void statistics(struct session *se, struct statistics *stat)
1287 {
1288     struct client_list *l;
1289     int count = 0;
1290
1291     memset(stat, 0, sizeof(*stat));
1292     stat->num_hits = 0;
1293     for (l = se->clients_active; l; l = l->next)
1294     {
1295         struct client *cl = l->client;
1296         if (!client_get_connection(cl))
1297             stat->num_no_connection++;
1298         stat->num_hits += client_get_hits(cl);
1299         switch (client_get_state(cl))
1300         {
1301         case Client_Connecting: stat->num_connecting++; break;
1302         case Client_Working: stat->num_working++; break;
1303         case Client_Idle: stat->num_idle++; break;
1304         case Client_Failed: stat->num_failed++; break;
1305         case Client_Error: stat->num_error++; break;
1306         default: break;
1307         }
1308         count++;
1309     }
1310     stat->num_records = se->total_records;
1311
1312     stat->num_clients = count;
1313 }
1314
1315 static struct record_metadata *record_metadata_init(
1316     NMEM nmem, const char *value, enum conf_metadata_type type,
1317     struct _xmlAttr *attr)
1318 {
1319     struct record_metadata *rec_md = record_metadata_create(nmem);
1320     struct record_metadata_attr **attrp = &rec_md->attributes;
1321     
1322     for (; attr; attr = attr->next)
1323     {
1324         if (attr->children && attr->children->content)
1325         {
1326             if (strcmp((const char *) attr->name, "type"))
1327             {  /* skip the "type" attribute.. Its value is already part of
1328                   the element in output (md-%s) and so repeating it here
1329                   is redundant */
1330                 *attrp = nmem_malloc(nmem, sizeof(**attrp));
1331                 (*attrp)->name =
1332                     nmem_strdup(nmem, (const char *) attr->name);
1333                 (*attrp)->value =
1334                     nmem_strdup(nmem, (const char *) attr->children->content);
1335                 attrp = &(*attrp)->next;
1336             }
1337         }
1338     }
1339     *attrp = 0;
1340
1341     if (type == Metadata_type_generic)
1342     {
1343         char *p = nmem_strdup(nmem, value);
1344
1345         p = normalize7bit_generic(p, " ,/.:([");
1346         
1347         rec_md->data.text.disp = p;
1348         rec_md->data.text.sort = 0;
1349     }
1350     else if (type == Metadata_type_year || type == Metadata_type_date)
1351     {
1352         int first, last;
1353         int longdate = 0;
1354
1355         if (type == Metadata_type_date)
1356             longdate = 1;
1357         if (extract7bit_dates((char *) value, &first, &last, longdate) < 0)
1358             return 0;
1359
1360         rec_md->data.number.min = first;
1361         rec_md->data.number.max = last;
1362     }
1363     else
1364         return 0;
1365     return rec_md;
1366 }
1367
1368 static int get_mergekey_from_doc(xmlDoc *doc, xmlNode *root, const char *name,
1369                                  struct conf_service *service, WRBUF norm_wr)
1370 {
1371     xmlNode *n;
1372     int no_found = 0;
1373     for (n = root->children; n; n = n->next)
1374     {
1375         if (n->type != XML_ELEMENT_NODE)
1376             continue;
1377         if (!strcmp((const char *) n->name, "metadata"))
1378         {
1379             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1380             if (type == NULL) {
1381                 yaz_log(YLOG_FATAL, "Missing type attribute on metadata element. Skipping!");
1382             }
1383             else if (!strcmp(name, (const char *) type))
1384             {
1385                 xmlChar *value = xmlNodeListGetString(doc, n->children, 1);
1386                 if (value)
1387                 {
1388                     const char *norm_str;
1389                     pp2_charset_token_t prt =
1390                         pp2_charset_token_create(service->charsets, "mergekey");
1391                     
1392                     pp2_charset_token_first(prt, (const char *) value, 0);
1393                     if (wrbuf_len(norm_wr) > 0)
1394                         wrbuf_puts(norm_wr, " ");
1395                     wrbuf_puts(norm_wr, name);
1396                     while ((norm_str =
1397                             pp2_charset_token_next(prt)))
1398                     {
1399                         if (*norm_str)
1400                         {
1401                             wrbuf_puts(norm_wr, " ");
1402                             wrbuf_puts(norm_wr, norm_str);
1403                         }
1404                     }
1405                     xmlFree(value);
1406                     pp2_charset_token_destroy(prt);
1407                     no_found++;
1408                 }
1409             }
1410             xmlFree(type);
1411         }
1412     }
1413     return no_found;
1414 }
1415
1416 static const char *get_mergekey(xmlDoc *doc, struct client *cl, int record_no,
1417                                 struct conf_service *service, NMEM nmem)
1418 {
1419     char *mergekey_norm = 0;
1420     xmlNode *root = xmlDocGetRootElement(doc);
1421     WRBUF norm_wr = wrbuf_alloc();
1422
1423     /* consider mergekey from XSL first */
1424     xmlChar *mergekey = xmlGetProp(root, (xmlChar *) "mergekey");
1425     if (mergekey)
1426     {
1427         const char *norm_str;
1428         pp2_charset_token_t prt =
1429             pp2_charset_token_create(service->charsets, "mergekey");
1430
1431         pp2_charset_token_first(prt, (const char *) mergekey, 0);
1432         while ((norm_str = pp2_charset_token_next(prt)))
1433         {
1434             if (*norm_str)
1435             {
1436                 if (wrbuf_len(norm_wr))
1437                     wrbuf_puts(norm_wr, " ");
1438                 wrbuf_puts(norm_wr, norm_str);
1439             }
1440         }
1441         pp2_charset_token_destroy(prt);
1442         xmlFree(mergekey);
1443     }
1444     else
1445     {
1446         /* no mergekey defined in XSL. Look for mergekey metadata instead */
1447         int field_id;
1448         for (field_id = 0; field_id < service->num_metadata; field_id++)
1449         {
1450             struct conf_metadata *ser_md = &service->metadata[field_id];
1451             if (ser_md->mergekey != Metadata_mergekey_no)
1452             {
1453                 int r = get_mergekey_from_doc(doc, root, ser_md->name,
1454                                               service, norm_wr);
1455                 if (r == 0 && ser_md->mergekey == Metadata_mergekey_required)
1456                 {
1457                     /* no mergekey on this one and it is required.. 
1458                        Generate unique key instead */
1459                     wrbuf_rewind(norm_wr);
1460                     break;
1461                 }
1462             }
1463         }
1464     }
1465
1466     /* generate unique key if none is not generated already or is empty */
1467     if (wrbuf_len(norm_wr) == 0)
1468     {
1469         wrbuf_printf(norm_wr, "position: %s-%d",
1470                      client_get_id(cl), record_no);
1471     }
1472     else
1473     {
1474         const char *lead = "content: ";
1475         wrbuf_insert(norm_wr, 0, lead, strlen(lead));
1476     }
1477     if (wrbuf_len(norm_wr) > 0)
1478         mergekey_norm = nmem_strdup(nmem, wrbuf_cstr(norm_wr));
1479     wrbuf_destroy(norm_wr);
1480     return mergekey_norm;
1481 }
1482
1483 /** \brief see if metadata for pz:recordfilter exists 
1484     \param root xml root element of normalized record
1485     \param sdb session database for client
1486     \retval 0 if there is no metadata for pz:recordfilter
1487     \retval 1 if there is metadata for pz:recordfilter
1488
1489     If there is no pz:recordfilter defined, this function returns 1
1490     as well.
1491 */
1492     
1493 static int check_record_filter(xmlNode *root, struct session_database *sdb)
1494 {
1495     int match = 0;
1496     xmlNode *n;
1497     const char *s;
1498     s = session_setting_oneval(sdb, PZ_RECORDFILTER);
1499
1500     if (!s || !*s)
1501         return 1;
1502
1503     for (n = root->children; n; n = n->next)
1504     {
1505         if (n->type != XML_ELEMENT_NODE)
1506             continue;
1507         if (!strcmp((const char *) n->name, "metadata"))
1508         {
1509             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1510             if (type)
1511             {
1512                 size_t len;
1513                 int substring;
1514                 const char *eq;
1515
1516                 if ((eq = strchr(s, '=')))
1517                     substring = 0;
1518                 else if ((eq = strchr(s, '~')))
1519                     substring = 1;
1520                 if (eq)
1521                     len = eq - s;
1522                 else
1523                     len = strlen(s);
1524                 if (len == strlen((const char *)type) &&
1525                     !memcmp((const char *) type, s, len))
1526                 {
1527                     xmlChar *value = xmlNodeGetContent(n);
1528                     if (value && *value)
1529                     {
1530                         if (!eq ||
1531                             (substring && strstr((const char *) value, eq+1)) ||
1532                             (!substring && !strcmp((const char *) value, eq + 1)))
1533                             match = 1;
1534                     }
1535                     xmlFree(value);
1536                 }
1537                 xmlFree(type);
1538             }
1539         }
1540     }
1541     return match;
1542 }
1543
1544
1545 static int ingest_to_cluster(struct client *cl,
1546                              xmlDoc *xdoc,
1547                              xmlNode *root,
1548                              int record_no,
1549                              const char *mergekey_norm);
1550
1551 /** \brief ingest XML record
1552     \param cl client holds the result set for record
1553     \param rec record buffer (0 terminated)
1554     \param record_no record position (1, 2, ..)
1555     \param nmem working NMEM
1556     \retval 0 OK
1557     \retval -1 failure
1558     \retval -2 Filtered
1559 */
1560 int ingest_record(struct client *cl, const char *rec,
1561                   int record_no, NMEM nmem)
1562 {
1563     struct session *se = client_get_session(cl);
1564     int ret = 0;
1565     struct session_database *sdb = client_get_database(cl);
1566     struct conf_service *service = se->service;
1567     xmlDoc *xdoc = normalize_record(se, sdb, service, rec, nmem);
1568     xmlNode *root;
1569     const char *mergekey_norm;
1570     
1571     if (!xdoc)
1572         return -1;
1573     
1574     root = xmlDocGetRootElement(xdoc);
1575     
1576     if (!check_record_filter(root, sdb))
1577     {
1578         session_log(se, YLOG_LOG, "Filtered out record no %d from %s", record_no, sdb->database->id);
1579         xmlFreeDoc(xdoc);
1580         return -2;
1581     }
1582     
1583     mergekey_norm = get_mergekey(xdoc, cl, record_no, service, nmem);
1584     if (!mergekey_norm)
1585     {
1586         session_log(se, YLOG_WARN, "Got no mergekey");
1587         xmlFreeDoc(xdoc);
1588         return -1;
1589     }
1590     session_enter(se);
1591     if (client_get_session(cl) == se)
1592         ret = ingest_to_cluster(cl, xdoc, root, record_no, mergekey_norm);
1593     session_leave(se);
1594     
1595     xmlFreeDoc(xdoc);
1596     return ret;
1597 }
1598
1599 // Skip record on non-zero
1600 static int check_limit_local(struct client *cl,
1601                              struct record *record,
1602                              int record_no)
1603 {
1604     int skip_record = 0;
1605     struct session *se = client_get_session(cl);
1606     struct conf_service *service = se->service;
1607     NMEM nmem_tmp = nmem_create();
1608     struct session_database *sdb = client_get_database(cl);
1609     int l = 0;
1610     while (!skip_record)
1611     {
1612         struct conf_metadata *ser_md = 0;
1613         struct record_metadata *rec_md = 0;
1614         int md_field_id;
1615         char **values = 0;
1616         int i, num_v = 0;
1617         
1618         const char *name = client_get_facet_limit_local(cl, sdb, &l, nmem_tmp, &num_v, &values);
1619         if (!name)
1620             break;
1621         
1622         md_field_id = conf_service_metadata_field_id(service, name);
1623         if (md_field_id < 0)
1624         {
1625             skip_record = 1;
1626             break;
1627         }
1628         ser_md = &service->metadata[md_field_id];
1629         rec_md = record->metadata[md_field_id];
1630         yaz_log(YLOG_DEBUG, "check limit local %s", name);
1631         for (i = 0; i < num_v; )
1632         {
1633             if (rec_md)
1634             {
1635                 if (ser_md->type == Metadata_type_year 
1636                     || ser_md->type == Metadata_type_date)
1637                 {
1638                     int y = atoi(values[i]);
1639                     if (y >= rec_md->data.number.min 
1640                         && y <= rec_md->data.number.max)
1641                         break;
1642                 }
1643                 else
1644                 {
1645                     yaz_log(YLOG_DEBUG, "cmp: '%s' '%s'", rec_md->data.text.disp, values[i]);
1646                     if (!strcmp(rec_md->data.text.disp, values[i]))
1647                     {
1648                         // Value equals, should not be filtered.
1649                         break;
1650                     }
1651                 }
1652                 rec_md = rec_md->next;
1653             }
1654             else
1655             {
1656                 rec_md = record->metadata[md_field_id];
1657                 i++;
1658             }
1659         }
1660         // At end , not match
1661         if (i == num_v)
1662         {
1663             skip_record = 1;
1664             break;
1665         }
1666     }
1667     nmem_destroy(nmem_tmp);
1668     return skip_record;
1669 }
1670                              
1671 static int ingest_to_cluster(struct client *cl,
1672                              xmlDoc *xdoc,
1673                              xmlNode *root,
1674                              int record_no,
1675                              const char *mergekey_norm)
1676 {
1677     xmlNode *n;
1678     xmlChar *type = 0;
1679     xmlChar *value = 0;
1680     struct session *se = client_get_session(cl);
1681     struct conf_service *service = se->service;
1682     int term_factor = 1;
1683     struct record_cluster *cluster;
1684     struct session_database *sdb = client_get_database(cl);
1685     struct record *record = record_create(se->nmem, 
1686                                           service->num_metadata,
1687                                           service->num_sortkeys, cl,
1688                                           record_no);
1689
1690     for (n = root->children; n; n = n->next)
1691     {
1692         if (type)
1693             xmlFree(type);
1694         if (value)
1695             xmlFree(value);
1696         type = value = 0;
1697         
1698         if (n->type != XML_ELEMENT_NODE)
1699             continue;
1700         if (!strcmp((const char *) n->name, "metadata"))
1701         {
1702             struct conf_metadata *ser_md = 0;
1703             struct record_metadata **wheretoput = 0;
1704             struct record_metadata *rec_md = 0;
1705             int md_field_id = -1;
1706             
1707             type = xmlGetProp(n, (xmlChar *) "type");
1708             value = xmlNodeListGetString(xdoc, n->children, 1);
1709             
1710             if (!type || !value || !*value)
1711                 continue;
1712             
1713             md_field_id 
1714                 = conf_service_metadata_field_id(service, (const char *) type);
1715             if (md_field_id < 0)
1716             {
1717                 if (se->number_of_warnings_unknown_metadata == 0)
1718                 {
1719                     session_log(se, YLOG_WARN, 
1720                             "Ignoring unknown metadata element: %s", type);
1721                 }
1722                 se->number_of_warnings_unknown_metadata++;
1723                 continue;
1724             }
1725            
1726             ser_md = &service->metadata[md_field_id];
1727
1728             // non-merged metadata
1729             rec_md = record_metadata_init(se->nmem, (const char *) value,
1730                                           ser_md->type, n->properties);
1731             if (!rec_md)
1732             {
1733                 session_log(se, YLOG_WARN, "bad metadata data '%s' "
1734                             "for element '%s'", value, type);
1735                 continue;
1736             }
1737             wheretoput = &record->metadata[md_field_id];
1738             while (*wheretoput)
1739                 wheretoput = &(*wheretoput)->next;
1740             *wheretoput = rec_md;
1741         }
1742     }
1743
1744     if (check_limit_local(cl, record, record_no))
1745     {
1746         session_log(se, YLOG_LOG, "Facet filtered out record no %d from %s",
1747                     record_no, sdb->database->id);
1748         if (type)
1749             xmlFree(type);
1750         if (value)
1751             xmlFree(value);
1752         return -2;
1753     }
1754     cluster = reclist_insert(se->reclist, service, record,
1755                              mergekey_norm, &se->total_merged);
1756     if (!cluster)
1757         return -1;
1758
1759     {
1760         const char *use_term_factor_str =
1761             session_setting_oneval(sdb, PZ_TERMLIST_TERM_FACTOR);
1762         if (use_term_factor_str && use_term_factor_str[0] == '1')
1763         {
1764             int maxrecs = client_get_maxrecs(cl);
1765             int hits = (int) client_get_hits(cl);
1766             term_factor = MAX(hits, maxrecs) /  MAX(1, maxrecs);
1767             assert(term_factor >= 1);
1768             yaz_log(YLOG_DEBUG, "Using term factor: %d (%d / %d)", term_factor, MAX(hits, maxrecs), MAX(1, maxrecs));
1769         }
1770     }
1771
1772     if (global_parameters.dump_records)
1773         session_log(se, YLOG_LOG, "Cluster id %s from %s (#%d)", cluster->recid,
1774                     sdb->database->id, record_no);
1775
1776
1777     relevance_newrec(se->relevance, cluster);
1778     
1779     // now parsing XML record and adding data to cluster or record metadata
1780     for (n = root->children; n; n = n->next)
1781     {
1782         pp2_charset_token_t prt;
1783         if (type)
1784             xmlFree(type);
1785         if (value)
1786             xmlFree(value);
1787         type = value = 0;
1788         
1789         if (n->type != XML_ELEMENT_NODE)
1790             continue;
1791         if (!strcmp((const char *) n->name, "metadata"))
1792         {
1793             struct conf_metadata *ser_md = 0;
1794             struct conf_sortkey *ser_sk = 0;
1795             struct record_metadata **wheretoput = 0;
1796             struct record_metadata *rec_md = 0;
1797             int md_field_id = -1;
1798             int sk_field_id = -1;
1799             const char *rank;
1800             xmlChar *xml_rank;
1801             
1802             type = xmlGetProp(n, (xmlChar *) "type");
1803             value = xmlNodeListGetString(xdoc, n->children, 1);
1804             
1805             if (!type || !value || !*value)
1806                 continue;
1807             
1808             md_field_id 
1809                 = conf_service_metadata_field_id(service, (const char *) type);
1810             if (md_field_id < 0)
1811                 continue;
1812             
1813             ser_md = &service->metadata[md_field_id];
1814
1815             if (ser_md->sortkey_offset >= 0)
1816             {
1817                 sk_field_id = ser_md->sortkey_offset;
1818                 ser_sk = &service->sortkeys[sk_field_id];
1819             }
1820
1821             // merged metadata
1822             rec_md = record_metadata_init(se->nmem, (const char *) value,
1823                                           ser_md->type, 0);
1824             if (!rec_md)
1825                 continue;
1826
1827             xml_rank = xmlGetProp(n, (xmlChar *) "rank");
1828             rank = xml_rank ? (const char *) xml_rank : ser_md->rank;
1829
1830             wheretoput = &cluster->metadata[md_field_id];
1831
1832             // and polulate with data:
1833             // assign cluster or record based on merge action
1834             if (ser_md->merge == Metadata_merge_unique)
1835             {
1836                 while (*wheretoput)
1837                 {
1838                     if (!strcmp((const char *) (*wheretoput)->data.text.disp, 
1839                                 rec_md->data.text.disp))
1840                         break;
1841                     wheretoput = &(*wheretoput)->next;
1842                 }
1843                 if (!*wheretoput)
1844                     *wheretoput = rec_md;
1845             }
1846             else if (ser_md->merge == Metadata_merge_longest)
1847             {
1848                 if (!*wheretoput 
1849                     || strlen(rec_md->data.text.disp) 
1850                     > strlen((*wheretoput)->data.text.disp))
1851                 {
1852                     *wheretoput = rec_md;
1853                     if (ser_sk)
1854                     {
1855                         const char *sort_str = 0;
1856                         int skip_article = 
1857                             ser_sk->type == Metadata_sortkey_skiparticle;
1858
1859                         if (!cluster->sortkeys[sk_field_id])
1860                             cluster->sortkeys[sk_field_id] = 
1861                                 nmem_malloc(se->nmem, 
1862                                             sizeof(union data_types));
1863                          
1864                         prt =
1865                             pp2_charset_token_create(service->charsets, "sort");
1866
1867                         pp2_charset_token_first(prt, rec_md->data.text.disp,
1868                                                 skip_article);
1869
1870                         pp2_charset_token_next(prt);
1871                          
1872                         sort_str = pp2_get_sort(prt);
1873                          
1874                         cluster->sortkeys[sk_field_id]->text.disp = 
1875                             rec_md->data.text.disp;
1876                         if (!sort_str)
1877                         {
1878                             sort_str = rec_md->data.text.disp;
1879                             session_log(se, YLOG_WARN, 
1880                                     "Could not make sortkey. Bug #1858");
1881                         }
1882                         cluster->sortkeys[sk_field_id]->text.sort = 
1883                             nmem_strdup(se->nmem, sort_str);
1884                         pp2_charset_token_destroy(prt);
1885                     }
1886                 }
1887             }
1888             else if (ser_md->merge == Metadata_merge_all)
1889             {
1890                 while (*wheretoput)
1891                     wheretoput = &(*wheretoput)->next;
1892                 *wheretoput = rec_md;
1893             }
1894             else if (ser_md->merge == Metadata_merge_range)
1895             {
1896                 if (!*wheretoput)
1897                 {
1898                     *wheretoput = rec_md;
1899                     if (ser_sk)
1900                         cluster->sortkeys[sk_field_id] 
1901                             = &rec_md->data;
1902                 }
1903                 else
1904                 {
1905                     int this_min = rec_md->data.number.min;
1906                     int this_max = rec_md->data.number.max;
1907                     if (this_min < (*wheretoput)->data.number.min)
1908                         (*wheretoput)->data.number.min = this_min;
1909                     if (this_max > (*wheretoput)->data.number.max)
1910                         (*wheretoput)->data.number.max = this_max;
1911                 }
1912             }
1913
1914             // ranking of _all_ fields enabled ... 
1915             if (rank)
1916             {
1917                 relevance_countwords(se->relevance, cluster, 
1918                                      (char *) value, rank, ser_md->name);
1919             }
1920
1921             // construct facets ... unless the client already has reported them
1922             if (ser_md->termlist && !client_has_facet(cl, (char *) type))
1923             {
1924                 if (ser_md->type == Metadata_type_year)
1925                 {
1926                     char year[64];
1927                     sprintf(year, "%d", rec_md->data.number.max);
1928
1929                     add_facet(se, (char *) type, year, term_factor);
1930                     if (rec_md->data.number.max != rec_md->data.number.min)
1931                     {
1932                         sprintf(year, "%d", rec_md->data.number.min);
1933                         add_facet(se, (char *) type, year, term_factor);
1934                     }
1935                 }
1936                 else
1937                     add_facet(se, (char *) type, (char *) value, term_factor);
1938             }
1939
1940             // cleaning up
1941             if (xml_rank)
1942                 xmlFree(xml_rank);
1943             xmlFree(type);
1944             xmlFree(value);
1945             type = value = 0;
1946         }
1947         else
1948         {
1949             if (se->number_of_warnings_unknown_elements == 0)
1950                 session_log(se, YLOG_WARN,
1951                         "Unexpected element in internal record: %s", n->name);
1952             se->number_of_warnings_unknown_elements++;
1953         }
1954     }
1955     if (type)
1956         xmlFree(type);
1957     if (value)
1958         xmlFree(value);
1959
1960     relevance_donerecord(se->relevance, cluster);
1961     se->total_records++;
1962
1963     return 0;
1964 }
1965
1966 void session_log(struct session *s, int level, const char *fmt, ...)
1967 {
1968     char buf[1024];
1969     va_list ap;
1970     va_start(ap, fmt);
1971
1972     yaz_vsnprintf(buf, sizeof(buf)-30, fmt, ap);
1973     yaz_log(level, "Session %u: %s", s->session_id, buf);
1974
1975     va_end(ap);
1976 }
1977
1978 /*
1979  * Local variables:
1980  * c-basic-offset: 4
1981  * c-file-style: "Stroustrup"
1982  * indent-tabs-mode: nil
1983  * End:
1984  * vim: shiftwidth=4 tabstop=8 expandtab
1985  */
1986