4a817a24a566436f08f3b63e34a6d306a92b0db5
[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, int position)
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->position = position;
639     se->sorted_results->next = 0;
640
641     se->reclist = reclist_create(se->nmem);
642 }
643
644 void session_sort(struct session *se, const char *field, int increasing,
645                   int position)
646 {
647     struct session_sorted_results *sr;
648     struct client_list *l;
649
650     session_enter(se);
651
652     yaz_log(YLOG_LOG, "session_sort field=%s increasing=%d position=%d", field, increasing, position);
653     /* see if we already have sorted for this critieria */
654     /* TODO I do not see the point in saving all previous sorts. Dont we re-sort anyway ? */
655     for (sr = se->sorted_results; sr; sr = sr->next)
656     {
657         if (!strcmp(field, sr->field) && increasing == sr->increasing && sr->position == position)
658             break;
659     }
660     if (sr)
661     {
662         session_log(se, YLOG_DEBUG, "search_sort: field=%s increasing=%d position=%d already fetched",
663                     field, increasing, position);
664         session_leave(se);
665         return;
666     }
667     session_log(se, YLOG_DEBUG, "search_sort: field=%s increasing=%d position=%d must fetch",
668                 field, increasing, position);
669     if (position)
670     {
671         yaz_log(YLOG_DEBUG, "Reset results due to position");
672         session_clear_set(se, field, increasing, position);
673     }
674     else {
675         sr = nmem_malloc(se->nmem, sizeof(*sr));
676         sr->field = nmem_strdup(se->nmem, field);
677         sr->increasing = increasing;
678         sr->position = position;
679         sr->next = se->sorted_results;
680         se->sorted_results = sr;
681     }
682     yaz_log(YLOG_DEBUG, "Restarting search for clients due to change in sort order");
683
684     for (l = se->clients_active; l; l = l->next)
685     {
686         struct client *cl = l->client;
687         if (client_get_state(cl) == Client_Connecting ||
688             client_get_state(cl) == Client_Idle ||
689             client_get_state(cl) == Client_Working) {
690             yaz_log(YLOG_DEBUG, "Client %s: Restarting search due to change in sort order", client_get_id(cl));
691             client_start_search(cl);
692         }
693     }
694     session_leave(se);
695 }
696
697 enum pazpar2_error_code session_search(struct session *se,
698                                        const char *query,
699                                        const char *startrecs,
700                                        const char *maxrecs,
701                                        const char *filter,
702                                        const char *limit,
703                                        const char **addinfo,
704                                        const char *sort_field,
705                                        int increasing)
706 {
707     int live_channels = 0;
708     int no_working = 0;
709     int no_failed_query = 0;
710     int no_failed_limit = 0;
711     struct client_list *l, *l0;
712     struct timeval tval;
713     facet_limits_t facet_limits;
714
715     session_log(se, YLOG_DEBUG, "Search");
716
717     *addinfo = 0;
718
719     if (se->settings_modified)
720         session_remove_cached_clients(se);
721     else
722         session_reset_active_clients(se, 0);
723
724     session_enter(se);
725     se->settings_modified = 0;
726     session_clear_set(se, sort_field, increasing, 0); /* hardcoded position */
727     relevance_destroy(&se->relevance);
728
729     live_channels = select_targets(se, filter);
730     if (!live_channels)
731     {
732         session_leave(se);
733         return PAZPAR2_NO_TARGETS;
734     }
735
736     yaz_gettimeofday(&tval);
737
738     tval.tv_sec += 5;
739
740     facet_limits = facet_limits_create(limit);
741     if (!facet_limits)
742     {
743         *addinfo = "limit";
744         session_leave(se);
745         return PAZPAR2_MALFORMED_PARAMETER_VALUE;
746     }
747
748     l0 = se->clients_active;
749     se->clients_active = 0;
750     session_leave(se);
751
752     for (l = l0; l; l = l->next)
753     {
754         int parse_ret;
755         struct client *cl = l->client;
756
757         if (prepare_map(se, client_get_database(cl)) < 0)
758             continue;
759
760         parse_ret = client_parse_query(cl, query, facet_limits, startrecs,
761                                        maxrecs, se->service->ccl_bibset);
762         if (parse_ret == -1)
763             no_failed_query++;
764         else if (parse_ret == -2)
765             no_failed_limit++;
766         else
767         {
768             int r =
769                 client_prep_connection(cl, se->service->z3950_operation_timeout,
770                                        se->service->z3950_session_timeout,
771                                        se->service->server->iochan_man,
772                                        &tval);
773             if (parse_ret == 1 && r == 2)
774             {
775                 session_log(se, YLOG_LOG, "client %s REUSE result", client_get_id(cl));
776                 client_reingest(cl);
777             }
778             else if (r)
779             {
780                 session_log(se, YLOG_LOG, "client %s NEW search", client_get_id(cl));
781                 client_start_search(cl);
782             }
783             no_working++;
784         }
785     }
786     facet_limits_destroy(facet_limits);
787     session_reset_active_clients(se, l0);
788
789     if (no_working == 0)
790     {
791         if (no_failed_query > 0)
792         {
793             *addinfo = "query";
794             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
795         }
796         else if (no_failed_limit > 0)
797         {
798             *addinfo = "limit";
799             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
800         }
801         else
802             return PAZPAR2_NO_TARGETS;
803     }
804     session_log(se, YLOG_LOG, "session_start_search done");
805     return PAZPAR2_NO_ERROR;
806 }
807
808 // Creates a new session_database object for a database
809 static void session_init_databases_fun(void *context, struct database *db)
810 {
811     struct session *se = (struct session *) context;
812     struct session_database *new = nmem_malloc(se->session_nmem, sizeof(*new));
813     int i;
814
815     new->database = db;
816
817     new->map = 0;
818     assert(db->settings);
819     new->settings = nmem_malloc(se->session_nmem,
820                                 sizeof(struct settings *) * db->num_settings);
821     new->num_settings = db->num_settings;
822     for (i = 0; i < db->num_settings; i++)
823     {
824         struct setting *setting = db->settings[i];
825         new->settings[i] = setting;
826     }
827     new->next = se->databases;
828     se->databases = new;
829 }
830
831 // Doesn't free memory associated with sdb -- nmem takes care of that
832 static void session_database_destroy(struct session_database *sdb)
833 {
834     sdb->map = 0;
835 }
836
837 // Initialize session_database list -- this represents this session's view
838 // of the database list -- subject to modification by the settings ws command
839 void session_init_databases(struct session *se)
840 {
841     se->databases = 0;
842     predef_grep_databases(se, se->service, session_init_databases_fun);
843 }
844
845 // Probably session_init_databases_fun should be refactored instead of
846 // called here.
847 static struct session_database *load_session_database(struct session *se,
848                                                       char *id)
849 {
850     struct database *db = new_database_inherit_settings(id, se->session_nmem, se->service->settings);
851     session_init_databases_fun((void*) se, db);
852
853     // New sdb is head of se->databases list
854     return se->databases;
855 }
856
857 // Find an existing session database. If not found, load it
858 static struct session_database *find_session_database(struct session *se,
859                                                       char *id)
860 {
861     struct session_database *sdb;
862
863     for (sdb = se->databases; sdb; sdb = sdb->next)
864         if (!strcmp(sdb->database->id, id))
865             return sdb;
866     return load_session_database(se, id);
867 }
868
869 // Apply a session override to a database
870 void session_apply_setting(struct session *se, char *dbname, char *setting,
871                            char *value)
872 {
873     struct session_database *sdb = find_session_database(se, dbname);
874     struct conf_service *service = se->service;
875     struct setting *new = nmem_malloc(se->session_nmem, sizeof(*new));
876     int offset = settings_create_offset(service, setting);
877
878     expand_settings_array(&sdb->settings, &sdb->num_settings, offset,
879                           se->session_nmem);
880     new->precedence = 0;
881     new->target = dbname;
882     new->name = setting;
883     new->value = value;
884     new->next = sdb->settings[offset];
885     sdb->settings[offset] = new;
886
887     se->settings_modified = 1;
888
889     // Force later recompute of settings-driven data structures
890     // (happens when a search starts and client connections are prepared)
891     switch (offset)
892     {
893     case PZ_XSLT:
894         if (sdb->map)
895         {
896             sdb->map = 0;
897         }
898         break;
899     }
900 }
901
902 void session_destroy(struct session *se)
903 {
904     struct session_database *sdb;
905     session_log(se, YLOG_DEBUG, "Destroying");
906     session_use(-1);
907     session_remove_cached_clients(se);
908
909     for (sdb = se->databases; sdb; sdb = sdb->next)
910         session_database_destroy(sdb);
911     normalize_cache_destroy(se->normalize_cache);
912     relevance_destroy(&se->relevance);
913     reclist_destroy(se->reclist);
914     if (nmem_total(se->nmem))
915         session_log(se, YLOG_DEBUG, "NMEN operation usage %zd", nmem_total(se->nmem));
916     if (nmem_total(se->session_nmem))
917         session_log(se, YLOG_DEBUG, "NMEN session usage %zd", nmem_total(se->session_nmem));
918     nmem_destroy(se->nmem);
919     service_destroy(se->service);
920     yaz_mutex_destroy(&se->session_mutex);
921 }
922
923 size_t session_get_memory_status(struct session *session) {
924     size_t session_nmem;
925     if (session == 0)
926         return 0;
927     session_enter(session);
928     session_nmem = nmem_total(session->nmem);
929     session_leave(session);
930     return session_nmem;
931 }
932
933
934 struct session *new_session(NMEM nmem, struct conf_service *service,
935                             unsigned session_id)
936 {
937     int i;
938     struct session *session = nmem_malloc(nmem, sizeof(*session));
939
940     char tmp_str[50];
941
942     sprintf(tmp_str, "session#%u", session_id);
943
944     session->session_id = session_id;
945     session_log(session, YLOG_DEBUG, "New");
946     session->service = service;
947     session->relevance = 0;
948     session->total_records = 0;
949     session->number_of_warnings_unknown_elements = 0;
950     session->number_of_warnings_unknown_metadata = 0;
951     session->num_termlists = 0;
952     session->reclist = 0;
953     session->clients_active = 0;
954     session->clients_cached = 0;
955     session->settings_modified = 0;
956     session->session_nmem = nmem;
957     session->nmem = nmem_create();
958     session->databases = 0;
959     for (i = 0; i <= SESSION_WATCH_MAX; i++)
960     {
961         session->watchlist[i].data = 0;
962         session->watchlist[i].fun = 0;
963     }
964     session->normalize_cache = normalize_cache_create();
965     session->session_mutex = 0;
966     pazpar2_mutex_create(&session->session_mutex, tmp_str);
967     session_use(1);
968     return session;
969 }
970
971 const char * client_get_suggestions_xml(struct client *cl, WRBUF wrbuf);
972
973 static struct hitsbytarget *hitsbytarget_nb(struct session *se,
974                                             int *count, NMEM nmem)
975 {
976     struct hitsbytarget *res = 0;
977     struct client_list *l;
978     size_t sz = 0;
979
980     for (l = se->clients_active; l; l = l->next)
981         sz++;
982
983     res = nmem_malloc(nmem, sizeof(*res) * sz);
984     *count = 0;
985     for (l = se->clients_active; l; l = l->next)
986     {
987         struct client *cl = l->client;
988         WRBUF w = wrbuf_alloc();
989         const char *name = session_setting_oneval(client_get_database(cl),
990                                                   PZ_NAME);
991
992         res[*count].id = client_get_id(cl);
993         res[*count].name = *name ? name : "Unknown";
994         res[*count].hits = client_get_hits(cl);
995         res[*count].approximation = client_get_approximation(cl);
996         res[*count].records = client_get_num_records(cl);
997         res[*count].filtered = client_get_num_records_filtered(cl);
998         res[*count].diagnostic =
999             client_get_diagnostic(cl, &res[*count].addinfo);
1000         res[*count].state = client_get_state_str(cl);
1001         res[*count].connected  = client_get_connection(cl) ? 1 : 0;
1002         session_settings_dump(se, client_get_database(cl), w);
1003         res[*count].settings_xml = nmem_strdup(nmem, wrbuf_cstr(w));
1004         wrbuf_rewind(w);
1005         wrbuf_puts(w, "");
1006         res[*count].suggestions_xml = nmem_strdup(nmem, client_get_suggestions_xml(cl, w));
1007         wrbuf_destroy(w);
1008         (*count)++;
1009     }
1010     return res;
1011 }
1012
1013 struct hitsbytarget *get_hitsbytarget(struct session *se, int *count, NMEM nmem)
1014 {
1015     struct hitsbytarget *p;
1016     session_enter(se);
1017     p = hitsbytarget_nb(se, count, nmem);
1018     session_leave(se);
1019     return p;
1020 }
1021
1022 // Compares two hitsbytarget nodes by hitcount
1023 static int cmp_ht(const void *p1, const void *p2)
1024 {
1025     const struct hitsbytarget *h1 = p1;
1026     const struct hitsbytarget *h2 = p2;
1027     return h2->hits - h1->hits;
1028 }
1029
1030 // Compares two hitsbytarget nodes by hitcount
1031 static int cmp_ht_approx(const void *p1, const void *p2)
1032 {
1033     const struct hitsbytarget *h1 = p1;
1034     const struct hitsbytarget *h2 = p2;
1035     return h2->approximation - h1->approximation;
1036 }
1037
1038 static int targets_termlist_nb(WRBUF wrbuf, struct session *se, int num,
1039                                NMEM nmem, int version)
1040 {
1041     struct hitsbytarget *ht;
1042     int count, i;
1043
1044     ht = hitsbytarget_nb(se, &count, nmem);
1045     if (version >= 2)
1046         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht_approx);
1047     else
1048         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht);
1049     for (i = 0; i < count && i < num && ht[i].hits > 0; i++)
1050     {
1051
1052         // do only print terms which have display names
1053
1054         wrbuf_puts(wrbuf, "<term>\n");
1055
1056         wrbuf_puts(wrbuf, "<id>");
1057         wrbuf_xmlputs(wrbuf, ht[i].id);
1058         wrbuf_puts(wrbuf, "</id>\n");
1059
1060         wrbuf_puts(wrbuf, "<name>");
1061         if (!ht[i].name || !ht[i].name[0])
1062             wrbuf_xmlputs(wrbuf, "NO TARGET NAME");
1063         else
1064             wrbuf_xmlputs(wrbuf, ht[i].name);
1065         wrbuf_puts(wrbuf, "</name>\n");
1066
1067         wrbuf_printf(wrbuf, "<frequency>" ODR_INT_PRINTF "</frequency>\n",
1068                      ht[i].hits);
1069
1070         if (version >= 2) {
1071             // Should not print if we know it isn't a approximation.
1072             wrbuf_printf(wrbuf, "<approximation>" ODR_INT_PRINTF "</approximation>\n", ht[i].approximation);
1073             wrbuf_printf(wrbuf, "<records>%d</records>\n", ht[i].records - ht[i].filtered);
1074             wrbuf_printf(wrbuf, "<filtered>%d</filtered>\n", ht[i].filtered);
1075         }
1076
1077         wrbuf_puts(wrbuf, "<state>");
1078         wrbuf_xmlputs(wrbuf, ht[i].state);
1079         wrbuf_puts(wrbuf, "</state>\n");
1080
1081         wrbuf_printf(wrbuf, "<diagnostic>%d</diagnostic>\n",
1082                      ht[i].diagnostic);
1083         wrbuf_puts(wrbuf, "</term>\n");
1084     }
1085     return count;
1086 }
1087
1088 void perform_termlist(struct http_channel *c, struct session *se,
1089                       const char *name, int num, int version)
1090 {
1091     int i, j;
1092     NMEM nmem_tmp = nmem_create();
1093     char **names;
1094     int num_names = 0;
1095
1096     if (!name)
1097         name = "*";
1098
1099     nmem_strsplit(nmem_tmp, ",", name, &names, &num_names);
1100
1101     session_enter(se);
1102
1103     for (j = 0; j < num_names; j++)
1104     {
1105         const char *tname;
1106         int must_generate_empty = 1; /* bug 5350 */
1107
1108         for (i = 0; i < se->num_termlists; i++)
1109         {
1110             tname = se->termlists[i].name;
1111             if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1112             {
1113                 struct termlist_score **p = 0;
1114                 int len;
1115
1116                 wrbuf_puts(c->wrbuf, "<list name=\"");
1117                 wrbuf_xmlputs(c->wrbuf, tname);
1118                 wrbuf_puts(c->wrbuf, "\">\n");
1119                 must_generate_empty = 0;
1120
1121                 p = termlist_highscore(se->termlists[i].termlist, &len,
1122                                        nmem_tmp);
1123                 if (p)
1124                 {
1125                     int i;
1126                     for (i = 0; i < len && i < num; i++)
1127                     {
1128                         // prevent sending empty term elements
1129                         if (!p[i]->display_term || !p[i]->display_term[0])
1130                             continue;
1131
1132                         wrbuf_puts(c->wrbuf, "<term>");
1133                         wrbuf_puts(c->wrbuf, "<name>");
1134                         wrbuf_xmlputs(c->wrbuf, p[i]->display_term);
1135                         wrbuf_puts(c->wrbuf, "</name>");
1136
1137                         wrbuf_printf(c->wrbuf,
1138                                      "<frequency>%d</frequency>",
1139                                      p[i]->frequency);
1140                         wrbuf_puts(c->wrbuf, "</term>\n");
1141                     }
1142                 }
1143                 wrbuf_puts(c->wrbuf, "</list>\n");
1144             }
1145         }
1146         tname = "xtargets";
1147         if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1148         {
1149             wrbuf_puts(c->wrbuf, "<list name=\"");
1150             wrbuf_xmlputs(c->wrbuf, tname);
1151             wrbuf_puts(c->wrbuf, "\">\n");
1152
1153             targets_termlist_nb(c->wrbuf, se, num, c->nmem, version);
1154             wrbuf_puts(c->wrbuf, "</list>\n");
1155             must_generate_empty = 0;
1156         }
1157         if (must_generate_empty)
1158         {
1159             wrbuf_puts(c->wrbuf, "<list name=\"");
1160             wrbuf_xmlputs(c->wrbuf, names[j]);
1161             wrbuf_puts(c->wrbuf, "\"/>\n");
1162         }
1163     }
1164     session_leave(se);
1165     nmem_destroy(nmem_tmp);
1166 }
1167
1168 #ifdef MISSING_HEADERS
1169 void report_nmem_stats(void)
1170 {
1171     size_t in_use, is_free;
1172
1173     nmem_get_memory_in_use(&in_use);
1174     nmem_get_memory_free(&is_free);
1175
1176     yaz_log(YLOG_LOG, "nmem stat: use=%ld free=%ld",
1177             (long) in_use, (long) is_free);
1178 }
1179 #endif
1180
1181 struct record_cluster *show_single_start(struct session *se, const char *id,
1182                                          struct record_cluster **prev_r,
1183                                          struct record_cluster **next_r)
1184 {
1185     struct record_cluster *r = 0;
1186
1187     session_enter(se);
1188     *prev_r = 0;
1189     *next_r = 0;
1190     if (se->reclist)
1191     {
1192         reclist_enter(se->reclist);
1193         while ((r = reclist_read_record(se->reclist)))
1194         {
1195             if (!strcmp(r->recid, id))
1196             {
1197                 *next_r = reclist_read_record(se->reclist);
1198                 break;
1199             }
1200             *prev_r = r;
1201         }
1202         reclist_leave(se->reclist);
1203     }
1204     if (!r)
1205         session_leave(se);
1206     return r;
1207 }
1208
1209 void show_single_stop(struct session *se, struct record_cluster *rec)
1210 {
1211     session_leave(se);
1212 }
1213
1214 struct record_cluster **show_range_start(struct session *se,
1215                                          struct reclist_sortparms *sp,
1216                                          int start, int *num, int *total, Odr_int *sumhits, Odr_int *approx_hits)
1217 {
1218     struct record_cluster **recs;
1219     struct reclist_sortparms *spp;
1220     int i;
1221 #if USE_TIMING
1222     yaz_timing_t t = yaz_timing_create();
1223 #endif
1224     session_enter(se);
1225     recs = nmem_malloc(se->nmem, *num * sizeof(struct record_cluster *));
1226     if (!se->relevance)
1227     {
1228         *num = 0;
1229         *total = 0;
1230         *sumhits = 0;
1231         *approx_hits = 0;
1232         recs = 0;
1233     }
1234     else
1235     {
1236         struct client_list *l;
1237
1238         for (spp = sp; spp; spp = spp->next)
1239             if (spp->type == Metadata_sortkey_relevance)
1240             {
1241                 relevance_prepare_read(se->relevance, se->reclist);
1242                 break;
1243             }
1244         reclist_sort(se->reclist, sp);
1245
1246         reclist_enter(se->reclist);
1247         *total = reclist_get_num_records(se->reclist);
1248
1249         *sumhits = 0;
1250         *approx_hits = 0;
1251         for (l = se->clients_active; l; l = l->next) {
1252             *sumhits += client_get_hits(l->client);
1253             *approx_hits += client_get_approximation(l->client);
1254         }
1255         for (i = 0; i < start; i++)
1256             if (!reclist_read_record(se->reclist))
1257             {
1258                 *num = 0;
1259                 recs = 0;
1260                 break;
1261             }
1262
1263         for (i = 0; i < *num; i++)
1264         {
1265             struct record_cluster *r = reclist_read_record(se->reclist);
1266             if (!r)
1267             {
1268                 *num = i;
1269                 break;
1270             }
1271             recs[i] = r;
1272         }
1273         reclist_leave(se->reclist);
1274     }
1275 #if USE_TIMING
1276     yaz_timing_stop(t);
1277     yaz_log(YLOG_LOG, "show %6.5f %3.2f %3.2f",
1278             yaz_timing_get_real(t), yaz_timing_get_user(t),
1279             yaz_timing_get_sys(t));
1280     yaz_timing_destroy(&t);
1281 #endif
1282     return recs;
1283 }
1284
1285 void show_range_stop(struct session *se, struct record_cluster **recs)
1286 {
1287     session_leave(se);
1288 }
1289
1290 void statistics(struct session *se, struct statistics *stat)
1291 {
1292     struct client_list *l;
1293     int count = 0;
1294
1295     memset(stat, 0, sizeof(*stat));
1296     stat->num_hits = 0;
1297     for (l = se->clients_active; l; l = l->next)
1298     {
1299         struct client *cl = l->client;
1300         if (!client_get_connection(cl))
1301             stat->num_no_connection++;
1302         stat->num_hits += client_get_hits(cl);
1303         switch (client_get_state(cl))
1304         {
1305         case Client_Connecting: stat->num_connecting++; break;
1306         case Client_Working: stat->num_working++; break;
1307         case Client_Idle: stat->num_idle++; break;
1308         case Client_Failed: stat->num_failed++; break;
1309         case Client_Error: stat->num_error++; break;
1310         default: break;
1311         }
1312         count++;
1313     }
1314     stat->num_records = se->total_records;
1315
1316     stat->num_clients = count;
1317 }
1318
1319 static struct record_metadata *record_metadata_init(
1320     NMEM nmem, const char *value, enum conf_metadata_type type,
1321     struct _xmlAttr *attr)
1322 {
1323     struct record_metadata *rec_md = record_metadata_create(nmem);
1324     struct record_metadata_attr **attrp = &rec_md->attributes;
1325
1326     for (; attr; attr = attr->next)
1327     {
1328         if (attr->children && attr->children->content)
1329         {
1330             if (strcmp((const char *) attr->name, "type"))
1331             {  /* skip the "type" attribute.. Its value is already part of
1332                   the element in output (md-%s) and so repeating it here
1333                   is redundant */
1334                 *attrp = nmem_malloc(nmem, sizeof(**attrp));
1335                 (*attrp)->name =
1336                     nmem_strdup(nmem, (const char *) attr->name);
1337                 (*attrp)->value =
1338                     nmem_strdup(nmem, (const char *) attr->children->content);
1339                 attrp = &(*attrp)->next;
1340             }
1341         }
1342     }
1343     *attrp = 0;
1344
1345     if (type == Metadata_type_generic)
1346     {
1347         char *p = nmem_strdup(nmem, value);
1348
1349         p = normalize7bit_generic(p, " ,/.:([");
1350
1351         rec_md->data.text.disp = p;
1352         rec_md->data.text.sort = 0;
1353     }
1354     else if (type == Metadata_type_year || type == Metadata_type_date)
1355     {
1356         int first, last;
1357         int longdate = 0;
1358
1359         if (type == Metadata_type_date)
1360             longdate = 1;
1361         if (extract7bit_dates((char *) value, &first, &last, longdate) < 0)
1362             return 0;
1363
1364         rec_md->data.number.min = first;
1365         rec_md->data.number.max = last;
1366     }
1367     else
1368         return 0;
1369     return rec_md;
1370 }
1371
1372 static int get_mergekey_from_doc(xmlDoc *doc, xmlNode *root, const char *name,
1373                                  struct conf_service *service, WRBUF norm_wr)
1374 {
1375     xmlNode *n;
1376     int no_found = 0;
1377     for (n = root->children; n; n = n->next)
1378     {
1379         if (n->type != XML_ELEMENT_NODE)
1380             continue;
1381         if (!strcmp((const char *) n->name, "metadata"))
1382         {
1383             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1384             if (type == NULL) {
1385                 yaz_log(YLOG_FATAL, "Missing type attribute on metadata element. Skipping!");
1386             }
1387             else if (!strcmp(name, (const char *) type))
1388             {
1389                 xmlChar *value = xmlNodeListGetString(doc, n->children, 1);
1390                 if (value)
1391                 {
1392                     const char *norm_str;
1393                     pp2_charset_token_t prt =
1394                         pp2_charset_token_create(service->charsets, "mergekey");
1395
1396                     pp2_charset_token_first(prt, (const char *) value, 0);
1397                     if (wrbuf_len(norm_wr) > 0)
1398                         wrbuf_puts(norm_wr, " ");
1399                     wrbuf_puts(norm_wr, name);
1400                     while ((norm_str =
1401                             pp2_charset_token_next(prt)))
1402                     {
1403                         if (*norm_str)
1404                         {
1405                             wrbuf_puts(norm_wr, " ");
1406                             wrbuf_puts(norm_wr, norm_str);
1407                         }
1408                     }
1409                     xmlFree(value);
1410                     pp2_charset_token_destroy(prt);
1411                     no_found++;
1412                 }
1413             }
1414             xmlFree(type);
1415         }
1416     }
1417     return no_found;
1418 }
1419
1420 static const char *get_mergekey(xmlDoc *doc, struct client *cl, int record_no,
1421                                 struct conf_service *service, NMEM nmem)
1422 {
1423     char *mergekey_norm = 0;
1424     xmlNode *root = xmlDocGetRootElement(doc);
1425     WRBUF norm_wr = wrbuf_alloc();
1426
1427     /* consider mergekey from XSL first */
1428     xmlChar *mergekey = xmlGetProp(root, (xmlChar *) "mergekey");
1429     if (mergekey)
1430     {
1431         const char *norm_str;
1432         pp2_charset_token_t prt =
1433             pp2_charset_token_create(service->charsets, "mergekey");
1434
1435         pp2_charset_token_first(prt, (const char *) mergekey, 0);
1436         while ((norm_str = pp2_charset_token_next(prt)))
1437         {
1438             if (*norm_str)
1439             {
1440                 if (wrbuf_len(norm_wr))
1441                     wrbuf_puts(norm_wr, " ");
1442                 wrbuf_puts(norm_wr, norm_str);
1443             }
1444         }
1445         pp2_charset_token_destroy(prt);
1446         xmlFree(mergekey);
1447     }
1448     else
1449     {
1450         /* no mergekey defined in XSL. Look for mergekey metadata instead */
1451         int field_id;
1452         for (field_id = 0; field_id < service->num_metadata; field_id++)
1453         {
1454             struct conf_metadata *ser_md = &service->metadata[field_id];
1455             if (ser_md->mergekey != Metadata_mergekey_no)
1456             {
1457                 int r = get_mergekey_from_doc(doc, root, ser_md->name,
1458                                               service, norm_wr);
1459                 if (r == 0 && ser_md->mergekey == Metadata_mergekey_required)
1460                 {
1461                     /* no mergekey on this one and it is required..
1462                        Generate unique key instead */
1463                     wrbuf_rewind(norm_wr);
1464                     break;
1465                 }
1466             }
1467         }
1468     }
1469
1470     /* generate unique key if none is not generated already or is empty */
1471     if (wrbuf_len(norm_wr) == 0)
1472     {
1473         wrbuf_printf(norm_wr, "position: %s-%d",
1474                      client_get_id(cl), record_no);
1475     }
1476     else
1477     {
1478         const char *lead = "content: ";
1479         wrbuf_insert(norm_wr, 0, lead, strlen(lead));
1480     }
1481     if (wrbuf_len(norm_wr) > 0)
1482         mergekey_norm = nmem_strdup(nmem, wrbuf_cstr(norm_wr));
1483     wrbuf_destroy(norm_wr);
1484     return mergekey_norm;
1485 }
1486
1487 /** \brief see if metadata for pz:recordfilter exists
1488     \param root xml root element of normalized record
1489     \param sdb session database for client
1490     \retval 0 if there is no metadata for pz:recordfilter
1491     \retval 1 if there is metadata for pz:recordfilter
1492
1493     If there is no pz:recordfilter defined, this function returns 1
1494     as well.
1495 */
1496
1497 static int check_record_filter(xmlNode *root, struct session_database *sdb)
1498 {
1499     int match = 0;
1500     xmlNode *n;
1501     const char *s;
1502     s = session_setting_oneval(sdb, PZ_RECORDFILTER);
1503
1504     if (!s || !*s)
1505         return 1;
1506
1507     for (n = root->children; n; n = n->next)
1508     {
1509         if (n->type != XML_ELEMENT_NODE)
1510             continue;
1511         if (!strcmp((const char *) n->name, "metadata"))
1512         {
1513             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1514             if (type)
1515             {
1516                 size_t len;
1517                 int substring;
1518                 const char *eq;
1519
1520                 if ((eq = strchr(s, '=')))
1521                     substring = 0;
1522                 else if ((eq = strchr(s, '~')))
1523                     substring = 1;
1524                 if (eq)
1525                     len = eq - s;
1526                 else
1527                     len = strlen(s);
1528                 if (len == strlen((const char *)type) &&
1529                     !memcmp((const char *) type, s, len))
1530                 {
1531                     xmlChar *value = xmlNodeGetContent(n);
1532                     if (value && *value)
1533                     {
1534                         if (!eq ||
1535                             (substring && strstr((const char *) value, eq+1)) ||
1536                             (!substring && !strcmp((const char *) value, eq + 1)))
1537                             match = 1;
1538                     }
1539                     xmlFree(value);
1540                 }
1541                 xmlFree(type);
1542             }
1543         }
1544     }
1545     return match;
1546 }
1547
1548
1549 static int ingest_to_cluster(struct client *cl,
1550                              xmlDoc *xdoc,
1551                              xmlNode *root,
1552                              int record_no,
1553                              const char *mergekey_norm);
1554
1555 /** \brief ingest XML record
1556     \param cl client holds the result set for record
1557     \param rec record buffer (0 terminated)
1558     \param record_no record position (1, 2, ..)
1559     \param nmem working NMEM
1560     \retval 0 OK
1561     \retval -1 failure
1562     \retval -2 Filtered
1563 */
1564 int ingest_record(struct client *cl, const char *rec,
1565                   int record_no, NMEM nmem)
1566 {
1567     struct session *se = client_get_session(cl);
1568     int ret = 0;
1569     struct session_database *sdb = client_get_database(cl);
1570     struct conf_service *service = se->service;
1571     xmlDoc *xdoc = normalize_record(se, sdb, service, rec, nmem);
1572     xmlNode *root;
1573     const char *mergekey_norm;
1574
1575     if (!xdoc)
1576         return -1;
1577
1578     root = xmlDocGetRootElement(xdoc);
1579
1580     if (!check_record_filter(root, sdb))
1581     {
1582         session_log(se, YLOG_LOG, "Filtered out record no %d from %s", record_no, sdb->database->id);
1583         xmlFreeDoc(xdoc);
1584         return -2;
1585     }
1586
1587     mergekey_norm = get_mergekey(xdoc, cl, record_no, service, nmem);
1588     if (!mergekey_norm)
1589     {
1590         session_log(se, YLOG_WARN, "Got no mergekey");
1591         xmlFreeDoc(xdoc);
1592         return -1;
1593     }
1594     session_enter(se);
1595     if (client_get_session(cl) == se)
1596         ret = ingest_to_cluster(cl, xdoc, root, record_no, mergekey_norm);
1597     session_leave(se);
1598
1599     xmlFreeDoc(xdoc);
1600     return ret;
1601 }
1602
1603 // Skip record on non-zero
1604 static int check_limit_local(struct client *cl,
1605                              struct record *record,
1606                              int record_no)
1607 {
1608     int skip_record = 0;
1609     struct session *se = client_get_session(cl);
1610     struct conf_service *service = se->service;
1611     NMEM nmem_tmp = nmem_create();
1612     struct session_database *sdb = client_get_database(cl);
1613     int l = 0;
1614     while (!skip_record)
1615     {
1616         struct conf_metadata *ser_md = 0;
1617         struct record_metadata *rec_md = 0;
1618         int md_field_id;
1619         char **values = 0;
1620         int i, num_v = 0;
1621
1622         const char *name = client_get_facet_limit_local(cl, sdb, &l, nmem_tmp, &num_v, &values);
1623         if (!name)
1624             break;
1625
1626         md_field_id = conf_service_metadata_field_id(service, name);
1627         if (md_field_id < 0)
1628         {
1629             skip_record = 1;
1630             break;
1631         }
1632         ser_md = &service->metadata[md_field_id];
1633         rec_md = record->metadata[md_field_id];
1634         yaz_log(YLOG_DEBUG, "check limit local %s", name);
1635         for (i = 0; i < num_v; )
1636         {
1637             if (rec_md)
1638             {
1639                 if (ser_md->type == Metadata_type_year
1640                     || ser_md->type == Metadata_type_date)
1641                 {
1642                     int y = atoi(values[i]);
1643                     if (y >= rec_md->data.number.min
1644                         && y <= rec_md->data.number.max)
1645                         break;
1646                 }
1647                 else
1648                 {
1649                     yaz_log(YLOG_DEBUG, "cmp: '%s' '%s'", rec_md->data.text.disp, values[i]);
1650                     if (!strcmp(rec_md->data.text.disp, values[i]))
1651                     {
1652                         // Value equals, should not be filtered.
1653                         break;
1654                     }
1655                 }
1656                 rec_md = rec_md->next;
1657             }
1658             else
1659             {
1660                 rec_md = record->metadata[md_field_id];
1661                 i++;
1662             }
1663         }
1664         // At end , not match
1665         if (i == num_v)
1666         {
1667             skip_record = 1;
1668             break;
1669         }
1670     }
1671     nmem_destroy(nmem_tmp);
1672     return skip_record;
1673 }
1674
1675 static int ingest_to_cluster(struct client *cl,
1676                              xmlDoc *xdoc,
1677                              xmlNode *root,
1678                              int record_no,
1679                              const char *mergekey_norm)
1680 {
1681     xmlNode *n;
1682     xmlChar *type = 0;
1683     xmlChar *value = 0;
1684     struct session *se = client_get_session(cl);
1685     struct conf_service *service = se->service;
1686     int term_factor = 1;
1687     struct record_cluster *cluster;
1688     struct session_database *sdb = client_get_database(cl);
1689     struct record *record = record_create(se->nmem,
1690                                           service->num_metadata,
1691                                           service->num_sortkeys, cl,
1692                                           record_no);
1693
1694     for (n = root->children; n; n = n->next)
1695     {
1696         if (type)
1697             xmlFree(type);
1698         if (value)
1699             xmlFree(value);
1700         type = value = 0;
1701
1702         if (n->type != XML_ELEMENT_NODE)
1703             continue;
1704         if (!strcmp((const char *) n->name, "metadata"))
1705         {
1706             struct conf_metadata *ser_md = 0;
1707             struct record_metadata **wheretoput = 0;
1708             struct record_metadata *rec_md = 0;
1709             int md_field_id = -1;
1710
1711             type = xmlGetProp(n, (xmlChar *) "type");
1712             value = xmlNodeListGetString(xdoc, n->children, 1);
1713
1714             if (!type || !value || !*value)
1715                 continue;
1716
1717             md_field_id
1718                 = conf_service_metadata_field_id(service, (const char *) type);
1719             if (md_field_id < 0)
1720             {
1721                 if (se->number_of_warnings_unknown_metadata == 0)
1722                 {
1723                     session_log(se, YLOG_WARN,
1724                             "Ignoring unknown metadata element: %s", type);
1725                 }
1726                 se->number_of_warnings_unknown_metadata++;
1727                 continue;
1728             }
1729
1730             ser_md = &service->metadata[md_field_id];
1731
1732             // non-merged metadata
1733             rec_md = record_metadata_init(se->nmem, (const char *) value,
1734                                           ser_md->type, n->properties);
1735             if (!rec_md)
1736             {
1737                 session_log(se, YLOG_WARN, "bad metadata data '%s' "
1738                             "for element '%s'", value, type);
1739                 continue;
1740             }
1741             wheretoput = &record->metadata[md_field_id];
1742             while (*wheretoput)
1743                 wheretoput = &(*wheretoput)->next;
1744             *wheretoput = rec_md;
1745         }
1746     }
1747
1748     if (check_limit_local(cl, record, record_no))
1749     {
1750         session_log(se, YLOG_LOG, "Facet filtered out record no %d from %s",
1751                     record_no, sdb->database->id);
1752         if (type)
1753             xmlFree(type);
1754         if (value)
1755             xmlFree(value);
1756         return -2;
1757     }
1758     cluster = reclist_insert(se->reclist, service, record,
1759                              mergekey_norm, &se->total_merged);
1760     if (!cluster)
1761         return -1;
1762
1763     {
1764         const char *use_term_factor_str =
1765             session_setting_oneval(sdb, PZ_TERMLIST_TERM_FACTOR);
1766         if (use_term_factor_str && use_term_factor_str[0] == '1')
1767         {
1768             int maxrecs = client_get_maxrecs(cl);
1769             int hits = (int) client_get_hits(cl);
1770             term_factor = MAX(hits, maxrecs) /  MAX(1, maxrecs);
1771             assert(term_factor >= 1);
1772             yaz_log(YLOG_DEBUG, "Using term factor: %d (%d / %d)", term_factor, MAX(hits, maxrecs), MAX(1, maxrecs));
1773         }
1774     }
1775
1776     if (global_parameters.dump_records)
1777         session_log(se, YLOG_LOG, "Cluster id %s from %s (#%d)", cluster->recid,
1778                     sdb->database->id, record_no);
1779
1780
1781     relevance_newrec(se->relevance, cluster);
1782
1783     // now parsing XML record and adding data to cluster or record metadata
1784     for (n = root->children; n; n = n->next)
1785     {
1786         pp2_charset_token_t prt;
1787         if (type)
1788             xmlFree(type);
1789         if (value)
1790             xmlFree(value);
1791         type = value = 0;
1792
1793         if (n->type != XML_ELEMENT_NODE)
1794             continue;
1795         if (!strcmp((const char *) n->name, "metadata"))
1796         {
1797             struct conf_metadata *ser_md = 0;
1798             struct conf_sortkey *ser_sk = 0;
1799             struct record_metadata **wheretoput = 0;
1800             struct record_metadata *rec_md = 0;
1801             int md_field_id = -1;
1802             int sk_field_id = -1;
1803             const char *rank;
1804             xmlChar *xml_rank;
1805
1806             type = xmlGetProp(n, (xmlChar *) "type");
1807             value = xmlNodeListGetString(xdoc, n->children, 1);
1808
1809             if (!type || !value || !*value)
1810                 continue;
1811
1812             md_field_id
1813                 = conf_service_metadata_field_id(service, (const char *) type);
1814             if (md_field_id < 0)
1815                 continue;
1816
1817             ser_md = &service->metadata[md_field_id];
1818
1819             if (ser_md->sortkey_offset >= 0)
1820             {
1821                 sk_field_id = ser_md->sortkey_offset;
1822                 ser_sk = &service->sortkeys[sk_field_id];
1823             }
1824
1825             // merged metadata
1826             rec_md = record_metadata_init(se->nmem, (const char *) value,
1827                                           ser_md->type, 0);
1828             if (!rec_md)
1829                 continue;
1830
1831             xml_rank = xmlGetProp(n, (xmlChar *) "rank");
1832             rank = xml_rank ? (const char *) xml_rank : ser_md->rank;
1833
1834             wheretoput = &cluster->metadata[md_field_id];
1835
1836             // and polulate with data:
1837             // assign cluster or record based on merge action
1838             if (ser_md->merge == Metadata_merge_unique)
1839             {
1840                 while (*wheretoput)
1841                 {
1842                     if (!strcmp((const char *) (*wheretoput)->data.text.disp,
1843                                 rec_md->data.text.disp))
1844                         break;
1845                     wheretoput = &(*wheretoput)->next;
1846                 }
1847                 if (!*wheretoput)
1848                     *wheretoput = rec_md;
1849             }
1850             else if (ser_md->merge == Metadata_merge_longest)
1851             {
1852                 if (!*wheretoput
1853                     || strlen(rec_md->data.text.disp)
1854                     > strlen((*wheretoput)->data.text.disp))
1855                 {
1856                     *wheretoput = rec_md;
1857                     if (ser_sk)
1858                     {
1859                         const char *sort_str = 0;
1860                         int skip_article =
1861                             ser_sk->type == Metadata_sortkey_skiparticle;
1862
1863                         if (!cluster->sortkeys[sk_field_id])
1864                             cluster->sortkeys[sk_field_id] =
1865                                 nmem_malloc(se->nmem,
1866                                             sizeof(union data_types));
1867
1868                         prt =
1869                             pp2_charset_token_create(service->charsets, "sort");
1870
1871                         pp2_charset_token_first(prt, rec_md->data.text.disp,
1872                                                 skip_article);
1873
1874                         pp2_charset_token_next(prt);
1875
1876                         sort_str = pp2_get_sort(prt);
1877
1878                         cluster->sortkeys[sk_field_id]->text.disp =
1879                             rec_md->data.text.disp;
1880                         if (!sort_str)
1881                         {
1882                             sort_str = rec_md->data.text.disp;
1883                             session_log(se, YLOG_WARN,
1884                                     "Could not make sortkey. Bug #1858");
1885                         }
1886                         cluster->sortkeys[sk_field_id]->text.sort =
1887                             nmem_strdup(se->nmem, sort_str);
1888                         pp2_charset_token_destroy(prt);
1889                     }
1890                 }
1891             }
1892             else if (ser_md->merge == Metadata_merge_all)
1893             {
1894                 while (*wheretoput)
1895                     wheretoput = &(*wheretoput)->next;
1896                 *wheretoput = rec_md;
1897             }
1898             else if (ser_md->merge == Metadata_merge_range)
1899             {
1900                 if (!*wheretoput)
1901                 {
1902                     *wheretoput = rec_md;
1903                     if (ser_sk)
1904                         cluster->sortkeys[sk_field_id]
1905                             = &rec_md->data;
1906                 }
1907                 else
1908                 {
1909                     int this_min = rec_md->data.number.min;
1910                     int this_max = rec_md->data.number.max;
1911                     if (this_min < (*wheretoput)->data.number.min)
1912                         (*wheretoput)->data.number.min = this_min;
1913                     if (this_max > (*wheretoput)->data.number.max)
1914                         (*wheretoput)->data.number.max = this_max;
1915                 }
1916             }
1917
1918             // ranking of _all_ fields enabled ...
1919             if (rank)
1920             {
1921                 relevance_countwords(se->relevance, cluster,
1922                                      (char *) value, rank, ser_md->name);
1923             }
1924
1925             // construct facets ... unless the client already has reported them
1926             if (ser_md->termlist && !client_has_facet(cl, (char *) type))
1927             {
1928                 if (ser_md->type == Metadata_type_year)
1929                 {
1930                     char year[64];
1931                     sprintf(year, "%d", rec_md->data.number.max);
1932
1933                     add_facet(se, (char *) type, year, term_factor);
1934                     if (rec_md->data.number.max != rec_md->data.number.min)
1935                     {
1936                         sprintf(year, "%d", rec_md->data.number.min);
1937                         add_facet(se, (char *) type, year, term_factor);
1938                     }
1939                 }
1940                 else
1941                     add_facet(se, (char *) type, (char *) value, term_factor);
1942             }
1943
1944             // cleaning up
1945             if (xml_rank)
1946                 xmlFree(xml_rank);
1947             xmlFree(type);
1948             xmlFree(value);
1949             type = value = 0;
1950         }
1951         else
1952         {
1953             if (se->number_of_warnings_unknown_elements == 0)
1954                 session_log(se, YLOG_WARN,
1955                         "Unexpected element in internal record: %s", n->name);
1956             se->number_of_warnings_unknown_elements++;
1957         }
1958     }
1959     if (type)
1960         xmlFree(type);
1961     if (value)
1962         xmlFree(value);
1963
1964     relevance_donerecord(se->relevance, cluster);
1965     se->total_records++;
1966
1967     return 0;
1968 }
1969
1970 void session_log(struct session *s, int level, const char *fmt, ...)
1971 {
1972     char buf[1024];
1973     va_list ap;
1974     va_start(ap, fmt);
1975
1976     yaz_vsnprintf(buf, sizeof(buf)-30, fmt, ap);
1977     yaz_log(level, "Session %u: %s", s->session_id, buf);
1978
1979     va_end(ap);
1980 }
1981
1982 /*
1983  * Local variables:
1984  * c-basic-offset: 4
1985  * c-file-style: "Stroustrup"
1986  * indent-tabs-mode: nil
1987  * End:
1988  * vim: shiftwidth=4 tabstop=8 expandtab
1989  */
1990