1927a1c13019ba20362af2f12b680080da038057
[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
60 #define USE_TIMING 0
61 #if USE_TIMING
62 #include <yaz/timing.h>
63 #endif
64
65 #include "ppmutex.h"
66 #include "parameters.h"
67 #include "session.h"
68 #include "eventl.h"
69 #include "http.h"
70 #include "termlists.h"
71 #include "reclists.h"
72 #include "relevance.h"
73 #include "database.h"
74 #include "client.h"
75 #include "settings.h"
76 #include "normalize7bit.h"
77
78 #include <libxml/tree.h>
79
80 #define MAX_CHUNK 15
81
82 #define MAX(a,b) ((a)>(b)?(a):(b))
83
84 // Note: Some things in this structure will eventually move to configuration
85 struct parameters global_parameters =
86 {
87     0,   // dump_records
88     0,   // debug_mode
89     0,   // predictable sessions
90 };
91
92 struct client_list {
93     struct client *client;
94     struct client_list *next;
95 };
96
97 /* session counting (1) , disable client counting (0) */
98 static YAZ_MUTEX g_session_mutex = 0;
99 static int no_sessions = 0;
100 static int no_session_total = 0;
101
102 static int session_use(int delta)
103 {
104     int sessions;
105     if (!g_session_mutex)
106         yaz_mutex_create(&g_session_mutex);
107     yaz_mutex_enter(g_session_mutex);
108     no_sessions += delta;
109     if (delta > 0)
110         no_session_total += delta;
111     sessions = no_sessions;
112     yaz_mutex_leave(g_session_mutex);
113     yaz_log(YLOG_DEBUG, "%s sessions=%d", delta == 0 ? "" : (delta > 0 ? "INC" : "DEC"), no_sessions);
114     return sessions;
115 }
116
117 int sessions_count(void)
118 {
119     return session_use(0);
120 }
121
122 int session_count_total(void)
123 {
124     int total = 0;
125     if (!g_session_mutex)
126         return 0;
127     yaz_mutex_enter(g_session_mutex);
128     total = no_session_total;
129     yaz_mutex_leave(g_session_mutex);
130     return total;
131 }
132
133 static void log_xml_doc(xmlDoc *doc)
134 {
135     FILE *lf = yaz_log_file();
136     xmlChar *result = 0;
137     int len = 0;
138 #if LIBXML_VERSION >= 20600
139     xmlDocDumpFormatMemory(doc, &result, &len, 1);
140 #else
141     xmlDocDumpMemory(doc, &result, &len);
142 #endif
143     if (lf && len)
144     {
145         (void) fwrite(result, 1, len, lf);
146         fprintf(lf, "\n");
147     }
148     xmlFree(result);
149 }
150
151 static void session_enter(struct session *s, const char *caller)
152 {
153     if (caller)
154         session_log(s, YLOG_DEBUG, "Session lock by %s", caller);
155     yaz_mutex_enter(s->session_mutex);
156 }
157
158 static void session_leave(struct session *s, const char *caller)
159 {
160     yaz_mutex_leave(s->session_mutex);
161     if (caller)
162         session_log(s, YLOG_DEBUG, "Session unlock by %s", caller);
163 }
164
165 static void session_normalize_facet(struct session *s, const char *type,
166                                     const char *value,
167                                     WRBUF display_wrbuf,
168                                     WRBUF facet_wrbuf)
169 {
170     struct conf_service *service = s->service;
171     pp2_charset_token_t prt;
172     const char *facet_component;
173     int i;
174     const char *icu_chain_id = 0;
175
176     for (i = 0; i < service->num_metadata; i++)
177         if (!strcmp((service->metadata + i)->name, type))
178             icu_chain_id = (service->metadata + i)->facetrule;
179     if (!icu_chain_id)
180         icu_chain_id = "facet";
181     prt = pp2_charset_token_create(service->charsets, icu_chain_id);
182     if (!prt)
183     {
184         yaz_log(YLOG_FATAL, "Unknown ICU chain '%s' for facet of type '%s'",
185                 icu_chain_id, type);
186         wrbuf_destroy(facet_wrbuf);
187         wrbuf_destroy(display_wrbuf);
188         return;
189     }
190     pp2_charset_token_first(prt, value, 0);
191     while ((facet_component = pp2_charset_token_next(prt)))
192     {
193         const char *display_component;
194         if (*facet_component)
195         {
196             if (wrbuf_len(facet_wrbuf))
197                 wrbuf_puts(facet_wrbuf, " ");
198             wrbuf_puts(facet_wrbuf, facet_component);
199         }
200         display_component = pp2_get_display(prt);
201         if (display_component)
202         {
203             if (wrbuf_len(display_wrbuf))
204                 wrbuf_puts(display_wrbuf, " ");
205             wrbuf_puts(display_wrbuf, display_component);
206         }
207     }
208     pp2_charset_token_destroy(prt);
209 }
210
211 void add_facet(struct session *s, const char *type, const char *value, int count)
212 {
213     WRBUF facet_wrbuf = wrbuf_alloc();
214     WRBUF display_wrbuf = wrbuf_alloc();
215
216     session_normalize_facet(s, type, value, display_wrbuf, facet_wrbuf);
217
218     if (wrbuf_len(facet_wrbuf))
219     {
220         int i;
221         for (i = 0; i < s->num_termlists; i++)
222             if (!strcmp(s->termlists[i].name, type))
223                 break;
224         if (i == s->num_termlists)
225         {
226             if (i == SESSION_MAX_TERMLISTS)
227             {
228                 session_log(s, YLOG_FATAL, "Too many termlists");
229                 wrbuf_destroy(facet_wrbuf);
230                 wrbuf_destroy(display_wrbuf);
231                 return;
232             }
233
234             s->termlists[i].name = nmem_strdup(s->nmem, type);
235             s->termlists[i].termlist = termlist_create(s->nmem);
236             s->num_termlists = i + 1;
237         }
238
239 #if 0
240         session_log(s, YLOG_LOG, "Facets for %s: %s norm:%s (%d)", type, value, wrbuf_cstr(facet_wrbuf), count);
241 #endif
242         termlist_insert(s->termlists[i].termlist, wrbuf_cstr(display_wrbuf),
243                         wrbuf_cstr(facet_wrbuf), count);
244     }
245     wrbuf_destroy(facet_wrbuf);
246     wrbuf_destroy(display_wrbuf);
247 }
248
249 static xmlDoc *record_to_xml(struct session *se,
250                              struct session_database *sdb, const char *rec)
251 {
252     struct database *db = sdb->database;
253     xmlDoc *rdoc = 0;
254
255     rdoc = xmlParseMemory(rec, strlen(rec));
256
257     if (!rdoc)
258     {
259         session_log(se, YLOG_WARN, "Non-wellformed XML");
260         return 0;
261     }
262
263     if (global_parameters.dump_records)
264     {
265         session_log(se, YLOG_LOG, "Un-normalized record from %s", db->id);
266         log_xml_doc(rdoc);
267     }
268
269     return rdoc;
270 }
271
272 #define MAX_XSLT_ARGS 16
273
274 // Add static values from session database settings if applicable
275 static void insert_settings_parameters(struct session_database *sdb,
276                                        struct conf_service *service,
277                                        char **parms,
278                                        NMEM nmem)
279 {
280     int i;
281     int nparms = 0;
282     int offset = 0;
283
284     for (i = 0; i < service->num_metadata; i++)
285     {
286         struct conf_metadata *md = &service->metadata[i];
287         int setting;
288
289         if (md->setting == Metadata_setting_parameter &&
290             (setting = settings_lookup_offset(service, md->name)) >= 0)
291         {
292             const char *val = session_setting_oneval(sdb, setting);
293             if (val && nparms < MAX_XSLT_ARGS)
294             {
295                 char *buf;
296                 int len = strlen(val);
297                 buf = nmem_malloc(nmem, len + 3);
298                 buf[0] = '\'';
299                 strcpy(buf + 1, val);
300                 buf[len+1] = '\'';
301                 buf[len+2] = '\0';
302                 parms[offset++] = md->name;
303                 parms[offset++] = buf;
304                 nparms++;
305             }
306         }
307     }
308     parms[offset] = 0;
309 }
310
311 // Add static values from session database settings if applicable
312 static void insert_settings_values(struct session_database *sdb, xmlDoc *doc,
313     struct conf_service *service)
314 {
315     int i;
316
317     for (i = 0; i < service->num_metadata; i++)
318     {
319         struct conf_metadata *md = &service->metadata[i];
320         int offset;
321
322         if (md->setting == Metadata_setting_postproc &&
323             (offset = settings_lookup_offset(service, md->name)) >= 0)
324         {
325             const char *val = session_setting_oneval(sdb, offset);
326             if (val)
327             {
328                 xmlNode *r = xmlDocGetRootElement(doc);
329                 xmlNode *n = xmlNewTextChild(r, 0, (xmlChar *) "metadata",
330                                              (xmlChar *) val);
331                 xmlSetProp(n, (xmlChar *) "type", (xmlChar *) md->name);
332             }
333         }
334     }
335 }
336
337 static xmlDoc *normalize_record(struct session *se,
338                                 struct session_database *sdb,
339                                 struct conf_service *service,
340                                 const char *rec, NMEM nmem)
341 {
342     xmlDoc *rdoc = record_to_xml(se, sdb, rec);
343
344     if (rdoc)
345     {
346         char *parms[MAX_XSLT_ARGS*2+1];
347
348         insert_settings_parameters(sdb, service, parms, nmem);
349
350         if (normalize_record_transform(sdb->map, &rdoc, (const char **)parms))
351         {
352             session_log(se, YLOG_WARN, "Normalize failed");
353         }
354         else
355         {
356             insert_settings_values(sdb, rdoc, service);
357
358             if (global_parameters.dump_records)
359             {
360                 session_log(se, YLOG_LOG, "Normalized record from %s",
361                             sdb->database->id);
362                 log_xml_doc(rdoc);
363             }
364         }
365     }
366     return rdoc;
367 }
368
369 void session_settings_dump(struct session *se,
370                            struct session_database *db,
371                            WRBUF w)
372 {
373     if (db->settings)
374     {
375         int i, num = db->num_settings;
376         for (i = 0; i < num; i++)
377         {
378             struct setting *s = db->settings[i];
379             for (;s ; s = s->next)
380             {
381                 wrbuf_puts(w, "<set name=\"");
382                 wrbuf_xmlputs(w, s->name);
383                 wrbuf_puts(w, "\" value=\"");
384                 wrbuf_xmlputs(w, s->value);
385                 wrbuf_puts(w, "\"/>");
386             }
387             if (db->settings[i])
388                 wrbuf_puts(w, "\n");
389         }
390     }
391 }
392
393 // Retrieve first defined value for 'name' for given database.
394 // Will be extended to take into account user associated with session
395 const char *session_setting_oneval(struct session_database *db, int offset)
396 {
397     if (offset >= db->num_settings || !db->settings[offset])
398         return "";
399     return db->settings[offset]->value;
400 }
401
402 // Prepare XSLT stylesheets for record normalization
403 // Structures are allocated on the session_wide nmem to avoid having
404 // to recompute this for every search. This would lead
405 // to leaking if a single session was to repeatedly change the PZ_XSLT
406 // setting. However, this is not a realistic use scenario.
407 static int prepare_map(struct session *se, struct session_database *sdb)
408 {
409     if (sdb->settings && !sdb->map)
410     {
411         const char *s;
412
413         if (sdb->settings[PZ_XSLT] &&
414             (s = session_setting_oneval(sdb, PZ_XSLT)))
415         {
416             char auto_stylesheet[256];
417
418             if (!strcmp(s, "auto"))
419             {
420                 const char *request_syntax = session_setting_oneval(
421                     sdb, PZ_REQUESTSYNTAX);
422                 if (request_syntax)
423                 {
424                     char *cp;
425                     yaz_snprintf(auto_stylesheet, sizeof(auto_stylesheet),
426                                  "%s.xsl", request_syntax);
427                     for (cp = auto_stylesheet; *cp; cp++)
428                     {
429                         /* deliberately only consider ASCII */
430                         if (*cp > 32 && *cp < 127)
431                             *cp = tolower(*cp);
432                     }
433                     s = auto_stylesheet;
434                 }
435                 else
436                 {
437                     session_log(se, YLOG_WARN,
438                                 "No pz:requestsyntax for auto stylesheet");
439                 }
440             }
441             sdb->map = normalize_cache_get(se->normalize_cache,
442                                            se->service, s);
443             if (!sdb->map)
444                 return -1;
445         }
446     }
447     return 0;
448 }
449
450 // called if watch should be removed because http_channel is to be destroyed
451 static void session_watch_cancel(void *data, struct http_channel *c,
452                                  void *data2)
453 {
454     struct session_watchentry *ent = data;
455
456     ent->fun = 0;
457     ent->data = 0;
458     ent->obs = 0;
459 }
460
461 // set watch. Returns 0=OK, -1 if watch is already set
462 int session_set_watch(struct session *s, int what,
463                       session_watchfun fun, void *data,
464                       struct http_channel *chan)
465 {
466     int ret;
467     session_enter(s, "session_set_watch");
468     if (s->watchlist[what].fun)
469         ret = -1;
470     else
471     {
472
473         s->watchlist[what].fun = fun;
474         s->watchlist[what].data = data;
475         s->watchlist[what].obs = http_add_observer(chan, &s->watchlist[what],
476                                                    session_watch_cancel);
477         ret = 0;
478     }
479     session_leave(s, "session_set_watch");
480     return ret;
481 }
482
483 void session_alert_watch(struct session *s, int what)
484 {
485     assert(s);
486     session_enter(s, "session_alert_watch");
487     if (s->watchlist[what].fun)
488     {
489         /* our watch is no longer associated with http_channel */
490         void *data;
491         session_watchfun fun;
492
493         http_remove_observer(s->watchlist[what].obs);
494         fun  = s->watchlist[what].fun;
495         data = s->watchlist[what].data;
496
497         /* reset watch before fun is invoked - in case fun wants to set
498            it again */
499         s->watchlist[what].fun = 0;
500         s->watchlist[what].data = 0;
501         s->watchlist[what].obs = 0;
502
503         session_leave(s, "session_alert_watch");
504         session_log(s, YLOG_DEBUG,
505                     "Alert Watch: %d calling function: %p", what, fun);
506         fun(data);
507     }
508     else
509         session_leave(s,"session_alert_watch");
510 }
511
512 //callback for grep_databases
513 static void select_targets_callback(struct session *se,
514                                     struct session_database *db)
515 {
516     struct client *cl;
517     struct client_list *l;
518
519     for (l = se->clients_cached; l; l = l->next)
520         if (client_get_database(l->client) == db)
521             break;
522
523     if (l)
524         cl = l->client;
525     else
526     {
527         cl = client_create(db->database->id);
528         client_set_database(cl, db);
529
530         l = xmalloc(sizeof(*l));
531         l->client = cl;
532         l->next = se->clients_cached;
533         se->clients_cached = l;
534     }
535     /* set session always. If may be 0 if client is not active */
536     client_set_session(cl, se);
537
538     l = xmalloc(sizeof(*l));
539     l->client = cl;
540     l->next = se->clients_active;
541     se->clients_active = l;
542 }
543
544 static void session_reset_active_clients(struct session *se,
545                                          struct client_list *new_list)
546 {
547     struct client_list *l;
548
549     session_enter(se, "session_reset_active_clients");
550     l = se->clients_active;
551     se->clients_active = new_list;
552     session_leave(se, "session_reset_active_clients");
553
554     while (l)
555     {
556         struct client_list *l_next = l->next;
557
558         client_lock(l->client);
559         client_set_session(l->client, 0); /* mark client inactive */
560         client_unlock(l->client);
561
562         xfree(l);
563         l = l_next;
564     }
565 }
566
567 static void session_remove_cached_clients(struct session *se)
568 {
569     struct client_list *l;
570
571     session_reset_active_clients(se, 0);
572
573     session_enter(se, "session_remove_cached_clients");
574     l = se->clients_cached;
575     se->clients_cached = 0;
576     session_leave(se, "session_remove_cached_clients");
577
578     while (l)
579     {
580         struct client_list *l_next = l->next;
581         client_lock(l->client);
582         client_set_session(l->client, 0);
583         client_set_database(l->client, 0);
584         client_unlock(l->client);
585         client_destroy(l->client);
586         xfree(l);
587         l = l_next;
588     }
589 }
590
591 // Associates a set of clients with a session;
592 // Note: Session-databases represent databases with per-session
593 // setting overrides
594 static int select_targets(struct session *se, const char *filter)
595 {
596     return session_grep_databases(se, filter, select_targets_callback);
597 }
598
599 int session_active_clients(struct session *s)
600 {
601     struct client_list *l;
602     int res = 0;
603
604     for (l = s->clients_active; l; l = l->next)
605         if (client_is_active(l->client))
606             res++;
607
608     return res;
609 }
610
611 int session_is_preferred_clients_ready(struct session *s)
612 {
613     struct client_list *l;
614     int res = 0;
615
616     for (l = s->clients_active; l; l = l->next)
617         if (client_is_active_preferred(l->client))
618             res++;
619     session_log(s, YLOG_DEBUG, "Has %d active preferred clients.", res);
620     return res == 0;
621 }
622
623 static void session_clear_set(struct session *se, struct reclist_sortparms *sp)
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->name = nmem_strdup(se->nmem, sp->name);
637     se->sorted_results->increasing = sp->increasing;
638     se->sorted_results->type = sp->type;
639     se->sorted_results->next = 0;
640
641     session_log(se, YLOG_DEBUG, "clear_set session_sort: field=%s increasing=%d type=%d configured",
642             sp->name, sp->increasing, sp->type);
643
644     se->reclist = reclist_create(se->nmem);
645 }
646
647 static void session_sort_unlocked(struct session *se, struct reclist_sortparms *sp)
648 {
649     struct reclist_sortparms *sr;
650     struct client_list *l;
651     const char *field = sp->name;
652     int increasing = sp->increasing;
653     int type  = sp->type;
654     int clients_research = 0;
655
656     yaz_log(YLOG_LOG, "session_sort field=%s increasing=%d type=%d", field, increasing, type);
657     /* see if we already have sorted for this criteria */
658     for (sr = se->sorted_results; sr; sr = sr->next)
659     {
660         if (!reclist_sortparms_cmp(sr,sp))
661             break;
662     }
663     if (sr)
664     {
665         session_log(se, YLOG_DEBUG, "search_sort: field=%s increasing=%d type=%d already fetched",
666                     field, increasing, type);
667         return;
668     }
669     session_log(se, YLOG_DEBUG, "search_sort: field=%s increasing=%d type=%d must fetch",
670                     field, increasing, type);
671
672     // We need to reset reclist on every sort that changes the records, not just for position
673     // So if just one client requires new searching, we need to clear set.
674     // Ask each of the client if sorting requires re-search due to native sort
675     // If it does it will require us to
676     for (l = se->clients_active; l; l = l->next)
677     {
678         struct client *cl = l->client;
679         // Assume no re-search is required.
680         client_parse_init(cl, 1);
681         clients_research += client_parse_sort(cl, sp);
682     }
683     if (clients_research) {
684         yaz_log(YLOG_DEBUG, "Reset results due to %d clients researching", clients_research);
685         session_clear_set(se, sp);
686     }
687     else {
688         // A new sorting based on same record set
689         sr = nmem_malloc(se->nmem, sizeof(*sr));
690         sr->name = nmem_strdup(se->nmem, field);
691         sr->increasing = increasing;
692         sr->type = type;
693         sr->next = se->sorted_results;
694         se->sorted_results = sr;
695         session_log(se, YLOG_DEBUG, "No research/ingesting done");
696         return ;
697     }
698     session_log(se, YLOG_DEBUG, "Re- search/ingesting for clients due to change in sort order");
699
700     for (l = se->clients_active; l; l = l->next)
701     {
702         struct client *cl = l->client;
703         if (client_get_state(cl) == Client_Connecting ||
704             client_get_state(cl) == Client_Idle ||
705             client_get_state(cl) == Client_Working) {
706             client_start_search(cl);
707         }
708         else {
709             yaz_log(YLOG_DEBUG, "Client %s: No re-start/ingest in show. Wrong client state: %d",
710                         client_get_id(cl), client_get_state(cl));
711         }
712
713     }
714 }
715
716 void session_sort(struct session *se, struct reclist_sortparms *sp) {
717     //session_enter(se, "session_sort");
718     session_sort_unlocked(se, sp);
719     //session_leave(se, "session_sort");
720 }
721
722
723 enum pazpar2_error_code session_search(struct session *se,
724                                        const char *query,
725                                        const char *startrecs,
726                                        const char *maxrecs,
727                                        const char *filter,
728                                        const char *limit,
729                                        const char **addinfo,
730                                        struct reclist_sortparms *sp)
731 {
732     int live_channels = 0;
733     int no_working = 0;
734     int no_failed_query = 0;
735     int no_failed_limit = 0;
736     struct client_list *l, *l0;
737     facet_limits_t facet_limits;
738     int same_sort_order = 0;
739
740     session_log(se, YLOG_DEBUG, "Search");
741
742     *addinfo = 0;
743
744     if (se->settings_modified) {
745         session_remove_cached_clients(se);
746     }
747     else
748         session_reset_active_clients(se, 0);
749
750     session_enter(se, "session_search");
751     se->settings_modified = 0;
752
753     if (se->sorted_results) {
754         if (!reclist_sortparms_cmp(se->sorted_results, sp))
755             same_sort_order = 1;
756     }
757     session_clear_set(se, sp);
758     relevance_destroy(&se->relevance);
759
760     live_channels = select_targets(se, filter);
761     if (!live_channels)
762     {
763         session_leave(se, "session_search");
764         return PAZPAR2_NO_TARGETS;
765     }
766
767     facet_limits = facet_limits_create(limit);
768     if (!facet_limits)
769     {
770         *addinfo = "limit";
771         session_leave(se, "session_search");
772         return PAZPAR2_MALFORMED_PARAMETER_VALUE;
773     }
774
775     l0 = se->clients_active;
776     se->clients_active = 0;
777     session_leave(se, "session_search");
778
779     for (l = l0; l; l = l->next)
780     {
781         int parse_ret;
782         struct client *cl = l->client;
783         client_parse_init(cl, 1);
784         if (prepare_map(se, client_get_database(cl)) < 0)
785             continue;
786
787         parse_ret = client_parse_query(cl, query, facet_limits, se->service->ccl_bibset);
788         if (parse_ret == -1)
789             no_failed_query++;
790         else if (parse_ret == -2)
791             no_failed_limit++;
792         else if (parse_ret < 0)
793             no_working++; /* other error, such as bad CCL map */
794         else
795         {
796             client_parse_range(cl, startrecs, maxrecs);
797             client_parse_sort(cl, sp);
798             client_start_search(cl);
799             no_working++;
800         }
801     }
802     facet_limits_destroy(facet_limits);
803     session_reset_active_clients(se, l0);
804
805     if (no_working == 0)
806     {
807         if (no_failed_query > 0)
808         {
809             *addinfo = "query";
810             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
811         }
812         else if (no_failed_limit > 0)
813         {
814             *addinfo = "limit";
815             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
816         }
817         else
818             return PAZPAR2_NO_TARGETS;
819     }
820     session_log(se, YLOG_LOG, "session_start_search done");
821     return PAZPAR2_NO_ERROR;
822 }
823
824 // Creates a new session_database object for a database
825 static void session_init_databases_fun(void *context, struct database *db)
826 {
827     struct session *se = (struct session *) context;
828     struct session_database *new = nmem_malloc(se->session_nmem, sizeof(*new));
829     int i;
830
831     new->database = db;
832
833     new->map = 0;
834     assert(db->settings);
835     new->settings = nmem_malloc(se->session_nmem,
836                                 sizeof(struct settings *) * db->num_settings);
837     new->num_settings = db->num_settings;
838     for (i = 0; i < db->num_settings; i++)
839     {
840         struct setting *setting = db->settings[i];
841         new->settings[i] = setting;
842     }
843     new->next = se->databases;
844     se->databases = new;
845 }
846
847 // Doesn't free memory associated with sdb -- nmem takes care of that
848 static void session_database_destroy(struct session_database *sdb)
849 {
850     sdb->map = 0;
851 }
852
853 // Initialize session_database list -- this represents this session's view
854 // of the database list -- subject to modification by the settings ws command
855 void session_init_databases(struct session *se)
856 {
857     se->databases = 0;
858     predef_grep_databases(se, se->service, session_init_databases_fun);
859 }
860
861 // Probably session_init_databases_fun should be refactored instead of
862 // called here.
863 static struct session_database *load_session_database(struct session *se,
864                                                       char *id)
865 {
866     struct database *db = new_database_inherit_settings(id, se->session_nmem, se->service->settings);
867     session_init_databases_fun((void*) se, db);
868
869     // New sdb is head of se->databases list
870     return se->databases;
871 }
872
873 // Find an existing session database. If not found, load it
874 static struct session_database *find_session_database(struct session *se,
875                                                       char *id)
876 {
877     struct session_database *sdb;
878
879     for (sdb = se->databases; sdb; sdb = sdb->next)
880         if (!strcmp(sdb->database->id, id))
881             return sdb;
882     return load_session_database(se, id);
883 }
884
885 // Apply a session override to a database
886 void session_apply_setting(struct session *se, char *dbname, char *setting,
887                            char *value)
888 {
889     struct session_database *sdb = find_session_database(se, dbname);
890     struct conf_service *service = se->service;
891     struct setting *new = nmem_malloc(se->session_nmem, sizeof(*new));
892     int offset = settings_create_offset(service, setting);
893
894     expand_settings_array(&sdb->settings, &sdb->num_settings, offset,
895                           se->session_nmem);
896     new->precedence = 0;
897     new->target = dbname;
898     new->name = setting;
899     new->value = value;
900     new->next = sdb->settings[offset];
901     sdb->settings[offset] = new;
902
903     se->settings_modified = 1;
904
905     // Force later recompute of settings-driven data structures
906     // (happens when a search starts and client connections are prepared)
907     switch (offset)
908     {
909     case PZ_XSLT:
910         if (sdb->map)
911         {
912             sdb->map = 0;
913         }
914         break;
915     }
916 }
917
918 void session_destroy(struct session *se)
919 {
920     struct session_database *sdb;
921     session_log(se, YLOG_DEBUG, "Destroying");
922     session_use(-1);
923     session_remove_cached_clients(se);
924
925     for (sdb = se->databases; sdb; sdb = sdb->next)
926         session_database_destroy(sdb);
927     normalize_cache_destroy(se->normalize_cache);
928     relevance_destroy(&se->relevance);
929     reclist_destroy(se->reclist);
930     if (nmem_total(se->nmem))
931         session_log(se, YLOG_DEBUG, "NMEN operation usage %zd", nmem_total(se->nmem));
932     if (nmem_total(se->session_nmem))
933         session_log(se, YLOG_DEBUG, "NMEN session usage %zd", nmem_total(se->session_nmem));
934     nmem_destroy(se->nmem);
935     service_destroy(se->service);
936     yaz_mutex_destroy(&se->session_mutex);
937 }
938
939 size_t session_get_memory_status(struct session *session) {
940     size_t session_nmem;
941     if (session == 0)
942         return 0;
943     session_enter(session, "session_get_memory_status");
944     session_nmem = nmem_total(session->nmem);
945     session_leave(session, "session_get_memory_status");
946     return session_nmem;
947 }
948
949
950 struct session *new_session(NMEM nmem, struct conf_service *service,
951                             unsigned session_id)
952 {
953     int i;
954     struct session *session = nmem_malloc(nmem, sizeof(*session));
955
956     char tmp_str[50];
957
958     sprintf(tmp_str, "session#%u", session_id);
959
960     session->session_id = session_id;
961     session_log(session, YLOG_DEBUG, "New");
962     session->service = service;
963     session->relevance = 0;
964     session->total_records = 0;
965     session->number_of_warnings_unknown_elements = 0;
966     session->number_of_warnings_unknown_metadata = 0;
967     session->num_termlists = 0;
968     session->reclist = 0;
969     session->clients_active = 0;
970     session->clients_cached = 0;
971     session->settings_modified = 0;
972     session->session_nmem = nmem;
973     session->nmem = nmem_create();
974     session->databases = 0;
975     session->sorted_results = 0;
976
977     for (i = 0; i <= SESSION_WATCH_MAX; i++)
978     {
979         session->watchlist[i].data = 0;
980         session->watchlist[i].fun = 0;
981     }
982     session->normalize_cache = normalize_cache_create();
983     session->session_mutex = 0;
984     pazpar2_mutex_create(&session->session_mutex, tmp_str);
985     session_use(1);
986     return session;
987 }
988
989 const char * client_get_suggestions_xml(struct client *cl, WRBUF wrbuf);
990
991 static struct hitsbytarget *hitsbytarget_nb(struct session *se,
992                                             int *count, NMEM nmem)
993 {
994     struct hitsbytarget *res = 0;
995     struct client_list *l;
996     size_t sz = 0;
997
998     for (l = se->clients_active; l; l = l->next)
999         sz++;
1000
1001     res = nmem_malloc(nmem, sizeof(*res) * sz);
1002     *count = 0;
1003     for (l = se->clients_active; l; l = l->next)
1004     {
1005         struct client *cl = l->client;
1006         WRBUF w = wrbuf_alloc();
1007         const char *name = session_setting_oneval(client_get_database(cl),
1008                                                   PZ_NAME);
1009
1010         res[*count].id = client_get_id(cl);
1011         res[*count].name = *name ? name : "Unknown";
1012         res[*count].hits = client_get_hits(cl);
1013         res[*count].approximation = client_get_approximation(cl);
1014         res[*count].records = client_get_num_records(cl);
1015         res[*count].filtered = client_get_num_records_filtered(cl);
1016         res[*count].diagnostic =
1017             client_get_diagnostic(cl, &res[*count].addinfo);
1018         res[*count].state = client_get_state_str(cl);
1019         res[*count].connected  = client_get_connection(cl) ? 1 : 0;
1020         session_settings_dump(se, client_get_database(cl), w);
1021         res[*count].settings_xml = nmem_strdup(nmem, wrbuf_cstr(w));
1022         wrbuf_rewind(w);
1023         wrbuf_puts(w, "");
1024         res[*count].suggestions_xml = nmem_strdup(nmem, client_get_suggestions_xml(cl, w));
1025         wrbuf_destroy(w);
1026         (*count)++;
1027     }
1028     return res;
1029 }
1030
1031 struct hitsbytarget *get_hitsbytarget(struct session *se, int *count, NMEM nmem)
1032 {
1033     struct hitsbytarget *p;
1034     session_enter(se, "get_hitsbytarget");
1035     p = hitsbytarget_nb(se, count, nmem);
1036     session_leave(se, "get_hitsbytarget");
1037     return p;
1038 }
1039
1040 // Compares two hitsbytarget nodes by hitcount
1041 static int cmp_ht(const void *p1, const void *p2)
1042 {
1043     const struct hitsbytarget *h1 = p1;
1044     const struct hitsbytarget *h2 = p2;
1045     return h2->hits - h1->hits;
1046 }
1047
1048 // Compares two hitsbytarget nodes by hitcount
1049 static int cmp_ht_approx(const void *p1, const void *p2)
1050 {
1051     const struct hitsbytarget *h1 = p1;
1052     const struct hitsbytarget *h2 = p2;
1053     return h2->approximation - h1->approximation;
1054 }
1055
1056 static int targets_termlist_nb(WRBUF wrbuf, struct session *se, int num,
1057                                NMEM nmem, int version)
1058 {
1059     struct hitsbytarget *ht;
1060     int count, i;
1061
1062     ht = hitsbytarget_nb(se, &count, nmem);
1063     if (version >= 2)
1064         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht_approx);
1065     else
1066         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht);
1067     for (i = 0; i < count && i < num && ht[i].hits > 0; i++)
1068     {
1069
1070         // do only print terms which have display names
1071
1072         wrbuf_puts(wrbuf, "<term>\n");
1073
1074         wrbuf_puts(wrbuf, "<id>");
1075         wrbuf_xmlputs(wrbuf, ht[i].id);
1076         wrbuf_puts(wrbuf, "</id>\n");
1077
1078         wrbuf_puts(wrbuf, "<name>");
1079         if (!ht[i].name || !ht[i].name[0])
1080             wrbuf_xmlputs(wrbuf, "NO TARGET NAME");
1081         else
1082             wrbuf_xmlputs(wrbuf, ht[i].name);
1083         wrbuf_puts(wrbuf, "</name>\n");
1084
1085         wrbuf_printf(wrbuf, "<frequency>" ODR_INT_PRINTF "</frequency>\n",
1086                      ht[i].hits);
1087
1088         if (version >= 2) {
1089             // Should not print if we know it isn't a approximation.
1090             wrbuf_printf(wrbuf, "<approximation>" ODR_INT_PRINTF "</approximation>\n", ht[i].approximation);
1091             wrbuf_printf(wrbuf, "<records>%d</records>\n", ht[i].records - ht[i].filtered);
1092             wrbuf_printf(wrbuf, "<filtered>%d</filtered>\n", ht[i].filtered);
1093         }
1094
1095         wrbuf_puts(wrbuf, "<state>");
1096         wrbuf_xmlputs(wrbuf, ht[i].state);
1097         wrbuf_puts(wrbuf, "</state>\n");
1098
1099         wrbuf_printf(wrbuf, "<diagnostic>%d</diagnostic>\n",
1100                      ht[i].diagnostic);
1101         wrbuf_puts(wrbuf, "</term>\n");
1102     }
1103     return count;
1104 }
1105
1106 void perform_termlist(struct http_channel *c, struct session *se,
1107                       const char *name, int num, int version)
1108 {
1109     int i, j;
1110     NMEM nmem_tmp = nmem_create();
1111     char **names;
1112     int num_names = 0;
1113
1114     if (!name)
1115         name = "*";
1116
1117     nmem_strsplit(nmem_tmp, ",", name, &names, &num_names);
1118
1119     session_enter(se, "perform_termlist");
1120
1121     for (j = 0; j < num_names; j++)
1122     {
1123         const char *tname;
1124         int must_generate_empty = 1; /* bug 5350 */
1125
1126         for (i = 0; i < se->num_termlists; i++)
1127         {
1128             tname = se->termlists[i].name;
1129             if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1130             {
1131                 struct termlist_score **p = 0;
1132                 int len;
1133
1134                 wrbuf_puts(c->wrbuf, "<list name=\"");
1135                 wrbuf_xmlputs(c->wrbuf, tname);
1136                 wrbuf_puts(c->wrbuf, "\">\n");
1137                 must_generate_empty = 0;
1138
1139                 p = termlist_highscore(se->termlists[i].termlist, &len,
1140                                        nmem_tmp);
1141                 if (p)
1142                 {
1143                     int i;
1144                     for (i = 0; i < len && i < num; i++)
1145                     {
1146                         // prevent sending empty term elements
1147                         if (!p[i]->display_term || !p[i]->display_term[0])
1148                             continue;
1149
1150                         wrbuf_puts(c->wrbuf, "<term>");
1151                         wrbuf_puts(c->wrbuf, "<name>");
1152                         wrbuf_xmlputs(c->wrbuf, p[i]->display_term);
1153                         wrbuf_puts(c->wrbuf, "</name>");
1154
1155                         wrbuf_printf(c->wrbuf,
1156                                      "<frequency>%d</frequency>",
1157                                      p[i]->frequency);
1158                         wrbuf_puts(c->wrbuf, "</term>\n");
1159                     }
1160                 }
1161                 wrbuf_puts(c->wrbuf, "</list>\n");
1162             }
1163         }
1164         tname = "xtargets";
1165         if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1166         {
1167             wrbuf_puts(c->wrbuf, "<list name=\"");
1168             wrbuf_xmlputs(c->wrbuf, tname);
1169             wrbuf_puts(c->wrbuf, "\">\n");
1170
1171             targets_termlist_nb(c->wrbuf, se, num, c->nmem, version);
1172             wrbuf_puts(c->wrbuf, "</list>\n");
1173             must_generate_empty = 0;
1174         }
1175         if (must_generate_empty)
1176         {
1177             wrbuf_puts(c->wrbuf, "<list name=\"");
1178             wrbuf_xmlputs(c->wrbuf, names[j]);
1179             wrbuf_puts(c->wrbuf, "\"/>\n");
1180         }
1181     }
1182     session_leave(se, "perform_termlist");
1183     nmem_destroy(nmem_tmp);
1184 }
1185
1186 #ifdef MISSING_HEADERS
1187 void report_nmem_stats(void)
1188 {
1189     size_t in_use, is_free;
1190
1191     nmem_get_memory_in_use(&in_use);
1192     nmem_get_memory_free(&is_free);
1193
1194     yaz_log(YLOG_LOG, "nmem stat: use=%ld free=%ld",
1195             (long) in_use, (long) is_free);
1196 }
1197 #endif
1198
1199 struct record_cluster *show_single_start(struct session *se, const char *id,
1200                                          struct record_cluster **prev_r,
1201                                          struct record_cluster **next_r)
1202 {
1203     struct record_cluster *r = 0;
1204
1205     session_enter(se, "show_single_start");
1206     *prev_r = 0;
1207     *next_r = 0;
1208     if (se->reclist)
1209     {
1210         reclist_enter(se->reclist);
1211         while ((r = reclist_read_record(se->reclist)))
1212         {
1213             if (!strcmp(r->recid, id))
1214             {
1215                 *next_r = reclist_read_record(se->reclist);
1216                 break;
1217             }
1218             *prev_r = r;
1219         }
1220         reclist_leave(se->reclist);
1221     }
1222     if (!r)
1223         session_leave(se, "show_single_start");
1224     return r;
1225 }
1226
1227 void show_single_stop(struct session *se, struct record_cluster *rec)
1228 {
1229     session_leave(se, "show_single_stop");
1230 }
1231
1232 struct record_cluster **show_range_start(struct session *se,
1233                                          struct reclist_sortparms *sp,
1234                                          int start, int *num, int *total, Odr_int *sumhits, Odr_int *approx_hits)
1235 {
1236     struct record_cluster **recs;
1237     struct reclist_sortparms *spp;
1238     int i;
1239 #if USE_TIMING
1240     yaz_timing_t t = yaz_timing_create();
1241 #endif
1242     session_enter(se, "show_range_start");
1243     recs = nmem_malloc(se->nmem, *num * sizeof(struct record_cluster *));
1244     if (!se->relevance)
1245     {
1246         *num = 0;
1247         *total = 0;
1248         *sumhits = 0;
1249         *approx_hits = 0;
1250         recs = 0;
1251     }
1252     else
1253     {
1254         struct client_list *l;
1255
1256         for (spp = sp; spp; spp = spp->next)
1257             if (spp->type == Metadata_sortkey_relevance)
1258             {
1259                 relevance_prepare_read(se->relevance, se->reclist);
1260                 break;
1261             }
1262         reclist_sort(se->reclist, sp);
1263
1264         reclist_enter(se->reclist);
1265         *total = reclist_get_num_records(se->reclist);
1266
1267         *sumhits = 0;
1268         *approx_hits = 0;
1269         for (l = se->clients_active; l; l = l->next) {
1270             *sumhits += client_get_hits(l->client);
1271             *approx_hits += client_get_approximation(l->client);
1272         }
1273         for (i = 0; i < start; i++)
1274             if (!reclist_read_record(se->reclist))
1275             {
1276                 *num = 0;
1277                 recs = 0;
1278                 break;
1279             }
1280
1281         for (i = 0; i < *num; i++)
1282         {
1283             struct record_cluster *r = reclist_read_record(se->reclist);
1284             if (!r)
1285             {
1286                 *num = i;
1287                 break;
1288             }
1289             recs[i] = r;
1290         }
1291         reclist_leave(se->reclist);
1292     }
1293 #if USE_TIMING
1294     yaz_timing_stop(t);
1295     yaz_log(YLOG_LOG, "show %6.5f %3.2f %3.2f",
1296             yaz_timing_get_real(t), yaz_timing_get_user(t),
1297             yaz_timing_get_sys(t));
1298     yaz_timing_destroy(&t);
1299 #endif
1300     return recs;
1301 }
1302
1303 void show_range_stop(struct session *se, struct record_cluster **recs)
1304 {
1305     session_leave(se, "show_range_stop");
1306 }
1307
1308 void statistics(struct session *se, struct statistics *stat)
1309 {
1310     struct client_list *l;
1311     int count = 0;
1312
1313     memset(stat, 0, sizeof(*stat));
1314     stat->num_hits = 0;
1315     for (l = se->clients_active; l; l = l->next)
1316     {
1317         struct client *cl = l->client;
1318         if (!client_get_connection(cl))
1319             stat->num_no_connection++;
1320         stat->num_hits += client_get_hits(cl);
1321         switch (client_get_state(cl))
1322         {
1323         case Client_Connecting: stat->num_connecting++; break;
1324         case Client_Working: stat->num_working++; break;
1325         case Client_Idle: stat->num_idle++; break;
1326         case Client_Failed: stat->num_failed++; break;
1327         case Client_Error: stat->num_error++; break;
1328         default: break;
1329         }
1330         count++;
1331     }
1332     stat->num_records = se->total_records;
1333
1334     stat->num_clients = count;
1335 }
1336
1337 static struct record_metadata *record_metadata_init(
1338     NMEM nmem, const char *value, enum conf_metadata_type type,
1339     struct _xmlAttr *attr)
1340 {
1341     struct record_metadata *rec_md = record_metadata_create(nmem);
1342     struct record_metadata_attr **attrp = &rec_md->attributes;
1343
1344     for (; attr; attr = attr->next)
1345     {
1346         if (attr->children && attr->children->content)
1347         {
1348             if (strcmp((const char *) attr->name, "type"))
1349             {  /* skip the "type" attribute.. Its value is already part of
1350                   the element in output (md-%s) and so repeating it here
1351                   is redundant */
1352                 *attrp = nmem_malloc(nmem, sizeof(**attrp));
1353                 (*attrp)->name =
1354                     nmem_strdup(nmem, (const char *) attr->name);
1355                 (*attrp)->value =
1356                     nmem_strdup(nmem, (const char *) attr->children->content);
1357                 attrp = &(*attrp)->next;
1358             }
1359         }
1360     }
1361     *attrp = 0;
1362
1363     if (type == Metadata_type_generic)
1364     {
1365         char *p = nmem_strdup(nmem, value);
1366
1367         p = normalize7bit_generic(p, " ,/.:([");
1368
1369         rec_md->data.text.disp = p;
1370         rec_md->data.text.sort = 0;
1371     }
1372     else if (type == Metadata_type_year || type == Metadata_type_date)
1373     {
1374         int first, last;
1375         int longdate = 0;
1376
1377         if (type == Metadata_type_date)
1378             longdate = 1;
1379         if (extract7bit_dates((char *) value, &first, &last, longdate) < 0)
1380             return 0;
1381
1382         rec_md->data.number.min = first;
1383         rec_md->data.number.max = last;
1384     }
1385     else
1386         return 0;
1387     return rec_md;
1388 }
1389
1390 static int get_mergekey_from_doc(xmlDoc *doc, xmlNode *root, const char *name,
1391                                  struct conf_service *service, WRBUF norm_wr)
1392 {
1393     xmlNode *n;
1394     int no_found = 0;
1395     for (n = root->children; n; n = n->next)
1396     {
1397         if (n->type != XML_ELEMENT_NODE)
1398             continue;
1399         if (!strcmp((const char *) n->name, "metadata"))
1400         {
1401             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1402             if (type == NULL) {
1403                 yaz_log(YLOG_FATAL, "Missing type attribute on metadata element. Skipping!");
1404             }
1405             else if (!strcmp(name, (const char *) type))
1406             {
1407                 xmlChar *value = xmlNodeListGetString(doc, n->children, 1);
1408                 if (value)
1409                 {
1410                     const char *norm_str;
1411                     pp2_charset_token_t prt =
1412                         pp2_charset_token_create(service->charsets, "mergekey");
1413
1414                     pp2_charset_token_first(prt, (const char *) value, 0);
1415                     if (wrbuf_len(norm_wr) > 0)
1416                         wrbuf_puts(norm_wr, " ");
1417                     wrbuf_puts(norm_wr, name);
1418                     while ((norm_str =
1419                             pp2_charset_token_next(prt)))
1420                     {
1421                         if (*norm_str)
1422                         {
1423                             wrbuf_puts(norm_wr, " ");
1424                             wrbuf_puts(norm_wr, norm_str);
1425                         }
1426                     }
1427                     xmlFree(value);
1428                     pp2_charset_token_destroy(prt);
1429                     no_found++;
1430                 }
1431             }
1432             xmlFree(type);
1433         }
1434     }
1435     return no_found;
1436 }
1437
1438 static const char *get_mergekey(xmlDoc *doc, struct client *cl, int record_no,
1439                                 struct conf_service *service, NMEM nmem)
1440 {
1441     char *mergekey_norm = 0;
1442     xmlNode *root = xmlDocGetRootElement(doc);
1443     WRBUF norm_wr = wrbuf_alloc();
1444
1445     /* consider mergekey from XSL first */
1446     xmlChar *mergekey = xmlGetProp(root, (xmlChar *) "mergekey");
1447     if (mergekey)
1448     {
1449         const char *norm_str;
1450         pp2_charset_token_t prt =
1451             pp2_charset_token_create(service->charsets, "mergekey");
1452
1453         pp2_charset_token_first(prt, (const char *) mergekey, 0);
1454         while ((norm_str = pp2_charset_token_next(prt)))
1455         {
1456             if (*norm_str)
1457             {
1458                 if (wrbuf_len(norm_wr))
1459                     wrbuf_puts(norm_wr, " ");
1460                 wrbuf_puts(norm_wr, norm_str);
1461             }
1462         }
1463         pp2_charset_token_destroy(prt);
1464         xmlFree(mergekey);
1465     }
1466     else
1467     {
1468         /* no mergekey defined in XSL. Look for mergekey metadata instead */
1469         int field_id;
1470         for (field_id = 0; field_id < service->num_metadata; field_id++)
1471         {
1472             struct conf_metadata *ser_md = &service->metadata[field_id];
1473             if (ser_md->mergekey != Metadata_mergekey_no)
1474             {
1475                 int r = get_mergekey_from_doc(doc, root, ser_md->name,
1476                                               service, norm_wr);
1477                 if (r == 0 && ser_md->mergekey == Metadata_mergekey_required)
1478                 {
1479                     /* no mergekey on this one and it is required..
1480                        Generate unique key instead */
1481                     wrbuf_rewind(norm_wr);
1482                     break;
1483                 }
1484             }
1485         }
1486     }
1487
1488     /* generate unique key if none is not generated already or is empty */
1489     if (wrbuf_len(norm_wr) == 0)
1490     {
1491         wrbuf_printf(norm_wr, "position: %s-%d",
1492                      client_get_id(cl), record_no);
1493     }
1494     else
1495     {
1496         const char *lead = "content: ";
1497         wrbuf_insert(norm_wr, 0, lead, strlen(lead));
1498     }
1499     if (wrbuf_len(norm_wr) > 0)
1500         mergekey_norm = nmem_strdup(nmem, wrbuf_cstr(norm_wr));
1501     wrbuf_destroy(norm_wr);
1502     return mergekey_norm;
1503 }
1504
1505 /** \brief see if metadata for pz:recordfilter exists
1506     \param root xml root element of normalized record
1507     \param sdb session database for client
1508     \retval 0 if there is no metadata for pz:recordfilter
1509     \retval 1 if there is metadata for pz:recordfilter
1510
1511     If there is no pz:recordfilter defined, this function returns 1
1512     as well.
1513 */
1514
1515 static int check_record_filter(xmlNode *root, struct session_database *sdb)
1516 {
1517     int match = 0;
1518     xmlNode *n;
1519     const char *s;
1520     s = session_setting_oneval(sdb, PZ_RECORDFILTER);
1521
1522     if (!s || !*s)
1523         return 1;
1524
1525     for (n = root->children; n; n = n->next)
1526     {
1527         if (n->type != XML_ELEMENT_NODE)
1528             continue;
1529         if (!strcmp((const char *) n->name, "metadata"))
1530         {
1531             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1532             if (type)
1533             {
1534                 size_t len;
1535                 int substring;
1536                 const char *eq;
1537
1538                 if ((eq = strchr(s, '=')))
1539                     substring = 0;
1540                 else if ((eq = strchr(s, '~')))
1541                     substring = 1;
1542                 if (eq)
1543                     len = eq - s;
1544                 else
1545                     len = strlen(s);
1546                 if (len == strlen((const char *)type) &&
1547                     !memcmp((const char *) type, s, len))
1548                 {
1549                     xmlChar *value = xmlNodeGetContent(n);
1550                     if (value && *value)
1551                     {
1552                         if (!eq ||
1553                             (substring && strstr((const char *) value, eq+1)) ||
1554                             (!substring && !strcmp((const char *) value, eq + 1)))
1555                             match = 1;
1556                     }
1557                     xmlFree(value);
1558                 }
1559                 xmlFree(type);
1560             }
1561         }
1562     }
1563     return match;
1564 }
1565
1566
1567 static int ingest_to_cluster(struct client *cl,
1568                              xmlDoc *xdoc,
1569                              xmlNode *root,
1570                              int record_no,
1571                              const char *mergekey_norm);
1572
1573 /** \brief ingest XML record
1574     \param cl client holds the result set for record
1575     \param rec record buffer (0 terminated)
1576     \param record_no record position (1, 2, ..)
1577     \param nmem working NMEM
1578     \retval 0 OK
1579     \retval -1 failure
1580     \retval -2 Filtered
1581 */
1582 int ingest_record(struct client *cl, const char *rec,
1583                   int record_no, NMEM nmem)
1584 {
1585     struct session *se = client_get_session(cl);
1586     int ret = 0;
1587     struct session_database *sdb = client_get_database(cl);
1588     struct conf_service *service = se->service;
1589     xmlDoc *xdoc = normalize_record(se, sdb, service, rec, nmem);
1590     xmlNode *root;
1591     const char *mergekey_norm;
1592
1593     if (!xdoc)
1594         return -1;
1595
1596     root = xmlDocGetRootElement(xdoc);
1597
1598     if (!check_record_filter(root, sdb))
1599     {
1600         session_log(se, YLOG_LOG, "Filtered out record no %d from %s", record_no, sdb->database->id);
1601         xmlFreeDoc(xdoc);
1602         return -2;
1603     }
1604
1605     mergekey_norm = get_mergekey(xdoc, cl, record_no, service, nmem);
1606     if (!mergekey_norm)
1607     {
1608         session_log(se, YLOG_WARN, "Got no mergekey");
1609         xmlFreeDoc(xdoc);
1610         return -1;
1611     }
1612     session_enter(se, "ingest_record");
1613     if (client_get_session(cl) == se)
1614         ret = ingest_to_cluster(cl, xdoc, root, record_no, mergekey_norm);
1615     session_leave(se, "ingest_record");
1616
1617     xmlFreeDoc(xdoc);
1618     return ret;
1619 }
1620
1621 // Skip record on non-zero
1622 static int check_limit_local(struct client *cl,
1623                              struct record *record,
1624                              int record_no)
1625 {
1626     int skip_record = 0;
1627     struct session *se = client_get_session(cl);
1628     struct conf_service *service = se->service;
1629     NMEM nmem_tmp = nmem_create();
1630     struct session_database *sdb = client_get_database(cl);
1631     int l = 0;
1632     while (!skip_record)
1633     {
1634         struct conf_metadata *ser_md = 0;
1635         struct record_metadata *rec_md = 0;
1636         int md_field_id;
1637         char **values = 0;
1638         int i, num_v = 0;
1639
1640         const char *name = client_get_facet_limit_local(cl, sdb, &l, nmem_tmp, &num_v, &values);
1641         if (!name)
1642             break;
1643
1644         md_field_id = conf_service_metadata_field_id(service, name);
1645         if (md_field_id < 0)
1646         {
1647             skip_record = 1;
1648             break;
1649         }
1650         ser_md = &service->metadata[md_field_id];
1651         rec_md = record->metadata[md_field_id];
1652         yaz_log(YLOG_DEBUG, "check limit local %s", name);
1653         for (i = 0; i < num_v; )
1654         {
1655             if (rec_md)
1656             {
1657                 if (ser_md->type == Metadata_type_year
1658                     || ser_md->type == Metadata_type_date)
1659                 {
1660                     int y = atoi(values[i]);
1661                     if (y >= rec_md->data.number.min
1662                         && y <= rec_md->data.number.max)
1663                         break;
1664                 }
1665                 else
1666                 {
1667                     yaz_log(YLOG_DEBUG, "cmp: '%s' '%s'", rec_md->data.text.disp, values[i]);
1668                     if (!strcmp(rec_md->data.text.disp, values[i]))
1669                     {
1670                         // Value equals, should not be filtered.
1671                         break;
1672                     }
1673                 }
1674                 rec_md = rec_md->next;
1675             }
1676             else
1677             {
1678                 rec_md = record->metadata[md_field_id];
1679                 i++;
1680             }
1681         }
1682         // At end , not match
1683         if (i == num_v)
1684         {
1685             skip_record = 1;
1686             break;
1687         }
1688     }
1689     nmem_destroy(nmem_tmp);
1690     return skip_record;
1691 }
1692
1693 static int ingest_to_cluster(struct client *cl,
1694                              xmlDoc *xdoc,
1695                              xmlNode *root,
1696                              int record_no,
1697                              const char *mergekey_norm)
1698 {
1699     xmlNode *n;
1700     xmlChar *type = 0;
1701     xmlChar *value = 0;
1702     struct session *se = client_get_session(cl);
1703     struct conf_service *service = se->service;
1704     int term_factor = 1;
1705     struct record_cluster *cluster;
1706     struct session_database *sdb = client_get_database(cl);
1707     struct record *record = record_create(se->nmem,
1708                                           service->num_metadata,
1709                                           service->num_sortkeys, cl,
1710                                           record_no);
1711
1712     for (n = root->children; n; n = n->next)
1713     {
1714         if (type)
1715             xmlFree(type);
1716         if (value)
1717             xmlFree(value);
1718         type = value = 0;
1719
1720         if (n->type != XML_ELEMENT_NODE)
1721             continue;
1722         if (!strcmp((const char *) n->name, "metadata"))
1723         {
1724             struct conf_metadata *ser_md = 0;
1725             struct record_metadata **wheretoput = 0;
1726             struct record_metadata *rec_md = 0;
1727             int md_field_id = -1;
1728
1729             type = xmlGetProp(n, (xmlChar *) "type");
1730             value = xmlNodeListGetString(xdoc, n->children, 1);
1731
1732             if (!type || !value || !*value)
1733                 continue;
1734
1735             md_field_id
1736                 = conf_service_metadata_field_id(service, (const char *) type);
1737             if (md_field_id < 0)
1738             {
1739                 if (se->number_of_warnings_unknown_metadata == 0)
1740                 {
1741                     session_log(se, YLOG_WARN,
1742                             "Ignoring unknown metadata element: %s", type);
1743                 }
1744                 se->number_of_warnings_unknown_metadata++;
1745                 continue;
1746             }
1747
1748             ser_md = &service->metadata[md_field_id];
1749
1750             // non-merged metadata
1751             rec_md = record_metadata_init(se->nmem, (const char *) value,
1752                                           ser_md->type, n->properties);
1753             if (!rec_md)
1754             {
1755                 session_log(se, YLOG_WARN, "bad metadata data '%s' "
1756                             "for element '%s'", value, type);
1757                 continue;
1758             }
1759             wheretoput = &record->metadata[md_field_id];
1760             while (*wheretoput)
1761                 wheretoput = &(*wheretoput)->next;
1762             *wheretoput = rec_md;
1763         }
1764     }
1765
1766     if (check_limit_local(cl, record, record_no))
1767     {
1768         session_log(se, YLOG_LOG, "Facet filtered out record no %d from %s",
1769                     record_no, sdb->database->id);
1770         if (type)
1771             xmlFree(type);
1772         if (value)
1773             xmlFree(value);
1774         return -2;
1775     }
1776     cluster = reclist_insert(se->reclist, service, record,
1777                              mergekey_norm, &se->total_merged);
1778     if (!cluster)
1779         return -1;
1780
1781     {
1782         const char *use_term_factor_str =
1783             session_setting_oneval(sdb, PZ_TERMLIST_TERM_FACTOR);
1784         if (use_term_factor_str && use_term_factor_str[0] == '1')
1785         {
1786             int maxrecs = client_get_maxrecs(cl);
1787             int hits = (int) client_get_hits(cl);
1788             term_factor = MAX(hits, maxrecs) /  MAX(1, maxrecs);
1789             assert(term_factor >= 1);
1790             yaz_log(YLOG_DEBUG, "Using term factor: %d (%d / %d)", term_factor, MAX(hits, maxrecs), MAX(1, maxrecs));
1791         }
1792     }
1793
1794     if (global_parameters.dump_records)
1795         session_log(se, YLOG_LOG, "Cluster id %s from %s (#%d)", cluster->recid,
1796                     sdb->database->id, record_no);
1797
1798
1799     relevance_newrec(se->relevance, cluster);
1800
1801     // now parsing XML record and adding data to cluster or record metadata
1802     for (n = root->children; n; n = n->next)
1803     {
1804         pp2_charset_token_t prt;
1805         if (type)
1806             xmlFree(type);
1807         if (value)
1808             xmlFree(value);
1809         type = value = 0;
1810
1811         if (n->type != XML_ELEMENT_NODE)
1812             continue;
1813         if (!strcmp((const char *) n->name, "metadata"))
1814         {
1815             struct conf_metadata *ser_md = 0;
1816             struct conf_sortkey *ser_sk = 0;
1817             struct record_metadata **wheretoput = 0;
1818             struct record_metadata *rec_md = 0;
1819             int md_field_id = -1;
1820             int sk_field_id = -1;
1821             const char *rank;
1822             xmlChar *xml_rank;
1823
1824             type = xmlGetProp(n, (xmlChar *) "type");
1825             value = xmlNodeListGetString(xdoc, n->children, 1);
1826
1827             if (!type || !value || !*value)
1828                 continue;
1829
1830             md_field_id
1831                 = conf_service_metadata_field_id(service, (const char *) type);
1832             if (md_field_id < 0)
1833                 continue;
1834
1835             ser_md = &service->metadata[md_field_id];
1836
1837             if (ser_md->sortkey_offset >= 0)
1838             {
1839                 sk_field_id = ser_md->sortkey_offset;
1840                 ser_sk = &service->sortkeys[sk_field_id];
1841             }
1842
1843             // merged metadata
1844             rec_md = record_metadata_init(se->nmem, (const char *) value,
1845                                           ser_md->type, 0);
1846             if (!rec_md)
1847                 continue;
1848
1849             xml_rank = xmlGetProp(n, (xmlChar *) "rank");
1850             rank = xml_rank ? (const char *) xml_rank : ser_md->rank;
1851
1852             wheretoput = &cluster->metadata[md_field_id];
1853
1854             // and polulate with data:
1855             // assign cluster or record based on merge action
1856             if (ser_md->merge == Metadata_merge_unique)
1857             {
1858                 while (*wheretoput)
1859                 {
1860                     if (!strcmp((const char *) (*wheretoput)->data.text.disp,
1861                                 rec_md->data.text.disp))
1862                         break;
1863                     wheretoput = &(*wheretoput)->next;
1864                 }
1865                 if (!*wheretoput)
1866                     *wheretoput = rec_md;
1867             }
1868             else if (ser_md->merge == Metadata_merge_longest)
1869             {
1870                 if (!*wheretoput
1871                     || strlen(rec_md->data.text.disp)
1872                     > strlen((*wheretoput)->data.text.disp))
1873                 {
1874                     *wheretoput = rec_md;
1875                     if (ser_sk)
1876                     {
1877                         const char *sort_str = 0;
1878                         int skip_article =
1879                             ser_sk->type == Metadata_sortkey_skiparticle;
1880
1881                         if (!cluster->sortkeys[sk_field_id])
1882                             cluster->sortkeys[sk_field_id] =
1883                                 nmem_malloc(se->nmem,
1884                                             sizeof(union data_types));
1885
1886                         prt =
1887                             pp2_charset_token_create(service->charsets, "sort");
1888
1889                         pp2_charset_token_first(prt, rec_md->data.text.disp,
1890                                                 skip_article);
1891
1892                         pp2_charset_token_next(prt);
1893
1894                         sort_str = pp2_get_sort(prt);
1895
1896                         cluster->sortkeys[sk_field_id]->text.disp =
1897                             rec_md->data.text.disp;
1898                         if (!sort_str)
1899                         {
1900                             sort_str = rec_md->data.text.disp;
1901                             session_log(se, YLOG_WARN,
1902                                     "Could not make sortkey. Bug #1858");
1903                         }
1904                         cluster->sortkeys[sk_field_id]->text.sort =
1905                             nmem_strdup(se->nmem, sort_str);
1906                         pp2_charset_token_destroy(prt);
1907                     }
1908                 }
1909             }
1910             else if (ser_md->merge == Metadata_merge_all)
1911             {
1912                 while (*wheretoput)
1913                     wheretoput = &(*wheretoput)->next;
1914                 *wheretoput = rec_md;
1915             }
1916             else if (ser_md->merge == Metadata_merge_range)
1917             {
1918                 if (!*wheretoput)
1919                 {
1920                     *wheretoput = rec_md;
1921                     if (ser_sk)
1922                         cluster->sortkeys[sk_field_id]
1923                             = &rec_md->data;
1924                 }
1925                 else
1926                 {
1927                     int this_min = rec_md->data.number.min;
1928                     int this_max = rec_md->data.number.max;
1929                     if (this_min < (*wheretoput)->data.number.min)
1930                         (*wheretoput)->data.number.min = this_min;
1931                     if (this_max > (*wheretoput)->data.number.max)
1932                         (*wheretoput)->data.number.max = this_max;
1933                 }
1934             }
1935
1936             // ranking of _all_ fields enabled ...
1937             if (rank)
1938             {
1939                 relevance_countwords(se->relevance, cluster,
1940                                      (char *) value, rank, ser_md->name);
1941             }
1942
1943             // construct facets ... unless the client already has reported them
1944             if (ser_md->termlist && !client_has_facet(cl, (char *) type))
1945             {
1946                 if (ser_md->type == Metadata_type_year)
1947                 {
1948                     char year[64];
1949                     sprintf(year, "%d", rec_md->data.number.max);
1950
1951                     add_facet(se, (char *) type, year, term_factor);
1952                     if (rec_md->data.number.max != rec_md->data.number.min)
1953                     {
1954                         sprintf(year, "%d", rec_md->data.number.min);
1955                         add_facet(se, (char *) type, year, term_factor);
1956                     }
1957                 }
1958                 else
1959                     add_facet(se, (char *) type, (char *) value, term_factor);
1960             }
1961
1962             // cleaning up
1963             if (xml_rank)
1964                 xmlFree(xml_rank);
1965             xmlFree(type);
1966             xmlFree(value);
1967             type = value = 0;
1968         }
1969         else
1970         {
1971             if (se->number_of_warnings_unknown_elements == 0)
1972                 session_log(se, YLOG_WARN,
1973                         "Unexpected element in internal record: %s", n->name);
1974             se->number_of_warnings_unknown_elements++;
1975         }
1976     }
1977     if (type)
1978         xmlFree(type);
1979     if (value)
1980         xmlFree(value);
1981
1982     relevance_donerecord(se->relevance, cluster);
1983     se->total_records++;
1984
1985     return 0;
1986 }
1987
1988 void session_log(struct session *s, int level, const char *fmt, ...)
1989 {
1990     char buf[1024];
1991     va_list ap;
1992     va_start(ap, fmt);
1993
1994     yaz_vsnprintf(buf, sizeof(buf)-30, fmt, ap);
1995     yaz_log(level, "Session %u: %s", s->session_id, buf);
1996
1997     va_end(ap);
1998 }
1999
2000 /*
2001  * Local variables:
2002  * c-basic-offset: 4
2003  * c-file-style: "Stroustrup"
2004  * indent-tabs-mode: nil
2005  * End:
2006  * vim: shiftwidth=4 tabstop=8 expandtab
2007  */
2008