Revise extended fetch algorithm
[pazpar2-moved-to-github.git] / src / session.c
1 /* This file is part of Pazpar2.
2    Copyright (C) 2006-2013 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_DEBUG, "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
738     session_log(se, YLOG_DEBUG, "Search");
739
740     *addinfo = 0;
741
742     if (se->settings_modified) {
743         session_remove_cached_clients(se);
744     }
745     else
746         session_reset_active_clients(se, 0);
747
748     session_enter(se, "session_search");
749     se->settings_modified = 0;
750
751     session_clear_set(se, sp);
752     relevance_destroy(&se->relevance);
753
754     live_channels = select_targets(se, filter);
755     if (!live_channels)
756     {
757         session_leave(se, "session_search");
758         return PAZPAR2_NO_TARGETS;
759     }
760
761     facet_limits_destroy(se->facet_limits);
762     se->facet_limits = facet_limits_create(limit);
763     if (!se->facet_limits)
764     {
765         *addinfo = "limit";
766         session_leave(se, "session_search");
767         return PAZPAR2_MALFORMED_PARAMETER_VALUE;
768     }
769
770     l0 = se->clients_active;
771     se->clients_active = 0;
772     session_leave(se, "session_search");
773
774     for (l = l0; l; l = l->next)
775     {
776         int parse_ret;
777         struct client *cl = l->client;
778         client_parse_init(cl, 1);
779         if (prepare_map(se, client_get_database(cl)) < 0)
780             continue;
781
782         parse_ret = client_parse_query(cl, query, se->facet_limits);
783         if (parse_ret == -1)
784             no_failed_query++;
785         else if (parse_ret == -2)
786             no_failed_limit++;
787         else if (parse_ret < 0)
788             no_working++; /* other error, such as bad CCL map */
789         else
790         {
791             client_parse_range(cl, startrecs, maxrecs);
792             client_parse_sort(cl, sp);
793             client_start_search(cl);
794             no_working++;
795         }
796     }
797     session_reset_active_clients(se, l0);
798
799     if (no_working == 0)
800     {
801         if (no_failed_query > 0)
802         {
803             *addinfo = "query";
804             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
805         }
806         else if (no_failed_limit > 0)
807         {
808             *addinfo = "limit";
809             return PAZPAR2_MALFORMED_PARAMETER_VALUE;
810         }
811         else
812             return PAZPAR2_NO_TARGETS;
813     }
814     session_log(se, YLOG_LOG, "session_start_search done");
815     return PAZPAR2_NO_ERROR;
816 }
817
818 // Creates a new session_database object for a database
819 static void session_init_databases_fun(void *context, struct database *db)
820 {
821     struct session *se = (struct session *) context;
822     struct session_database *new = nmem_malloc(se->session_nmem, sizeof(*new));
823     int i;
824
825     new->database = db;
826
827     new->map = 0;
828     assert(db->settings);
829     new->settings = nmem_malloc(se->session_nmem,
830                                 sizeof(struct settings *) * db->num_settings);
831     new->num_settings = db->num_settings;
832     for (i = 0; i < db->num_settings; i++)
833     {
834         struct setting *setting = db->settings[i];
835         new->settings[i] = setting;
836     }
837     new->next = se->databases;
838     se->databases = new;
839 }
840
841 // Doesn't free memory associated with sdb -- nmem takes care of that
842 static void session_database_destroy(struct session_database *sdb)
843 {
844     sdb->map = 0;
845 }
846
847 // Initialize session_database list -- this represents this session's view
848 // of the database list -- subject to modification by the settings ws command
849 void session_init_databases(struct session *se)
850 {
851     se->databases = 0;
852     predef_grep_databases(se, se->service, session_init_databases_fun);
853 }
854
855 // Probably session_init_databases_fun should be refactored instead of
856 // called here.
857 static struct session_database *load_session_database(struct session *se,
858                                                       char *id)
859 {
860     struct database *db = new_database_inherit_settings(id, se->session_nmem, se->service->settings);
861     session_init_databases_fun((void*) se, db);
862
863     // New sdb is head of se->databases list
864     return se->databases;
865 }
866
867 // Find an existing session database. If not found, load it
868 static struct session_database *find_session_database(struct session *se,
869                                                       char *id)
870 {
871     struct session_database *sdb;
872
873     for (sdb = se->databases; sdb; sdb = sdb->next)
874         if (!strcmp(sdb->database->id, id))
875             return sdb;
876     return load_session_database(se, id);
877 }
878
879 // Apply a session override to a database
880 void session_apply_setting(struct session *se, char *dbname, char *setting,
881                            char *value)
882 {
883     struct session_database *sdb = find_session_database(se, dbname);
884     struct conf_service *service = se->service;
885     struct setting *new = nmem_malloc(se->session_nmem, sizeof(*new));
886     int offset = settings_create_offset(service, setting);
887
888     expand_settings_array(&sdb->settings, &sdb->num_settings, offset,
889                           se->session_nmem);
890     new->precedence = 0;
891     new->target = dbname;
892     new->name = setting;
893     new->value = value;
894     new->next = sdb->settings[offset];
895     sdb->settings[offset] = new;
896
897     se->settings_modified = 1;
898
899     // Force later recompute of settings-driven data structures
900     // (happens when a search starts and client connections are prepared)
901     switch (offset)
902     {
903     case PZ_XSLT:
904         if (sdb->map)
905         {
906             sdb->map = 0;
907         }
908         break;
909     }
910 }
911
912 void session_destroy(struct session *se)
913 {
914     struct session_database *sdb;
915     session_log(se, YLOG_DEBUG, "Destroying");
916     session_use(-1);
917     session_remove_cached_clients(se);
918
919     for (sdb = se->databases; sdb; sdb = sdb->next)
920         session_database_destroy(sdb);
921     normalize_cache_destroy(se->normalize_cache);
922     relevance_destroy(&se->relevance);
923     reclist_destroy(se->reclist);
924     if (nmem_total(se->nmem))
925         session_log(se, YLOG_DEBUG, "NMEN operation usage %zd", nmem_total(se->nmem));
926     if (nmem_total(se->session_nmem))
927         session_log(se, YLOG_DEBUG, "NMEN session usage %zd", nmem_total(se->session_nmem));
928     facet_limits_destroy(se->facet_limits);
929     nmem_destroy(se->nmem);
930     service_destroy(se->service);
931     yaz_mutex_destroy(&se->session_mutex);
932 }
933
934 size_t session_get_memory_status(struct session *session) {
935     size_t session_nmem;
936     if (session == 0)
937         return 0;
938     session_enter(session, "session_get_memory_status");
939     session_nmem = nmem_total(session->nmem);
940     session_leave(session, "session_get_memory_status");
941     return session_nmem;
942 }
943
944
945 struct session *new_session(NMEM nmem, struct conf_service *service,
946                             unsigned session_id)
947 {
948     int i;
949     struct session *session = nmem_malloc(nmem, sizeof(*session));
950
951     char tmp_str[50];
952
953     sprintf(tmp_str, "session#%u", session_id);
954
955     session->session_id = session_id;
956     session_log(session, YLOG_DEBUG, "New");
957     session->service = service;
958     session->relevance = 0;
959     session->total_records = 0;
960     session->number_of_warnings_unknown_elements = 0;
961     session->number_of_warnings_unknown_metadata = 0;
962     session->num_termlists = 0;
963     session->reclist = 0;
964     session->clients_active = 0;
965     session->clients_cached = 0;
966     session->settings_modified = 0;
967     session->session_nmem = nmem;
968     session->nmem = nmem_create();
969     session->databases = 0;
970     session->sorted_results = 0;
971     session->facet_limits = 0;
972
973     for (i = 0; i <= SESSION_WATCH_MAX; i++)
974     {
975         session->watchlist[i].data = 0;
976         session->watchlist[i].fun = 0;
977     }
978     session->normalize_cache = normalize_cache_create();
979     session->session_mutex = 0;
980     pazpar2_mutex_create(&session->session_mutex, tmp_str);
981     session_use(1);
982     return session;
983 }
984
985 const char * client_get_suggestions_xml(struct client *cl, WRBUF wrbuf);
986
987 static struct hitsbytarget *hitsbytarget_nb(struct session *se,
988                                             int *count, NMEM nmem)
989 {
990     struct hitsbytarget *res = 0;
991     struct client_list *l;
992     size_t sz = 0;
993
994     for (l = se->clients_active; l; l = l->next)
995         sz++;
996
997     res = nmem_malloc(nmem, sizeof(*res) * sz);
998     *count = 0;
999     for (l = se->clients_active; l; l = l->next)
1000     {
1001         struct client *cl = l->client;
1002         WRBUF w = wrbuf_alloc();
1003         const char *name = session_setting_oneval(client_get_database(cl),
1004                                                   PZ_NAME);
1005
1006         res[*count].id = client_get_id(cl);
1007         res[*count].name = *name ? name : "Unknown";
1008         res[*count].hits = client_get_hits(cl);
1009         res[*count].approximation = client_get_approximation(cl);
1010         res[*count].records = client_get_num_records(cl);
1011         res[*count].filtered = client_get_num_records_filtered(cl);
1012         res[*count].diagnostic =
1013             client_get_diagnostic(cl, &res[*count].message,
1014                                   &res[*count].addinfo);
1015         res[*count].state = client_get_state_str(cl);
1016         res[*count].connected  = client_get_connection(cl) ? 1 : 0;
1017         session_settings_dump(se, client_get_database(cl), w);
1018         res[*count].settings_xml = nmem_strdup(nmem, wrbuf_cstr(w));
1019         wrbuf_rewind(w);
1020         wrbuf_puts(w, "");
1021         res[*count].suggestions_xml = nmem_strdup(nmem, client_get_suggestions_xml(cl, w));
1022         wrbuf_destroy(w);
1023         (*count)++;
1024     }
1025     return res;
1026 }
1027
1028 struct hitsbytarget *get_hitsbytarget(struct session *se, int *count, NMEM nmem)
1029 {
1030     struct hitsbytarget *p;
1031     session_enter(se, "get_hitsbytarget");
1032     p = hitsbytarget_nb(se, count, nmem);
1033     session_leave(se, "get_hitsbytarget");
1034     return p;
1035 }
1036
1037 // Compares two hitsbytarget nodes by hitcount
1038 static int cmp_ht(const void *p1, const void *p2)
1039 {
1040     const struct hitsbytarget *h1 = p1;
1041     const struct hitsbytarget *h2 = p2;
1042     return h2->hits - h1->hits;
1043 }
1044
1045 // Compares two hitsbytarget nodes by hitcount
1046 static int cmp_ht_approx(const void *p1, const void *p2)
1047 {
1048     const struct hitsbytarget *h1 = p1;
1049     const struct hitsbytarget *h2 = p2;
1050     return h2->approximation - h1->approximation;
1051 }
1052
1053 static int targets_termlist_nb(WRBUF wrbuf, struct session *se, int num,
1054                                NMEM nmem, int version)
1055 {
1056     struct hitsbytarget *ht;
1057     int count, i;
1058
1059     ht = hitsbytarget_nb(se, &count, nmem);
1060     if (version >= 2)
1061         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht_approx);
1062     else
1063         qsort(ht, count, sizeof(struct hitsbytarget), cmp_ht);
1064     for (i = 0; i < count && i < num && ht[i].hits > 0; i++)
1065     {
1066
1067         // do only print terms which have display names
1068
1069         wrbuf_puts(wrbuf, "<term>\n");
1070
1071         wrbuf_puts(wrbuf, "<id>");
1072         wrbuf_xmlputs(wrbuf, ht[i].id);
1073         wrbuf_puts(wrbuf, "</id>\n");
1074
1075         wrbuf_puts(wrbuf, "<name>");
1076         if (!ht[i].name || !ht[i].name[0])
1077             wrbuf_xmlputs(wrbuf, "NO TARGET NAME");
1078         else
1079             wrbuf_xmlputs(wrbuf, ht[i].name);
1080         wrbuf_puts(wrbuf, "</name>\n");
1081
1082         wrbuf_printf(wrbuf, "<frequency>" ODR_INT_PRINTF "</frequency>\n",
1083                      ht[i].hits);
1084
1085         if (version >= 2) {
1086             // Should not print if we know it isn't a approximation.
1087             wrbuf_printf(wrbuf, "<approximation>" ODR_INT_PRINTF "</approximation>\n", ht[i].approximation);
1088             wrbuf_printf(wrbuf, "<records>%d</records>\n", ht[i].records - ht[i].filtered);
1089             wrbuf_printf(wrbuf, "<filtered>%d</filtered>\n", ht[i].filtered);
1090         }
1091
1092         wrbuf_puts(wrbuf, "<state>");
1093         wrbuf_xmlputs(wrbuf, ht[i].state);
1094         wrbuf_puts(wrbuf, "</state>\n");
1095
1096         wrbuf_printf(wrbuf, "<diagnostic>%d</diagnostic>\n",
1097                      ht[i].diagnostic);
1098         wrbuf_puts(wrbuf, "</term>\n");
1099     }
1100     return count;
1101 }
1102
1103 void perform_termlist(struct http_channel *c, struct session *se,
1104                       const char *name, int num, int version)
1105 {
1106     int i, j;
1107     NMEM nmem_tmp = nmem_create();
1108     char **names;
1109     int num_names = 0;
1110
1111     if (!name)
1112         name = "*";
1113
1114     nmem_strsplit(nmem_tmp, ",", name, &names, &num_names);
1115
1116     session_enter(se, "perform_termlist");
1117
1118     for (j = 0; j < num_names; j++)
1119     {
1120         const char *tname;
1121         int must_generate_empty = 1; /* bug 5350 */
1122
1123         for (i = 0; i < se->num_termlists; i++)
1124         {
1125             tname = se->termlists[i].name;
1126             if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1127             {
1128                 struct termlist_score **p = 0;
1129                 int len;
1130
1131                 wrbuf_puts(c->wrbuf, "<list name=\"");
1132                 wrbuf_xmlputs(c->wrbuf, tname);
1133                 wrbuf_puts(c->wrbuf, "\">\n");
1134                 must_generate_empty = 0;
1135
1136                 p = termlist_highscore(se->termlists[i].termlist, &len,
1137                                        nmem_tmp);
1138                 if (p)
1139                 {
1140                     int i;
1141                     for (i = 0; i < len && i < num; i++)
1142                     {
1143                         // prevent sending empty term elements
1144                         if (!p[i]->display_term || !p[i]->display_term[0])
1145                             continue;
1146
1147                         wrbuf_puts(c->wrbuf, "<term>");
1148                         wrbuf_puts(c->wrbuf, "<name>");
1149                         wrbuf_xmlputs(c->wrbuf, p[i]->display_term);
1150                         wrbuf_puts(c->wrbuf, "</name>");
1151
1152                         wrbuf_printf(c->wrbuf,
1153                                      "<frequency>%d</frequency>",
1154                                      p[i]->frequency);
1155                         wrbuf_puts(c->wrbuf, "</term>\n");
1156                     }
1157                 }
1158                 wrbuf_puts(c->wrbuf, "</list>\n");
1159             }
1160         }
1161         tname = "xtargets";
1162         if (!strcmp(names[j], tname) || !strcmp(names[j], "*"))
1163         {
1164             wrbuf_puts(c->wrbuf, "<list name=\"");
1165             wrbuf_xmlputs(c->wrbuf, tname);
1166             wrbuf_puts(c->wrbuf, "\">\n");
1167
1168             targets_termlist_nb(c->wrbuf, se, num, c->nmem, version);
1169             wrbuf_puts(c->wrbuf, "</list>\n");
1170             must_generate_empty = 0;
1171         }
1172         if (must_generate_empty)
1173         {
1174             wrbuf_puts(c->wrbuf, "<list name=\"");
1175             wrbuf_xmlputs(c->wrbuf, names[j]);
1176             wrbuf_puts(c->wrbuf, "\"/>\n");
1177         }
1178     }
1179     session_leave(se, "perform_termlist");
1180     nmem_destroy(nmem_tmp);
1181 }
1182
1183 #ifdef MISSING_HEADERS
1184 void report_nmem_stats(void)
1185 {
1186     size_t in_use, is_free;
1187
1188     nmem_get_memory_in_use(&in_use);
1189     nmem_get_memory_free(&is_free);
1190
1191     yaz_log(YLOG_LOG, "nmem stat: use=%ld free=%ld",
1192             (long) in_use, (long) is_free);
1193 }
1194 #endif
1195
1196 struct record_cluster *show_single_start(struct session *se, const char *id,
1197                                          struct record_cluster **prev_r,
1198                                          struct record_cluster **next_r)
1199 {
1200     struct record_cluster *r = 0;
1201
1202     session_enter(se, "show_single_start");
1203     *prev_r = 0;
1204     *next_r = 0;
1205     if (se->reclist)
1206     {
1207         reclist_limit(se->reclist, se);
1208
1209         reclist_enter(se->reclist);
1210         while ((r = reclist_read_record(se->reclist)))
1211         {
1212             if (!strcmp(r->recid, id))
1213             {
1214                 *next_r = reclist_read_record(se->reclist);
1215                 break;
1216             }
1217             *prev_r = r;
1218         }
1219         reclist_leave(se->reclist);
1220     }
1221     if (!r)
1222         session_leave(se, "show_single_start");
1223     return r;
1224 }
1225
1226 void show_single_stop(struct session *se, struct record_cluster *rec)
1227 {
1228     session_leave(se, "show_single_stop");
1229 }
1230
1231
1232 int session_fetch_more(struct session *se)
1233 {
1234     struct client_list *l;
1235     int ret = 0;
1236
1237     for (l = se->clients_active; l; l = l->next)
1238     {
1239         struct client *cl = l->client;
1240         if (client_get_state(cl) == Client_Idle)
1241         {
1242             if (client_fetch_more(cl))
1243             {
1244                 session_log(se, YLOG_LOG, "%s: more to fetch",
1245                             client_get_id(cl));
1246                 ret = 1;
1247             }
1248             else
1249             {
1250                 session_log(se, YLOG_LOG, "%s: no more to fetch",
1251                             client_get_id(cl));
1252             }
1253         }
1254         else
1255         {
1256             session_log(se, YLOG_LOG, "%s: no fetch due to state=%s",
1257                         client_get_id(cl), client_get_state_str(cl));
1258         }
1259
1260     }
1261     return ret;
1262 }
1263
1264 struct record_cluster **show_range_start(struct session *se,
1265                                          struct reclist_sortparms *sp,
1266                                          int start, int *num, int *total,
1267                                          Odr_int *sumhits, Odr_int *approx_hits,
1268                                          void (*show_records_ready)(void *data),
1269                                          struct http_channel *chan)
1270 {
1271     struct record_cluster **recs = 0;
1272     struct reclist_sortparms *spp;
1273     struct client_list *l;
1274     int i;
1275 #if USE_TIMING
1276     yaz_timing_t t = yaz_timing_create();
1277 #endif
1278     session_enter(se, "show_range_start");
1279     *sumhits = 0;
1280     *approx_hits = 0;
1281     *total = 0;
1282     reclist_limit(se->reclist, se);
1283     if (se->relevance)
1284     {
1285         for (spp = sp; spp; spp = spp->next)
1286             if (spp->type == Metadata_sortkey_relevance)
1287             {
1288                 relevance_prepare_read(se->relevance, se->reclist);
1289                 break;
1290             }
1291         for (l = se->clients_active; l; l = l->next) {
1292             *sumhits += client_get_hits(l->client);
1293             *approx_hits += client_get_approximation(l->client);
1294         }
1295     }
1296     reclist_sort(se->reclist, sp);
1297
1298     reclist_enter(se->reclist);
1299     *total = reclist_get_num_records(se->reclist);
1300
1301     for (l = se->clients_active; l; l = l->next)
1302         client_update_show_stat(l->client, 0);
1303
1304     for (i = 0; i < start; i++)
1305     {
1306         struct record_cluster *r = reclist_read_record(se->reclist);
1307         if (!r)
1308         {
1309             *num = 0;
1310             break;
1311         }
1312         else
1313         {
1314             struct record *rec = r->records;
1315             for (;rec; rec = rec->next)
1316                 client_update_show_stat(rec->client, 1);
1317         }
1318     }
1319     if (*num > 0)
1320         recs =
1321             nmem_malloc(se->nmem, *num * sizeof(struct record_cluster *));
1322     for (i = 0; i < *num; i++)
1323     {
1324         struct record_cluster *r = reclist_read_record(se->reclist);
1325         if (!r)
1326         {
1327             *num = i;
1328             break;
1329         }
1330         else
1331         {
1332             struct record *rec = r->records;
1333             for (;rec; rec = rec->next)
1334                 client_update_show_stat(rec->client, 1);
1335             recs[i] = r;
1336         }
1337     }
1338     reclist_leave(se->reclist);
1339 #if USE_TIMING
1340     yaz_timing_stop(t);
1341     yaz_log(YLOG_LOG, "show %6.5f %3.2f %3.2f",
1342             yaz_timing_get_real(t), yaz_timing_get_user(t),
1343             yaz_timing_get_sys(t));
1344     yaz_timing_destroy(&t);
1345 #endif
1346
1347     if (!session_fetch_more(se))
1348         session_log(se, YLOG_LOG, "can not fetch more");
1349     else
1350     {
1351         show_range_stop(se, recs);
1352         session_log(se, YLOG_LOG, "fetching more in progress");
1353         if (session_set_watch(se, SESSION_WATCH_SHOW,
1354                               show_records_ready, chan, chan))
1355         {
1356             session_log(se, YLOG_WARN, "Ignoring show block");
1357             session_enter(se, "show_range_start");
1358         }
1359         else
1360         {
1361             session_log(se, YLOG_LOG, "session watch OK");
1362             return 0;
1363         }
1364     }
1365     return recs;
1366 }
1367
1368 void show_range_stop(struct session *se, struct record_cluster **recs)
1369 {
1370     session_leave(se, "show_range_stop");
1371 }
1372
1373 void statistics(struct session *se, struct statistics *stat)
1374 {
1375     struct client_list *l;
1376     int count = 0;
1377
1378     memset(stat, 0, sizeof(*stat));
1379     stat->num_hits = 0;
1380     for (l = se->clients_active; l; l = l->next)
1381     {
1382         struct client *cl = l->client;
1383         if (!client_get_connection(cl))
1384             stat->num_no_connection++;
1385         stat->num_hits += client_get_hits(cl);
1386         switch (client_get_state(cl))
1387         {
1388         case Client_Connecting: stat->num_connecting++; break;
1389         case Client_Working: stat->num_working++; break;
1390         case Client_Idle: stat->num_idle++; break;
1391         case Client_Failed: stat->num_failed++; break;
1392         case Client_Error: stat->num_error++; break;
1393         default: break;
1394         }
1395         count++;
1396     }
1397     stat->num_records = se->total_records;
1398
1399     stat->num_clients = count;
1400 }
1401
1402 static struct record_metadata *record_metadata_init(
1403     NMEM nmem, const char *value, enum conf_metadata_type type,
1404     struct _xmlAttr *attr)
1405 {
1406     struct record_metadata *rec_md = record_metadata_create(nmem);
1407     struct record_metadata_attr **attrp = &rec_md->attributes;
1408
1409     for (; attr; attr = attr->next)
1410     {
1411         if (attr->children && attr->children->content)
1412         {
1413             if (strcmp((const char *) attr->name, "type"))
1414             {  /* skip the "type" attribute.. Its value is already part of
1415                   the element in output (md-%s) and so repeating it here
1416                   is redundant */
1417                 *attrp = nmem_malloc(nmem, sizeof(**attrp));
1418                 (*attrp)->name =
1419                     nmem_strdup(nmem, (const char *) attr->name);
1420                 (*attrp)->value =
1421                     nmem_strdup(nmem, (const char *) attr->children->content);
1422                 attrp = &(*attrp)->next;
1423             }
1424         }
1425     }
1426     *attrp = 0;
1427
1428     if (type == Metadata_type_generic)
1429     {
1430         char *p = nmem_strdup(nmem, value);
1431
1432         p = normalize7bit_generic(p, " ,/.:([");
1433
1434         rec_md->data.text.disp = p;
1435         rec_md->data.text.sort = 0;
1436     }
1437     else if (type == Metadata_type_year || type == Metadata_type_date)
1438     {
1439         int first, last;
1440         int longdate = 0;
1441
1442         if (type == Metadata_type_date)
1443             longdate = 1;
1444         if (extract7bit_dates((char *) value, &first, &last, longdate) < 0)
1445             return 0;
1446
1447         rec_md->data.number.min = first;
1448         rec_md->data.number.max = last;
1449     }
1450     else
1451         return 0;
1452     return rec_md;
1453 }
1454
1455 static int get_mergekey_from_doc(xmlDoc *doc, xmlNode *root, const char *name,
1456                                  struct conf_service *service, WRBUF norm_wr)
1457 {
1458     xmlNode *n;
1459     int no_found = 0;
1460     for (n = root->children; n; n = n->next)
1461     {
1462         if (n->type != XML_ELEMENT_NODE)
1463             continue;
1464         if (!strcmp((const char *) n->name, "metadata"))
1465         {
1466             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1467             if (type == NULL) {
1468                 yaz_log(YLOG_FATAL, "Missing type attribute on metadata element. Skipping!");
1469             }
1470             else if (!strcmp(name, (const char *) type))
1471             {
1472                 xmlChar *value = xmlNodeListGetString(doc, n->children, 1);
1473                 if (value)
1474                 {
1475                     const char *norm_str;
1476                     pp2_charset_token_t prt =
1477                         pp2_charset_token_create(service->charsets, "mergekey");
1478
1479                     pp2_charset_token_first(prt, (const char *) value, 0);
1480                     if (wrbuf_len(norm_wr) > 0)
1481                         wrbuf_puts(norm_wr, " ");
1482                     wrbuf_puts(norm_wr, name);
1483                     while ((norm_str =
1484                             pp2_charset_token_next(prt)))
1485                     {
1486                         if (*norm_str)
1487                         {
1488                             wrbuf_puts(norm_wr, " ");
1489                             wrbuf_puts(norm_wr, norm_str);
1490                         }
1491                     }
1492                     xmlFree(value);
1493                     pp2_charset_token_destroy(prt);
1494                     no_found++;
1495                 }
1496             }
1497             xmlFree(type);
1498         }
1499     }
1500     return no_found;
1501 }
1502
1503 static const char *get_mergekey(xmlDoc *doc, struct client *cl, int record_no,
1504                                 struct conf_service *service, NMEM nmem)
1505 {
1506     char *mergekey_norm = 0;
1507     xmlNode *root = xmlDocGetRootElement(doc);
1508     WRBUF norm_wr = wrbuf_alloc();
1509
1510     /* consider mergekey from XSL first */
1511     xmlChar *mergekey = xmlGetProp(root, (xmlChar *) "mergekey");
1512     if (mergekey)
1513     {
1514         const char *norm_str;
1515         pp2_charset_token_t prt =
1516             pp2_charset_token_create(service->charsets, "mergekey");
1517
1518         pp2_charset_token_first(prt, (const char *) mergekey, 0);
1519         while ((norm_str = pp2_charset_token_next(prt)))
1520         {
1521             if (*norm_str)
1522             {
1523                 if (wrbuf_len(norm_wr))
1524                     wrbuf_puts(norm_wr, " ");
1525                 wrbuf_puts(norm_wr, norm_str);
1526             }
1527         }
1528         pp2_charset_token_destroy(prt);
1529         xmlFree(mergekey);
1530     }
1531     else
1532     {
1533         /* no mergekey defined in XSL. Look for mergekey metadata instead */
1534         int field_id;
1535         for (field_id = 0; field_id < service->num_metadata; field_id++)
1536         {
1537             struct conf_metadata *ser_md = &service->metadata[field_id];
1538             if (ser_md->mergekey != Metadata_mergekey_no)
1539             {
1540                 int r = get_mergekey_from_doc(doc, root, ser_md->name,
1541                                               service, norm_wr);
1542                 if (r == 0 && ser_md->mergekey == Metadata_mergekey_required)
1543                 {
1544                     /* no mergekey on this one and it is required..
1545                        Generate unique key instead */
1546                     wrbuf_rewind(norm_wr);
1547                     break;
1548                 }
1549             }
1550         }
1551     }
1552
1553     /* generate unique key if none is not generated already or is empty */
1554     if (wrbuf_len(norm_wr) == 0)
1555     {
1556         wrbuf_printf(norm_wr, "position: %s-%d",
1557                      client_get_id(cl), record_no);
1558     }
1559     else
1560     {
1561         const char *lead = "content: ";
1562         wrbuf_insert(norm_wr, 0, lead, strlen(lead));
1563     }
1564     if (wrbuf_len(norm_wr) > 0)
1565         mergekey_norm = nmem_strdup(nmem, wrbuf_cstr(norm_wr));
1566     wrbuf_destroy(norm_wr);
1567     return mergekey_norm;
1568 }
1569
1570 /** \brief see if metadata for pz:recordfilter exists
1571     \param root xml root element of normalized record
1572     \param sdb session database for client
1573     \retval 0 if there is no metadata for pz:recordfilter
1574     \retval 1 if there is metadata for pz:recordfilter
1575
1576     If there is no pz:recordfilter defined, this function returns 1
1577     as well.
1578 */
1579
1580 static int check_record_filter(xmlNode *root, struct session_database *sdb)
1581 {
1582     int match = 0;
1583     xmlNode *n;
1584     const char *s;
1585     s = session_setting_oneval(sdb, PZ_RECORDFILTER);
1586
1587     if (!s || !*s)
1588         return 1;
1589
1590     for (n = root->children; n; n = n->next)
1591     {
1592         if (n->type != XML_ELEMENT_NODE)
1593             continue;
1594         if (!strcmp((const char *) n->name, "metadata"))
1595         {
1596             xmlChar *type = xmlGetProp(n, (xmlChar *) "type");
1597             if (type)
1598             {
1599                 size_t len;
1600                 int substring;
1601                 const char *eq;
1602
1603                 if ((eq = strchr(s, '=')))
1604                     substring = 0;
1605                 else if ((eq = strchr(s, '~')))
1606                     substring = 1;
1607                 if (eq)
1608                     len = eq - s;
1609                 else
1610                     len = strlen(s);
1611                 if (len == strlen((const char *)type) &&
1612                     !memcmp((const char *) type, s, len))
1613                 {
1614                     xmlChar *value = xmlNodeGetContent(n);
1615                     if (value && *value)
1616                     {
1617                         if (!eq ||
1618                             (substring && strstr((const char *) value, eq+1)) ||
1619                             (!substring && !strcmp((const char *) value, eq + 1)))
1620                             match = 1;
1621                     }
1622                     xmlFree(value);
1623                 }
1624                 xmlFree(type);
1625             }
1626         }
1627     }
1628     return match;
1629 }
1630
1631
1632 static int ingest_to_cluster(struct client *cl,
1633                              xmlDoc *xdoc,
1634                              xmlNode *root,
1635                              int record_no,
1636                              const char *mergekey_norm);
1637
1638 /** \brief ingest XML record
1639     \param cl client holds the result set for record
1640     \param rec record buffer (0 terminated)
1641     \param record_no record position (1, 2, ..)
1642     \param nmem working NMEM
1643     \retval 0 OK
1644     \retval -1 failure
1645     \retval -2 Filtered
1646 */
1647 int ingest_record(struct client *cl, const char *rec,
1648                   int record_no, NMEM nmem)
1649 {
1650     struct session *se = client_get_session(cl);
1651     int ret = 0;
1652     struct session_database *sdb = client_get_database(cl);
1653     struct conf_service *service = se->service;
1654     xmlDoc *xdoc = normalize_record(se, sdb, service, rec, nmem);
1655     xmlNode *root;
1656     const char *mergekey_norm;
1657
1658     if (!xdoc)
1659         return -1;
1660
1661     root = xmlDocGetRootElement(xdoc);
1662
1663     if (!check_record_filter(root, sdb))
1664     {
1665         session_log(se, YLOG_LOG, "Filtered out record no %d from %s", record_no, sdb->database->id);
1666         xmlFreeDoc(xdoc);
1667         return -2;
1668     }
1669
1670     mergekey_norm = get_mergekey(xdoc, cl, record_no, service, nmem);
1671     if (!mergekey_norm)
1672     {
1673         session_log(se, YLOG_WARN, "Got no mergekey");
1674         xmlFreeDoc(xdoc);
1675         return -1;
1676     }
1677     session_enter(se, "ingest_record");
1678     if (client_get_session(cl) == se)
1679         ret = ingest_to_cluster(cl, xdoc, root, record_no, mergekey_norm);
1680     session_leave(se, "ingest_record");
1681
1682     xmlFreeDoc(xdoc);
1683     return ret;
1684 }
1685
1686 //    struct conf_metadata *ser_md = &service->metadata[md_field_id];
1687 //    struct record_metadata *rec_md = record->metadata[md_field_id];
1688 static int match_metadata_local(struct conf_metadata *ser_md,
1689                                 struct record_metadata *rec_md0,
1690                                 char **values, int num_v)
1691 {
1692     int i;
1693     struct record_metadata *rec_md = rec_md0;
1694     for (i = 0; i < num_v; )
1695     {
1696         if (rec_md)
1697         {
1698             if (ser_md->type == Metadata_type_year
1699                 || ser_md->type == Metadata_type_date)
1700             {
1701                 int y = atoi(values[i]);
1702                 if (y >= rec_md->data.number.min
1703                     && y <= rec_md->data.number.max)
1704                     break;
1705             }
1706             else
1707             {
1708                 yaz_log(YLOG_DEBUG, "cmp: '%s' '%s'", rec_md->data.text.disp, values[i]);
1709                 if (!strcmp(rec_md->data.text.disp, values[i]))
1710                 {
1711                     // Value equals, should not be filtered.
1712                     break;
1713                 }
1714             }
1715             rec_md = rec_md->next;
1716         }
1717         else
1718         {
1719             rec_md = rec_md0;
1720             i++;
1721         }
1722     }
1723     return i < num_v ? 1 : 0;
1724 }
1725
1726 int session_check_cluster_limit(struct session *se, struct record_cluster *rec)
1727 {
1728     int i;
1729     struct conf_service *service = se->service;
1730     int ret = 1;
1731     const char *name;
1732     const char *value;
1733     NMEM nmem_tmp = nmem_create();
1734
1735     for (i = 0; (name = facet_limits_get(se->facet_limits, i, &value)); i++)
1736     {
1737         int j;
1738         for (j = 0; j < service->num_metadata; j++)
1739         {
1740             struct conf_metadata *md = service->metadata + j;
1741             if (!strcmp(md->name, name) && md->limitcluster)
1742             {
1743                 char **values = 0;
1744                 int num = 0;
1745                 int md_field_id =
1746                     conf_service_metadata_field_id(service,
1747                                                    md->limitcluster);
1748
1749                 if (md_field_id < 0)
1750                 {
1751                     ret = 0;
1752                     break;
1753                 }
1754
1755                 nmem_strsplit_escape2(nmem_tmp, "|", value, &values,
1756                                       &num, 1, '\\', 1);
1757
1758                 if (!match_metadata_local(&service->metadata[md_field_id],
1759                                           rec->metadata[md_field_id],
1760                                           values, num))
1761                 {
1762                     ret = 0;
1763                     break;
1764                 }
1765             }
1766         }
1767     }
1768     nmem_destroy(nmem_tmp);
1769     return ret;
1770 }
1771
1772 // Skip record on non-zero
1773 static int check_limit_local(struct client *cl,
1774                              struct record *record,
1775                              int record_no)
1776 {
1777     int skip_record = 0;
1778     struct session *se = client_get_session(cl);
1779     struct conf_service *service = se->service;
1780     NMEM nmem_tmp = nmem_create();
1781     struct session_database *sdb = client_get_database(cl);
1782     int l = 0;
1783     while (!skip_record)
1784     {
1785         int md_field_id;
1786         char **values = 0;
1787         int num_v = 0;
1788         const char *name =
1789             client_get_facet_limit_local(cl, sdb, &l, nmem_tmp,
1790                                          &num_v, &values);
1791         if (!name)
1792             break;
1793
1794         if (!strcmp(name, "*"))
1795         {
1796             for (md_field_id = 0; md_field_id < service->num_metadata;
1797                  md_field_id++)
1798             {
1799                 if (match_metadata_local(
1800                         &service->metadata[md_field_id],
1801                         record->metadata[md_field_id],
1802                         values, num_v))
1803                     break;
1804             }
1805             if (md_field_id == service->num_metadata)
1806                 skip_record = 1;
1807         }
1808         else
1809         {
1810             md_field_id = conf_service_metadata_field_id(service, name);
1811             if (md_field_id < 0)
1812             {
1813                 skip_record = 1;
1814                 break;
1815             }
1816             if (!match_metadata_local(
1817                     &service->metadata[md_field_id],
1818                     record->metadata[md_field_id],
1819                     values, num_v))
1820             {
1821                 skip_record = 1;
1822             }
1823         }
1824     }
1825     nmem_destroy(nmem_tmp);
1826     return skip_record;
1827 }
1828
1829 static int ingest_to_cluster(struct client *cl,
1830                              xmlDoc *xdoc,
1831                              xmlNode *root,
1832                              int record_no,
1833                              const char *mergekey_norm)
1834 {
1835     xmlNode *n;
1836     xmlChar *type = 0;
1837     xmlChar *value = 0;
1838     struct session *se = client_get_session(cl);
1839     struct conf_service *service = se->service;
1840     int term_factor = 1;
1841     struct record_cluster *cluster;
1842     struct record_metadata **metadata0;
1843     struct session_database *sdb = client_get_database(cl);
1844     struct record *record = record_create(se->nmem,
1845                                           service->num_metadata,
1846                                           service->num_sortkeys, cl,
1847                                           record_no);
1848
1849     for (n = root->children; n; n = n->next)
1850     {
1851         if (type)
1852             xmlFree(type);
1853         if (value)
1854             xmlFree(value);
1855         type = value = 0;
1856
1857         if (n->type != XML_ELEMENT_NODE)
1858             continue;
1859         if (!strcmp((const char *) n->name, "metadata"))
1860         {
1861             struct conf_metadata *ser_md = 0;
1862             struct record_metadata **wheretoput = 0;
1863             struct record_metadata *rec_md = 0;
1864             int md_field_id = -1;
1865
1866             type = xmlGetProp(n, (xmlChar *) "type");
1867             value = xmlNodeListGetString(xdoc, n->children, 1);
1868
1869             if (!type || !value || !*value)
1870                 continue;
1871
1872             md_field_id
1873                 = conf_service_metadata_field_id(service, (const char *) type);
1874             if (md_field_id < 0)
1875             {
1876                 if (se->number_of_warnings_unknown_metadata == 0)
1877                 {
1878                     session_log(se, YLOG_WARN,
1879                             "Ignoring unknown metadata element: %s", type);
1880                 }
1881                 se->number_of_warnings_unknown_metadata++;
1882                 continue;
1883             }
1884
1885             ser_md = &service->metadata[md_field_id];
1886
1887             // non-merged metadata
1888             rec_md = record_metadata_init(se->nmem, (const char *) value,
1889                                           ser_md->type, n->properties);
1890             if (!rec_md)
1891             {
1892                 session_log(se, YLOG_WARN, "bad metadata data '%s' "
1893                             "for element '%s'", value, type);
1894                 continue;
1895             }
1896             wheretoput = &record->metadata[md_field_id];
1897             while (*wheretoput)
1898                 wheretoput = &(*wheretoput)->next;
1899             *wheretoput = rec_md;
1900         }
1901     }
1902
1903     if (check_limit_local(cl, record, record_no))
1904     {
1905         session_log(se, YLOG_LOG, "Facet filtered out record no %d from %s",
1906                     record_no, sdb->database->id);
1907         if (type)
1908             xmlFree(type);
1909         if (value)
1910             xmlFree(value);
1911         return -2;
1912     }
1913     cluster = reclist_insert(se->reclist, service, record,
1914                              mergekey_norm, &se->total_merged);
1915     if (!cluster)
1916         return 0; // complete match with existing record
1917
1918     {
1919         const char *use_term_factor_str =
1920             session_setting_oneval(sdb, PZ_TERMLIST_TERM_FACTOR);
1921         if (use_term_factor_str && use_term_factor_str[0] == '1')
1922         {
1923             int maxrecs = client_get_maxrecs(cl);
1924             int hits = (int) client_get_hits(cl);
1925             term_factor = MAX(hits, maxrecs) /  MAX(1, maxrecs);
1926             assert(term_factor >= 1);
1927             yaz_log(YLOG_DEBUG, "Using term factor: %d (%d / %d)", term_factor, MAX(hits, maxrecs), MAX(1, maxrecs));
1928         }
1929     }
1930
1931     if (global_parameters.dump_records)
1932         session_log(se, YLOG_LOG, "Cluster id %s from %s (#%d)", cluster->recid,
1933                     sdb->database->id, record_no);
1934
1935
1936     relevance_newrec(se->relevance, cluster);
1937
1938     // original metadata, to check if first existence of a field
1939     metadata0 = xmalloc(sizeof(*metadata0) * service->num_metadata);
1940     memcpy(metadata0, cluster->metadata,
1941            sizeof(*metadata0) * service->num_metadata);
1942
1943     // now parsing XML record and adding data to cluster or record metadata
1944     for (n = root->children; n; n = n->next)
1945     {
1946         pp2_charset_token_t prt;
1947         if (type)
1948             xmlFree(type);
1949         if (value)
1950             xmlFree(value);
1951         type = value = 0;
1952
1953         if (n->type != XML_ELEMENT_NODE)
1954             continue;
1955         if (!strcmp((const char *) n->name, "metadata"))
1956         {
1957             struct conf_metadata *ser_md = 0;
1958             struct conf_sortkey *ser_sk = 0;
1959             struct record_metadata **wheretoput = 0;
1960             struct record_metadata *rec_md = 0;
1961             int md_field_id = -1;
1962             int sk_field_id = -1;
1963             const char *rank;
1964             xmlChar *xml_rank;
1965
1966             type = xmlGetProp(n, (xmlChar *) "type");
1967             value = xmlNodeListGetString(xdoc, n->children, 1);
1968
1969             if (!type || !value || !*value)
1970                 continue;
1971
1972             md_field_id
1973                 = conf_service_metadata_field_id(service, (const char *) type);
1974             if (md_field_id < 0)
1975                 continue;
1976
1977             ser_md = &service->metadata[md_field_id];
1978
1979             if (ser_md->sortkey_offset >= 0)
1980             {
1981                 sk_field_id = ser_md->sortkey_offset;
1982                 ser_sk = &service->sortkeys[sk_field_id];
1983             }
1984
1985             // merged metadata
1986             rec_md = record_metadata_init(se->nmem, (const char *) value,
1987                                           ser_md->type, 0);
1988
1989             // see if the field was not in cluster already (from beginning)
1990
1991             if (!rec_md)
1992                 continue;
1993
1994             xml_rank = xmlGetProp(n, (xmlChar *) "rank");
1995             rank = xml_rank ? (const char *) xml_rank : ser_md->rank;
1996
1997             wheretoput = &cluster->metadata[md_field_id];
1998
1999             if (ser_md->merge == Metadata_merge_first)
2000             {
2001                 if (!metadata0[md_field_id])
2002                 {
2003                     while (*wheretoput)
2004                         wheretoput = &(*wheretoput)->next;
2005                     *wheretoput = rec_md;
2006                 }
2007             }
2008             else if (ser_md->merge == Metadata_merge_unique)
2009             {
2010                 while (*wheretoput)
2011                 {
2012                     if (!strcmp((const char *) (*wheretoput)->data.text.disp,
2013                                 rec_md->data.text.disp))
2014                         break;
2015                     wheretoput = &(*wheretoput)->next;
2016                 }
2017                 if (!*wheretoput)
2018                     *wheretoput = rec_md;
2019             }
2020             else if (ser_md->merge == Metadata_merge_longest)
2021             {
2022                 if (!*wheretoput
2023                     || strlen(rec_md->data.text.disp)
2024                     > strlen((*wheretoput)->data.text.disp))
2025                 {
2026                     *wheretoput = rec_md;
2027                     if (ser_sk)
2028                     {
2029                         const char *sort_str = 0;
2030                         int skip_article =
2031                             ser_sk->type == Metadata_sortkey_skiparticle;
2032
2033                         if (!cluster->sortkeys[sk_field_id])
2034                             cluster->sortkeys[sk_field_id] =
2035                                 nmem_malloc(se->nmem,
2036                                             sizeof(union data_types));
2037
2038                         prt =
2039                             pp2_charset_token_create(service->charsets, "sort");
2040
2041                         pp2_charset_token_first(prt, rec_md->data.text.disp,
2042                                                 skip_article);
2043
2044                         pp2_charset_token_next(prt);
2045
2046                         sort_str = pp2_get_sort(prt);
2047
2048                         cluster->sortkeys[sk_field_id]->text.disp =
2049                             rec_md->data.text.disp;
2050                         if (!sort_str)
2051                         {
2052                             sort_str = rec_md->data.text.disp;
2053                             session_log(se, YLOG_WARN,
2054                                     "Could not make sortkey. Bug #1858");
2055                         }
2056                         cluster->sortkeys[sk_field_id]->text.sort =
2057                             nmem_strdup(se->nmem, sort_str);
2058                         pp2_charset_token_destroy(prt);
2059                     }
2060                 }
2061             }
2062             else if (ser_md->merge == Metadata_merge_all)
2063             {
2064                 while (*wheretoput)
2065                     wheretoput = &(*wheretoput)->next;
2066                 *wheretoput = rec_md;
2067             }
2068             else if (ser_md->merge == Metadata_merge_range)
2069             {
2070                 if (!*wheretoput)
2071                 {
2072                     *wheretoput = rec_md;
2073                     if (ser_sk)
2074                         cluster->sortkeys[sk_field_id]
2075                             = &rec_md->data;
2076                 }
2077                 else
2078                 {
2079                     int this_min = rec_md->data.number.min;
2080                     int this_max = rec_md->data.number.max;
2081                     if (this_min < (*wheretoput)->data.number.min)
2082                         (*wheretoput)->data.number.min = this_min;
2083                     if (this_max > (*wheretoput)->data.number.max)
2084                         (*wheretoput)->data.number.max = this_max;
2085                 }
2086             }
2087
2088             // ranking of _all_ fields enabled ...
2089             if (rank)
2090             {
2091                 relevance_countwords(se->relevance, cluster,
2092                                      (char *) value, rank, ser_md->name);
2093             }
2094
2095             // construct facets ... unless the client already has reported them
2096             if (ser_md->termlist && !client_has_facet(cl, (char *) type))
2097             {
2098                 if (ser_md->type == Metadata_type_year)
2099                 {
2100                     char year[64];
2101                     sprintf(year, "%d", rec_md->data.number.max);
2102
2103                     add_facet(se, (char *) type, year, term_factor);
2104                     if (rec_md->data.number.max != rec_md->data.number.min)
2105                     {
2106                         sprintf(year, "%d", rec_md->data.number.min);
2107                         add_facet(se, (char *) type, year, term_factor);
2108                     }
2109                 }
2110                 else
2111                     add_facet(se, (char *) type, (char *) value, term_factor);
2112             }
2113
2114             // cleaning up
2115             if (xml_rank)
2116                 xmlFree(xml_rank);
2117             xmlFree(type);
2118             xmlFree(value);
2119             type = value = 0;
2120         }
2121         else
2122         {
2123             if (se->number_of_warnings_unknown_elements == 0)
2124                 session_log(se, YLOG_WARN,
2125                         "Unexpected element in internal record: %s", n->name);
2126             se->number_of_warnings_unknown_elements++;
2127         }
2128     }
2129     if (type)
2130         xmlFree(type);
2131     if (value)
2132         xmlFree(value);
2133
2134     xfree(metadata0);
2135     relevance_donerecord(se->relevance, cluster);
2136     se->total_records++;
2137
2138     return 0;
2139 }
2140
2141 void session_log(struct session *s, int level, const char *fmt, ...)
2142 {
2143     char buf[1024];
2144     va_list ap;
2145     va_start(ap, fmt);
2146
2147     yaz_vsnprintf(buf, sizeof(buf)-30, fmt, ap);
2148     yaz_log(level, "Session %u: %s", s->session_id, buf);
2149
2150     va_end(ap);
2151 }
2152
2153 /*
2154  * Local variables:
2155  * c-basic-offset: 4
2156  * c-file-style: "Stroustrup"
2157  * indent-tabs-mode: nil
2158  * End:
2159  * vim: shiftwidth=4 tabstop=8 expandtab
2160  */
2161