+ }
+ qsort(col, no_terms_collect, sizeof(*col), term_qsort_handle);
+}
+
+static struct term_collect *term_collect_create(zebra_strmap_t sm,
+ int no_terms_collect,
+ NMEM nmem)
+{
+ const char *term;
+ void *data_buf;
+ size_t data_len;
+ zebra_strmap_it it;
+ struct term_collect *col = nmem_malloc(nmem,
+ sizeof *col *no_terms_collect);
+ int i;
+ for (i = 0; i < no_terms_collect; i++)
+ {
+ col[i].term = 0;
+ col[i].oc = 0;
+ col[i].set_occur = 0;
+ }
+ /* iterate over terms and collect the most frequent ones */
+ it = zebra_strmap_it_create(sm);
+ while ((term = zebra_strmap_it_next(it, &data_buf, &data_len)))
+ {
+ /* invariant:
+ col[0] has lowest oc . col[no_terms_collect-1] has highest oc */
+ int oc = *(int*) data_buf;
+ int j = 0;
+ /* insertion may be slow but terms terms will be "infrequent" and
+ thus number of iterations should be small below
+ */
+ while (j < no_terms_collect && oc > col[j].oc)
+ j++;
+ if (j)
+ { /* oc <= col[j] and oc > col[j-1] */
+ --j;
+ memmove(col, col+1, sizeof(*col) * j);
+ col[j].term = term;
+ col[j].oc = oc;
+ }
+ }
+ zebra_strmap_it_destroy(it);
+ return col;
+}
+
+static int perform_facet_sort(ZebraHandle zh, int no_ord, int *ord_array,
+ zebra_strmap_t *map_array,
+ int num_recs, ZebraMetaRecord *poset)
+{
+ int rec_i;
+ WRBUF w = wrbuf_alloc();
+ int ord_i;
+
+ for (ord_i = 0; ord_i < no_ord; ord_i++)
+ {
+ for (rec_i = 0; rec_i < num_recs; rec_i++)
+ {
+ if (!poset[rec_i].sysno)
+ continue;
+
+ zebra_sort_sysno(zh->reg->sort_index, poset[rec_i].sysno);
+ zebra_sort_type(zh->reg->sort_index, ord_array[ord_i]);
+
+ wrbuf_rewind(w);
+ if (zebra_sort_read(zh->reg->sort_index, 0, w))
+ {
+ zebra_strmap_t sm = map_array[ord_i];
+ int off = 0;
+ while (off != wrbuf_len(w))
+ {
+ const char *str = wrbuf_buf(w) + off;
+ int *freq = zebra_strmap_lookup(sm, str, 0, 0);
+ if (freq)
+ (*freq)++;
+ else
+ {
+ int v = 1;
+ zebra_strmap_add(sm, str, &v, sizeof v);
+ }
+ off += strlen(str)+1;
+ }
+ }
+ }
+ }
+ wrbuf_destroy(w);
+ return 0;
+}
+
+
+static int perform_facet_index(ZebraHandle zh,
+ struct special_fetch_s *fi,
+ int no_ord, int *ord_array,
+ zebra_strmap_t *map_array,
+ int num_recs, ZebraMetaRecord *poset,
+ struct index_spec *spec_list)
+{
+ int max_chunks = 2;
+ int rec_i;
+ res_get_int(zh->res, "facetMaxChunks", &max_chunks);
+
+ for (rec_i = 0; rec_i < num_recs; rec_i++)
+ {
+ int ret;
+ int j;
+ zint sysnos[MAX_SYSNOS_PER_RECORD];
+ int no_sysnos = MAX_SYSNOS_PER_RECORD;
+ if (!poset[rec_i].sysno)
+ continue;
+ ret = zebra_result_recid_to_sysno(zh, fi->setname,
+ poset[rec_i].sysno,
+ sysnos, &no_sysnos);
+ assert(no_sysnos > 0);
+ yaz_log(YLOG_DEBUG, "Analyzing rec=%d ISAM sysno=" ZINT_FORMAT " chunks=%d",
+ rec_i, poset[rec_i].sysno, no_sysnos);
+ for (j = 0; j < no_sysnos && j < max_chunks; j++)
+ {
+ size_t slen;
+ const char *str;
+ struct it_key key_in;
+ Record rec = rec_get(zh->reg->records, sysnos[j]);
+ zebra_rec_keys_t keys = zebra_rec_keys_open();
+ zebra_rec_keys_set_buf(keys, rec->info[recInfo_delKeys],
+ rec->size[recInfo_delKeys], 0);
+
+ yaz_log(YLOG_DEBUG, "rec %d " ZINT_FORMAT " %s",
+ j, sysnos[j], zebra_rec_keys_empty(keys) ? "empty" : "non-empty");
+ if (zebra_rec_keys_rewind(keys))
+ {
+ while (zebra_rec_keys_read(keys, &str, &slen, &key_in))
+ {
+ int ord_i;
+ struct index_spec *spec;
+ for (spec = spec_list, ord_i = 0; ord_i < no_ord;
+ ord_i++, spec = spec->next)
+ {
+ int ord = CAST_ZINT_TO_INT(key_in.mem[0]);
+ if (ord == ord_array[ord_i] &&
+ str[0] != FIRST_IN_FIELD_CHAR)
+ {
+ int *freq;
+ zebra_strmap_t sm = map_array[ord_i];
+
+ freq = zebra_strmap_lookup(sm, str, 0, 0);
+ if (freq)
+ (*freq)++;
+ else
+ {
+ int v = 1;
+ zebra_strmap_add(sm, str, &v, sizeof v);
+ }
+ }
+ }
+ }
+ }
+ zebra_rec_keys_close(keys);
+ rec_free(&rec);
+ }
+ }
+ return 0;
+}
+
+static int perform_facet(ZebraHandle zh,
+ struct special_fetch_s *fi,
+ WRBUF result,
+ int num_recs, ZebraMetaRecord *poset,
+ struct index_spec *spec_list,
+ int no_ord, int *ord_array,
+ int use_xml,
+ zinfo_index_category_t cat)
+{
+ int i;
+ int ret = 0;
+ WRBUF wr = result;
+ struct index_spec *spec;
+ yaz_timing_t timing = yaz_timing_create();
+ zebra_strmap_t *map_array
+ = nmem_malloc(fi->nmem, sizeof *map_array * no_ord);
+ for (i = 0; i < no_ord; i++)
+ map_array[i] = zebra_strmap_create();
+
+ if (cat == zinfo_index_category_sort)
+ perform_facet_sort(zh, no_ord, ord_array, map_array,
+ num_recs, poset);
+ else
+ perform_facet_index(zh, fi, no_ord, ord_array, map_array,
+ num_recs, poset, spec_list);
+ yaz_timing_stop(timing);
+ yaz_log(YLOG_LOG, "facet first phase real=%4.2f cat=%s",
+ yaz_timing_get_real(timing),
+ (cat == zinfo_index_category_sort) ? "sort" : "index");
+ yaz_timing_start(timing);
+ for (spec = spec_list, i = 0; i < no_ord; i++, spec = spec->next)
+ {
+ int j;
+ NMEM nmem = nmem_create();
+ struct term_collect *col;
+ int no_collect_terms = 20;
+
+ if (spec->extra)
+ no_collect_terms = atoi(spec->extra);
+ if (no_collect_terms < 1)
+ no_collect_terms = 1;
+ col = term_collect_create(map_array[i], no_collect_terms, nmem);
+ term_collect_freq(zh, col, no_collect_terms, ord_array[i],
+ resultSetRef(zh, fi->setname),
+ cat == zinfo_index_category_sort ? 1.0 : -1.0);
+
+ if (use_xml)
+ wrbuf_printf(wr, " <facet type=\"%s\" index=\"%s\">\n",
+ spec->index_type, spec->index_name);