-/* $Id: zrpn.c,v 1.220 2006-06-22 23:06:06 adam Exp $
+/* $Id: zrpn.c,v 1.221 2006-06-23 11:21:38 adam Exp $
Copyright (C) 1995-2006
Index Data ApS
break;
case 3:
case 102:
- case 103:
case -1:
if (!**term_sub)
return 1;
strcat(term_tmp, term_component);
strcat(term_tmp, ")");
break;
+ case 103:
+ yaz_log(log_level_rpn, "Relation always matches");
+ /* skip to end of term (we don't care what it is) */
+ while (**term_sub != '\0')
+ (*term_sub)++;
+ break;
default:
*error_code = YAZ_BIB1_UNSUPP_RELATION_ATTRIBUTE;
return 0;
const char *term_ref_id_str = 0;
*rset = 0;
- term_limits_APT(zh, zapt, &hits_limit_value, &term_ref_id_str,
- stream);
+ term_limits_APT(zh, zapt, &hits_limit_value, &term_ref_id_str, stream);
grep_info->isam_p_indx = 0;
res = string_term(zh, zapt, term_sub, attributeSet, stream, grep_info,
reg_type, complete_flag, num_bases, basenames,
char ord_buf[32];
int ord_len, i;
- termp = *term_sub;
+ termp = *term_sub; /* start of term for each database */
if (zebraExplain_curDatabase (zh->reg->zei, basenames[base_no]))
{
return ZEBRA_FAIL;
}
- if (zebra_apt_get_ord(zh, zapt, zinfo_index_category_index,
- reg_type, xpath_use,
+ if (zebra_apt_get_ord(zh, zapt, reg_type, xpath_use,
attributeSet, &ord) != ZEBRA_OK)
continue;
+ bases_ok++;
+
*ol = ord_list_append(stream, *ol, ord);
-
- if (prefix_len)
- term_dict[prefix_len++] = '|';
- else
- term_dict[prefix_len++] = '(';
-
ord_len = key_SU_encode (ord, ord_buf);
+
+ term_dict[prefix_len++] = '(';
for (i = 0; i<ord_len; i++)
{
- term_dict[prefix_len++] = 1;
+ term_dict[prefix_len++] = 1; /* our internal regexp escape char */
term_dict[prefix_len++] = ord_buf[i];
}
-
- bases_ok++;
-
term_dict[prefix_len++] = ')';
term_dict[prefix_len] = '\0';
j = prefix_len;
{
case -1: /* not specified */
case 100: /* do not truncate */
- if (!string_relation (zh, zapt, &termp, term_dict,
- attributeSet,
- reg_type, space_split, term_dst,
- &relation_error))
+ if (!string_relation(zh, zapt, &termp, term_dict,
+ attributeSet,
+ reg_type, space_split, term_dst,
+ &relation_error))
{
if (relation_error)
{
struct grep_info grep_info;
const char *termp = termz;
int alloc_sets = 0;
- int empty_term = *termz ? 0 : 1;
- empty_term = 0;
*num_result_sets = 0;
*term_dst = 0;
if (grep_info_prepare(zh, zapt, &grep_info, reg_type) == ZEBRA_FAIL)
break;
(*num_result_sets)++;
- if (empty_term)
- break;
if (!*termp)
break;
}
return ZEBRA_OK;
}
-
-static ZEBRA_RES always_term(ZebraHandle zh, Z_AttributesPlusTerm *zapt,
- oid_value attributeSet, NMEM stream,
- struct grep_info *grep_info,
- int reg_type, int complete_flag,
- int num_bases, char **basenames,
- const char *xpath_use,
- struct ord_list **ol)
-{
- char term_dict[2*IT_MAX_WORD+4000];
- int r, base_no;
- struct rpn_char_map_info rcmi;
-
- int bases_ok = 0; /* no of databases with OK attribute */
-
- *ol = ord_list_create(stream);
-
- rpn_char_map_prepare (zh->reg, reg_type, &rcmi);
-
- for (base_no = 0; base_no < num_bases; base_no++)
- {
- int ord = -1;
- int regex_range = 0;
- int max_pos, prefix_len = 0;
- char ord_buf[32];
- int ord_len, i;
-
- if (zebraExplain_curDatabase (zh->reg->zei, basenames[base_no]))
- {
- zebra_setError(zh, YAZ_BIB1_DATABASE_UNAVAILABLE,
- basenames[base_no]);
- return ZEBRA_FAIL;
- }
-
- if (zebra_apt_get_ord(zh, zapt, zinfo_index_category_alwaysmatches,
- reg_type, xpath_use,
- attributeSet, &ord) != ZEBRA_OK)
- continue;
-
- *ol = ord_list_append(stream, *ol, ord);
-
- if (prefix_len)
- term_dict[prefix_len++] = '|';
- else
- term_dict[prefix_len++] = '(';
-
- ord_len = key_SU_encode (ord, ord_buf);
- for (i = 0; i<ord_len; i++)
- {
- term_dict[prefix_len++] = 1;
- term_dict[prefix_len++] = ord_buf[i];
- }
-
- term_dict[prefix_len++] = ')';
- term_dict[prefix_len] = '\0';
-
- bases_ok++;
-
- r = dict_lookup_grep(zh->reg->dict, term_dict, regex_range,
- grep_info, &max_pos,
- ord_len /* number of "exact" chars */,
- grep_handle);
- }
- if (!bases_ok)
- return ZEBRA_FAIL;
- return ZEBRA_OK;
-}
-
-static ZEBRA_RES rpn_search_APT_alwaysmatches(ZebraHandle zh,
- Z_AttributesPlusTerm *zapt,
- const char *termz_org,
- oid_value attributeSet,
- NMEM stream,
- int reg_type, int complete_flag,
- const char *rank_type,
- const char *xpath_use,
- int num_bases, char **basenames,
- NMEM rset_nmem,
- RSET *rset,
- struct rset_key_control *kc)
-{
- const char *term_dst = "always";
- struct grep_info grep_info;
- zint hits_limit_value;
- const char *term_ref_id_str = 0;
- ZEBRA_RES res;
- struct ord_list *ol;
-
- term_limits_APT(zh, zapt, &hits_limit_value, &term_ref_id_str,
- stream);
- if (grep_info_prepare(zh, zapt, &grep_info, reg_type) == ZEBRA_FAIL)
- return ZEBRA_FAIL;
-
- grep_info.isam_p_indx = 0;
-
- res = always_term(zh, zapt, attributeSet, stream, &grep_info,
- reg_type, complete_flag, num_bases, basenames,
- xpath_use, &ol);
- if (res == ZEBRA_OK)
- {
- *rset = rset_trunc(zh, grep_info.isam_p_buf,
- grep_info.isam_p_indx, term_dst, strlen(term_dst),
- rank_type, 1 /* preserve pos */,
- zapt->term->which, rset_nmem,
- kc, kc->scope, ol, reg_type, hits_limit_value,
- term_ref_id_str);
- if (!*rset)
- res = ZEBRA_FAIL;
- }
- grep_info_delete (&grep_info);
- return res;
-}
-
static ZEBRA_RES rpn_search_APT_phrase(ZebraHandle zh,
Z_AttributesPlusTerm *zapt,
const char *termz_org,
yaz_log(log_level_rpn, "numeric relation value=%d", relation_value);
- if (!term_100(zh->reg->zebra_maps, reg_type, term_sub, term_tmp, 1,
- term_dst))
- return 0;
- term_value = atoi (term_tmp);
switch (relation_value)
{
case 1:
yaz_log(log_level_rpn, "Relation <");
+ if (!term_100(zh->reg->zebra_maps, reg_type, term_sub, term_tmp, 1,
+ term_dst))
+ return 0;
+ term_value = atoi (term_tmp);
gen_regular_rel(term_tmp, term_value-1, 1);
break;
case 2:
yaz_log(log_level_rpn, "Relation <=");
+ if (!term_100(zh->reg->zebra_maps, reg_type, term_sub, term_tmp, 1,
+ term_dst))
+ return 0;
+ term_value = atoi (term_tmp);
gen_regular_rel(term_tmp, term_value, 1);
break;
case 4:
yaz_log(log_level_rpn, "Relation >=");
+ if (!term_100(zh->reg->zebra_maps, reg_type, term_sub, term_tmp, 1,
+ term_dst))
+ return 0;
+ term_value = atoi (term_tmp);
gen_regular_rel(term_tmp, term_value, 0);
break;
case 5:
yaz_log(log_level_rpn, "Relation >");
+ if (!term_100(zh->reg->zebra_maps, reg_type, term_sub, term_tmp, 1,
+ term_dst))
+ return 0;
+ term_value = atoi (term_tmp);
gen_regular_rel(term_tmp, term_value+1, 0);
break;
case -1:
case 3:
yaz_log(log_level_rpn, "Relation =");
+ if (!term_100(zh->reg->zebra_maps, reg_type, term_sub, term_tmp, 1,
+ term_dst))
+ return 0;
+ term_value = atoi (term_tmp);
sprintf(term_tmp, "(0*%d)", term_value);
break;
+ case 103:
+ /* term_tmp untouched.. */
+ while (**term_sub != '\0')
+ (*term_sub)++;
+ break;
default:
*error_code = YAZ_BIB1_UNSUPP_RELATION_ATTRIBUTE;
return 0;
return ZEBRA_FAIL;
}
- if (zebra_apt_get_ord(zh, zapt, zinfo_index_category_index,
- reg_type, xpath_use, attributeSet, &ord)
- != ZEBRA_OK)
+ if (zebra_apt_get_ord(zh, zapt, reg_type, xpath_use,
+ attributeSet, &ord) != ZEBRA_OK)
continue;
+ bases_ok++;
*ol = ord_list_append(stream, *ol, ord);
- if (prefix_len)
- term_dict[prefix_len++] = '|';
- else
- term_dict[prefix_len++] = '(';
-
ord_len = key_SU_encode (ord, ord_buf);
+
+ term_dict[prefix_len++] = '(';
for (i = 0; i < ord_len; i++)
{
term_dict[prefix_len++] = 1;
}
term_dict[prefix_len++] = ')';
term_dict[prefix_len] = '\0';
- bases_ok++;
+
if (!numeric_relation(zh, zapt, &termp, term_dict,
attributeSet, grep_info, &max_pos, reg_type,
term_dst, &relation_error))
if (!result_sets[num_result_sets])
break;
num_result_sets++;
+ if (!*termp)
+ break;
}
grep_info_delete(&grep_info);
- if (termp)
- {
- int i;
- for (i = 0; i<num_result_sets; i++)
- rset_delete(result_sets[i]);
- return ZEBRA_FAIL;
- }
+
+ if (res != ZEBRA_OK)
+ return res;
if (num_result_sets == 0)
*rset = rset_create_null(rset_nmem, kc, 0);
- if (num_result_sets == 1)
+ else if (num_result_sets == 1)
*rset = result_sets[0];
else
- *rset = rset_create_and(rset_nmem, kc, kc->scope,
+ *rset = rset_create_and(rset_nmem, kc, kc->scope,
num_result_sets, result_sets);
if (!*rset)
- return ZEBRA_FAIL;
+ return ZEBRA_FAIL;
return ZEBRA_OK;
}
xpath_use = ZEBRA_XPATH_ATTR_CDATA; /* last step is attribute */
else
xpath_use = ZEBRA_XPATH_CDATA; /* searching for cdata */
+
+ if (1)
+ {
+ AttrType relation;
+ int relation_value;
+
+ attr_init_APT(&relation, zapt, 2);
+ relation_value = attr_find(&relation, NULL);
+
+ if (relation_value == 103) /* alwaysmatches */
+ {
+ *rset = 0; /* signal no "term" set */
+ return rpn_search_xpath(zh, num_bases, basenames,
+ stream, rank_type, *rset,
+ xpath_len, xpath, rset_nmem, rset, kc);
+ }
+ }
}
/* search using one of the various search type strategies
num_bases, basenames, rset_nmem,
rset, kc);
}
- else if (!strcmp(search_type, "always"))
- {
- if (xpath_len >= 0) /* alwaysmatches and X-Path ? */
- {
- *rset = 0; /* signal no "term" set */
- return rpn_search_xpath(zh, num_bases, basenames,
- stream, rank_type, *rset,
- xpath_len, xpath, rset_nmem, rset, kc);
- }
- else
- {
- res = rpn_search_APT_alwaysmatches(zh, zapt, termz,
- attributeSet, stream,
- reg_id, complete_flag,
- rank_type,
- xpath_use,
- num_bases, basenames, rset_nmem,
- rset, kc);
- }
- }
else
{
zebra_setError(zh, YAZ_BIB1_UNSUPP_SEARCH, 0);
*num_entries = 0;
return ZEBRA_FAIL;
}
- if (zebra_apt_get_ord(zh, zapt, zinfo_index_category_index,
- index_type, 0, attributeset, &ord)
+ if (zebra_apt_get_ord(zh, zapt, index_type, 0, attributeset, &ord)
!= ZEBRA_OK)
continue;
ords[ord_no++] = ord;