static void
roqet_query_walk(rasqal_query *rq, FILE *fh, int indent) {
rasqal_query_verb verb;
int i;
rasqal_graph_pattern* gp;
raptor_sequence *seq;
verb = rasqal_query_get_verb(rq);
roqet_write_indent(fh, indent);
fprintf(fh, "query verb: %s\n", rasqal_query_verb_as_string(verb));
i = rasqal_query_get_distinct(rq);
if(i != 0) {
roqet_write_indent(fh, indent);
fprintf(fh, "query asks for distinct results\n");
}
i = rasqal_query_get_limit(rq);
if(i >= 0) {
roqet_write_indent(fh, indent);
fprintf(fh, "query asks for result limits %d\n", i);
}
i = rasqal_query_get_offset(rq);
if(i >= 0) {
roqet_write_indent(fh, indent);
fprintf(fh, "query asks for result offset %d\n", i);
}
seq = rasqal_query_get_bound_variable_sequence(rq);
if(seq && raptor_sequence_size(seq) > 0) {
fprintf(fh, "query bound variables (%d): ",
raptor_sequence_size(seq));
i = 0;
while(1) {
rasqal_variable* v = (rasqal_variable*)raptor_sequence_get_at(seq, i);
if(!v)
break;
if(i > 0)
fputs(", ", fh);
roqet_query_write_variable(fh, v);
i++;
}
fputc('\n', fh);
}
gp = rasqal_query_get_query_graph_pattern(rq);
if(!gp)
return;
seq = rasqal_query_get_construct_triples_sequence(rq);
if(seq && raptor_sequence_size(seq) > 0) {
roqet_write_indent(fh, indent);
fprintf(fh, "query construct triples (%d) {\n",
raptor_sequence_size(seq));
i = 0;
while(1) {
rasqal_triple* t = rasqal_query_get_construct_triple(rq, i);
if(!t)
break;
roqet_write_indent(fh, indent + 2);
fprintf(fh, "triple #%d { ", i);
rasqal_triple_print(t, fh);
fputs(" }\n", fh);
i++;
}
roqet_write_indent(fh, indent);
fputs("}\n", fh);
}
/* look for binding rows */
seq = rasqal_query_get_bindings_variables_sequence(rq);
if(seq) {
roqet_write_indent(fh, indent);
fprintf(fh, "bindings variables (%d): ", raptor_sequence_size(seq));
i = 0;
while(1) {
rasqal_variable* v = rasqal_query_get_bindings_variable(rq, i);
if(!v)
break;
if(i > 0)
fputs(", ", fh);
roqet_query_write_variable(fh, v);
i++;
}
fputc('\n', fh);
seq = rasqal_query_get_bindings_rows_sequence(rq);
fprintf(fh, "bindings rows (%d) {\n", raptor_sequence_size(seq));
i = 0;
while(1) {
rasqal_row* row;
row = rasqal_query_get_bindings_row(rq, i);
if(!row)
break;
roqet_write_indent(fh, indent + 2);
fprintf(fh, "row #%d { ", i);
rasqal_row_print(row, fh);
fputs("}\n", fh);
i++;
}
}
fputs("query ", fh);
roqet_graph_pattern_walk(gp, -1, fh, indent);
}
int fs_bind_freq(fs_query_state *qs, fs_query *q, int block, rasqal_triple *t)
{
int ret = 100;
#if DEBUG_OPTIMISER
char dir = 'X';
#endif
if (!fs_opt_is_const(q->bb[block], t->subject) && !fs_opt_is_const(q->bb[block], t->predicate) &&
!fs_opt_is_const(q->bb[block], t->object) && !fs_opt_is_const(q->bb[block], t->origin)) {
#if DEBUG_OPTIMISER
dir = '?';
#endif
ret = INT_MAX;
} else if (!fs_opt_is_const(q->bb[block], t->subject) &&
!fs_opt_is_const(q->bb[block], t->object)) {
#if DEBUG_OPTIMISER
dir = '?';
#endif
ret = INT_MAX - 100;
} else if (qs->freq_s && fs_opt_num_vals(q->bb[block], t->subject) == 1 &&
fs_opt_num_vals(q->bb[block], t->predicate) == 1) {
#if DEBUG_OPTIMISER
dir = 's';
#endif
ret = calc_freq(q, block, qs->freq_s, t->subject, t->predicate);
} else if (qs->freq_o && fs_opt_num_vals(q->bb[block], t->object) == 1 &&
fs_opt_num_vals(q->bb[block], t->predicate) == 1) {
#if DEBUG_OPTIMISER
dir = 'o';
#endif
ret = calc_freq(q, block, qs->freq_o, t->object, t->predicate) +
q->segments * 50;
} else if (qs->freq_s && fs_opt_num_vals(q->bb[block], t->subject) == 1) {
#if DEBUG_OPTIMISER
dir = 's';
#endif
ret = calc_freq(q, block, qs->freq_s, t->subject, NULL);
} else if (qs->freq_o && fs_opt_num_vals(q->bb[block], t->object) == 1) {
#if DEBUG_OPTIMISER
dir = 'o';
#endif
ret = calc_freq(q, block, qs->freq_s, t->object, NULL) +
q->segments * 50;
/* cluases for if we have no freq data */
} else if (fs_opt_num_vals(q->bb[block], t->subject) < 1000000 &&
fs_opt_num_vals(q->bb[block], t->predicate) < 100 &&
fs_opt_num_vals(q->bb[block], t->object) == INT_MAX) {
#if DEBUG_OPTIMISER
dir = 's';
#endif
ret = fs_opt_num_vals(q->bb[block], t->subject) * fs_opt_num_vals(q->bb[block], t->predicate);
if (!fs_opt_is_bound(q->bb[block], t->subject) &&
!fs_opt_is_bound(q->bb[block], t->predicate) &&
!fs_opt_is_bound(q->bb[block], t->object)) {
ret *= (fs_binding_length(q->bb[block]) * 100);
}
} else if (fs_opt_num_vals(q->bb[block], t->object) < 1000000 &&
fs_opt_num_vals(q->bb[block], t->predicate) < 100 &&
fs_opt_num_vals(q->bb[block], t->subject) == INT_MAX) {
#if DEBUG_OPTIMISER
dir = 'o';
#endif
ret = fs_opt_num_vals(q->bb[block], t->predicate) * fs_opt_num_vals(q->bb[block], t->object);
if (!fs_opt_is_bound(q->bb[block], t->subject) &&
!fs_opt_is_bound(q->bb[block], t->predicate) &&
!fs_opt_is_bound(q->bb[block], t->object)) {
ret *= (fs_binding_length(q->bb[block]) * 100);
}
}
#if DEBUG_OPTIMISER
if (q->flags & FS_QUERY_EXPLAIN) {
printf("freq(%c, ", dir);
rasqal_triple_print(t, stdout);
printf(") = %d\n", ret);
}
#endif
return ret;
}
static void
roqet_graph_pattern_walk(rasqal_graph_pattern *gp, int gp_index,
FILE *fh, int indent) {
int triple_index = 0;
rasqal_graph_pattern_operator op;
int seen;
raptor_sequence *seq;
int idx;
rasqal_expression* expr;
rasqal_variable* var;
rasqal_literal* literal;
op = rasqal_graph_pattern_get_operator(gp);
roqet_write_indent(fh, indent);
fprintf(fh, "%s graph pattern",
rasqal_graph_pattern_operator_as_string(op));
idx = rasqal_graph_pattern_get_index(gp);
if(idx >= 0)
fprintf(fh, "[%d]", idx);
if(gp_index >= 0)
fprintf(fh, " #%d", gp_index);
fputs(" {\n", fh);
indent += 2;
/* look for LET variable and value */
var = rasqal_graph_pattern_get_variable(gp);
if(var) {
roqet_write_indent(fh, indent);
fprintf(fh, "%s := ", var->name);
rasqal_expression_print(var->expression, fh);
}
/* look for GRAPH literal */
literal = rasqal_graph_pattern_get_origin(gp);
if(literal) {
roqet_write_indent(fh, indent);
fputs("origin ", fh);
rasqal_literal_print(literal, fh);
fputc('\n', fh);
}
/* look for SERVICE literal */
literal = rasqal_graph_pattern_get_service(gp);
if(literal) {
roqet_write_indent(fh, indent);
rasqal_literal_print(literal, fh);
fputc('\n', fh);
}
/* look for triples */
seen = 0;
while(1) {
rasqal_triple* t;
t = rasqal_graph_pattern_get_triple(gp, triple_index);
if(!t)
break;
if(!seen) {
roqet_write_indent(fh, indent);
fputs("triples {\n", fh);
seen = 1;
}
roqet_write_indent(fh, indent + 2);
fprintf(fh, "triple #%d { ", triple_index);
rasqal_triple_print(t, fh);
fputs(" }\n", fh);
triple_index++;
}
if(seen) {
roqet_write_indent(fh, indent);
fputs("}\n", fh);
}
/* look for sub-graph patterns */
seq = rasqal_graph_pattern_get_sub_graph_pattern_sequence(gp);
if(seq && raptor_sequence_size(seq) > 0) {
roqet_write_indent(fh, indent);
fprintf(fh, "sub-graph patterns (%d) {\n", raptor_sequence_size(seq));
gp_index = 0;
while(1) {
rasqal_graph_pattern* sgp;
sgp = rasqal_graph_pattern_get_sub_graph_pattern(gp, gp_index);
if(!sgp)
break;
roqet_graph_pattern_walk(sgp, gp_index, fh, indent + 2);
gp_index++;
}
roqet_write_indent(fh, indent);
fputs("}\n", fh);
}
/* look for filter */
expr = rasqal_graph_pattern_get_filter_expression(gp);
if(expr) {
roqet_write_indent(fh, indent);
fputs("filter { ", fh);
rasqal_expression_print(expr, fh);
fputs("}\n", fh);
}
indent -= 2;
roqet_write_indent(fh, indent);
fputs("}\n", fh);
}
int fs_optimise_triple_pattern(fs_query_state *qs, fs_query *q, int block, rasqal_triple *patt[], int length, int start)
{
if (length - start < 2 || q->opt_level < 1) {
return 1;
}
rasqal_triple **pbuf = malloc(length * sizeof(rasqal_triple *));
memcpy(pbuf, patt, sizeof(rasqal_triple *) * length);
memset(patt, 0, length * sizeof(rasqal_triple *));
int append_pos = start;
for (int i=0; i<start; i++) {
pbuf[i] = patt[i];
}
/* roughly sort into order:
* const subject and predicate
* const predicate and object
* const subject
* const object
* const graph
* const predicate
* all variable
*/
#if 0
/* this code complicates things greatly, so I've removed it for now - swh
should maybe be reexamined if/when we get histograms back */
for (int i=start; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_bind_freq(qs, q, block, pbuf[i]) == 1) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
#endif
for (int i=start; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_opt_is_const(q->bb[block], pbuf[i]->subject) && fs_opt_is_const(q->bb[block], pbuf[i]->predicate) && fs_opt_is_bound(q->bb[block], pbuf[i]->object)) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
/* triples like :s :p ?o, where :p != rdf:type */
for (int i=0; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_opt_is_const(q->bb[block], pbuf[i]->subject) && fs_opt_is_const(q->bb[block], pbuf[i]->predicate) && fs_opt_is_bound(q->bb[block], pbuf[i]->object) && !fs_opt_literal_is_rdf_type(pbuf[i]->predicate)) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
/* triples like :s :p _, where :p != rdf:type */
for (int i=0; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_opt_is_const(q->bb[block], pbuf[i]->subject) && fs_opt_is_const(q->bb[block], pbuf[i]->predicate) && !fs_opt_literal_is_rdf_type(pbuf[i]->predicate)) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
/* triples like ?s :p :o, where :p != rdf:type */
for (int i=0; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_opt_is_bound(q->bb[block], pbuf[i]->subject) && fs_opt_is_const(q->bb[block], pbuf[i]->predicate) && fs_opt_is_const(q->bb[block], pbuf[i]->object) && !fs_opt_literal_is_rdf_type(pbuf[i]->predicate)) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
/* triples like ?s rdf:type :o */
for (int i=0; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_opt_is_bound(q->bb[block], pbuf[i]->subject) && fs_opt_is_const(q->bb[block], pbuf[i]->predicate) && fs_opt_is_const(q->bb[block], pbuf[i]->object)) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
for (int i=0; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_opt_is_const(q->bb[block], pbuf[i]->subject) && fs_opt_is_bound(q->bb[block], pbuf[i]->object)) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
for (int i=0; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_opt_is_const(q->bb[block], pbuf[i]->object) && fs_opt_is_bound(q->bb[block], pbuf[i]->subject)) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
for (int i=0; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_opt_is_const(q->bb[block], pbuf[i]->predicate) && (fs_opt_is_bound(q->bb[block], pbuf[i]->subject) || fs_opt_is_bound(q->bb[block], pbuf[i]->object))) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
for (int i=0; i<length; i++) {
if (!pbuf[i]) {
continue;
}
if (fs_opt_is_const(q->bb[block], pbuf[i]->origin)) {
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
}
for (int i=0; i<length; i++) {
if (!pbuf[i]) {
continue;
}
patt[append_pos++] = pbuf[i];
pbuf[i] = NULL;
}
free(pbuf);
if (append_pos != length) {
fs_error(LOG_CRIT, "Optimser mismatch error");
}
#ifdef DEBUG_OPTIMISER
printf("optimiser choices look like:\n");
for (int i=start; i<length; i++) {
printf("%4d: ", i);
rasqal_triple_print(patt[i], stdout);
printf("\n");
}
#endif
/* If the next two or more pattern's subjects are both variables, we might be able
* to multi reverse bind them */
if (var_name(patt[start]->subject) && var_name(patt[start+1]->subject) &&
!var_name(patt[start]->predicate) &&
!var_name(patt[start]->object) &&
fs_opt_num_vals(q->bb[block], patt[start]->predicate) == 1 &&
fs_opt_num_vals(q->bb[block], patt[start]->origin) == 0 &&
fs_opt_num_vals(q->bb[block], patt[start+1]->origin) == 0) {
char *svname = var_name(patt[start]->subject);
int count = 1;
while (start+count < length &&
!fs_opt_is_const(q->bb[block], patt[start+count]->subject) &&
!strcmp(svname, var_name(patt[start+count]->subject)) &&
!var_name(patt[start+count]->object) &&
!var_name(patt[start+count]->predicate)) {
count++;
}
/* if we found a reverse bind pair then we may as well use that, rather
* than pressing on and using the freq data to pick an order, the
* backend has more complete information */
if (count > 1) return count;
}
if (length - start > 1) {
int freq_a = fs_bind_freq(qs, q, block, patt[start]);
int freq_b = fs_bind_freq(qs, q, block, patt[start+1]);
/* the 2nd is cheaper than the 1st, then swap them */
if (freq_b < freq_a) {
rasqal_triple *tmp = patt[start];
patt[start] = patt[start+1];
patt[start+1] = tmp;
}
}
return 1;
}