Changeset: fd5a3cab2c5b for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=fd5a3cab2c5b
Modified Files:
monetdb5/modules/kernel/algebra.mal
sql/backends/monet5/rel_bin.c
sql/backends/monet5/sql_gencode.c
sql/backends/monet5/sql_statement.c
sql/backends/monet5/sql_statement.h
sql/server/rel_select.c
sql/test/BugTracker-2010/Tests/ORDER_BY_over_UNION_EXCEPT_INTERSECT.Bug-2606.stable.out
Branch: default
Log Message:
cleanup except/intersect implementations. removed use of
stmt_union/stmt_diff.
diffs (truncated from 708 to 300 lines):
diff --git a/monetdb5/modules/kernel/algebra.mal
b/monetdb5/modules/kernel/algebra.mal
--- a/monetdb5/modules/kernel/algebra.mal
+++ b/monetdb5/modules/kernel/algebra.mal
@@ -72,20 +72,20 @@ comment "Produces a new BAT with groups
#command join( left:bat[:any_1,:any_2], right:bat[:any_2,:any_3],
estimate:lng) :bat[:any_1,:any_3]
#address ALGjoinestimate;
+command join(l:bat[:any_1,:any_2], rl:bat[:any_3,:any_2],
rh:bat[:any_3,:any_2], li:bit, hi:bit) :bat[:any_1,:any_3]
+address ALGrangejoin;
+
command leftjoin( left:bat[:any_1,:any_2], right:bat[:any_2,:any_3])
:bat[:any_1,:any_3]
address ALGleftjoin;
command leftjoin( left:bat[:any_1,:any_2], right:bat[:any_2,:any_3],
estimate:lng) :bat[:any_1,:any_3]
address ALGleftjoinestimate;
+#END OF HEADLESS PLANS
+
command leftfetchjoin ( left:bat[:oid,:oid], right:bat[:oid,:any_3] )
:bat[:oid,:any_3]
address ALGleftfetchjoin
comment "Hook directly into the left fetch join implementation.";
-command join(l:bat[:any_1,:any_2], rl:bat[:any_3,:any_2],
rh:bat[:any_3,:any_2], li:bit, hi:bit) :bat[:any_1,:any_3]
-address ALGrangejoin;
-
-#END OF HEADLESS PLANS
-
command revert( b:bat[:oid,:any_1]) :bat[:oid,:any_1]
address ALGrevert
comment "Returns a BAT copy with buns in reverse order";
diff --git a/sql/backends/monet5/rel_bin.c b/sql/backends/monet5/rel_bin.c
--- a/sql/backends/monet5/rel_bin.c
+++ b/sql/backends/monet5/rel_bin.c
@@ -753,7 +753,6 @@ exp_bin(mvc *sql, sql_exp *e, stmt *left
static stmt *check_types(mvc *sql, sql_subtype *ct, stmt *s, check_type tpe);
-/* TODO pass optional selection */
static stmt *
stmt_col( mvc *sql, sql_column *c, stmt *del)
{
@@ -1470,7 +1469,7 @@ releqjoin( mvc *sql, list *l1, list *l2,
if (cmp_op == cmp_equal)
f = sql_bind_func(sql->sa, sql->session->schema, "=",
tail_type(le), tail_type(le), F_FUNC);
- else /* cmp_equal_nil ? */
+ else /* TODO cmp_equal_nil */
f = sql_bind_func(sql->sa, sql->session->schema, "=",
tail_type(le), tail_type(le), F_FUNC);
assert(f);
@@ -1878,6 +1877,29 @@ rel2bin_distinct(mvc *sql, stmt *s, stmt
}
static stmt *
+rel_rename(mvc *sql, sql_rel *rel, stmt *sub)
+{
+ if (rel->exps) {
+ node *en, *n;
+ list *l = sa_list(sql->sa);
+
+ for( en = rel->exps->h, n = sub->op4.lval->h; en && n; en =
en->next, n = n->next ) {
+ sql_exp *exp = en->data;
+ stmt *s = n->data;
+
+ if (!s) {
+ assert(0);
+ return NULL;
+ }
+ s = stmt_rename(sql, rel, exp, s);
+ list_append(l, s);
+ }
+ sub = stmt_list(sql->sa, l);
+ }
+ return sub;
+}
+
+static stmt *
rel2bin_union( mvc *sql, sql_rel *rel, list *refs)
{
list *l;
@@ -1901,45 +1923,18 @@ rel2bin_union( mvc *sql, sql_rel *rel, l
char *nme = column_name(sql->sa, c1);
stmt *s;
- /* append isn't save, ie use union
- (also not save loses unique head oids)
-
- so we create append on copies.
- TODO: mark columns non base columns, ie were no
- copy is needed
- */
s = stmt_append(sql->sa, Column(sql->sa, c1), c2);
s = stmt_alias(sql->sa, s, rnme, nme);
list_append(l, s);
}
sub = stmt_list(sql->sa, l);
- /* union exp list is a rename only */
- if (rel->exps) {
- node *en, *n;
- list *l = sa_list(sql->sa);
-
- for( en = rel->exps->h, n = sub->op4.lval->h; en && n; en =
en->next, n = n->next ) {
- sql_exp *exp = en->data;
- stmt *s = n->data;
-
- if (!s) {
- assert(0);
- return NULL;
- }
- s = stmt_rename(sql, rel, exp, s);
- list_append(l, s);
- }
- sub = stmt_list(sql->sa, l);
- }
-
+ sub = rel_rename(sql, rel, sub);
if (need_distinct(rel))
sub = rel2bin_distinct(sql, sub, NULL);
return sub;
}
-/* Both EXCEPT and INTERSECT need work, current versions aren't mergetable
save
- * (bails out on the gen_group) */
static stmt *
rel2bin_except( mvc *sql, sql_rel *rel, list *refs)
{
@@ -1947,12 +1942,13 @@ rel2bin_except( mvc *sql, sql_rel *rel,
list *stmts;
node *n, *m;
stmt *left = NULL, *right = NULL, *sub;
+ sql_subfunc *min;
stmt *lg = NULL, *rg = NULL;
stmt *lgrp = NULL, *rgrp = NULL;
- stmt *lext = NULL, *rext = NULL;
- stmt *lcnt = NULL, *rcnt = NULL;
- stmt *s, *lm, *rm, *ecnt = NULL;
+ stmt *lext = NULL, *rext = NULL, *next = NULL;
+ stmt *lcnt = NULL, *rcnt = NULL, *ncnt = NULL, *zero = NULL;
+ stmt *s, *lm, *rm;
list *lje = sa_list(sql->sa);
list *rje = sa_list(sql->sa);
@@ -1963,18 +1959,13 @@ rel2bin_except( mvc *sql, sql_rel *rel,
if (!left || !right)
return NULL;
left = row2cols(sql, left);
-
- /* construct relation */
- stmts = sa_list(sql->sa);
+ right = row2cols(sql, right);
+
/*
- * The multi column intersect is handled using group by's and
+ * The multi column except is handled using group by's and
* group size counts on both sides of the intersect. We then
- * return for each group of A with min(A.count,B.count),
+ * return for each group of L with min(L.count,R.count),
* number of rows.
- *
- * The problem with this approach is that the groups should
- * have equal group identifiers. So we take the union of all
- * columns before the group by.
*/
for (n = left->op4.lval->h; n; n = n->next) {
lg = stmt_group(sql->sa, column(sql->sa, n->data), lgrp, lext,
lcnt);
@@ -1994,10 +1985,12 @@ rel2bin_except( mvc *sql, sql_rel *rel,
stmt_group_done(lg);
stmt_group_done(rg);
+ if (need_distinct(rel)) {
+ lcnt = stmt_const(sql->sa, lcnt, stmt_atom_wrd(sql->sa, 1));
+ rcnt = stmt_const(sql->sa, rcnt, stmt_atom_wrd(sql->sa, 1));
+ }
+
/* now find the matching groups */
-
- /* TODO change to left outer join semantics to keep those in A not in B
*/
- /* would need outerjoin eqjoin and outer project code, cleans up
following mess */
for (n = left->op4.lval->h, m = right->op4.lval->h; n && m; n =
n->next, m = m->next) {
stmt *l = column(sql->sa, n->data);
stmt *r = column(sql->sa, m->data);
@@ -2011,33 +2004,32 @@ rel2bin_except( mvc *sql, sql_rel *rel,
lm = stmt_result(sql->sa, s, 0);
rm = stmt_result(sql->sa, s, 1);
- /* the join of the groups removed those in A but not in B,
- * we need these later so keep these in 'ecnt' */
- ecnt = stmt_diff(sql->sa, lcnt, stmt_reverse(sql->sa, lm));
-
- /*if (!distinct) */
- {
- stmt *glcnt, *grcnt, *o;
- sql_subfunc *sub;
-
- /* nil + count -> ? */
- glcnt = stmt_project(sql->sa, lm, lcnt);
- grcnt = stmt_project(sql->sa, rm, rcnt);
-
- sub = sql_bind_func(sql->sa, sql->session->schema, "sql_sub",
wrd, wrd, F_FUNC);
- s = stmt_binop(sql->sa, glcnt, grcnt, sub); /* use count */
-
- /* now we need to add the groups which weren't in B */
- lcnt = stmt_reorder_project(sql->sa, stmt_reverse(sql->sa, lm),
s);
- s = stmt_union(sql->sa, ecnt, lcnt);
- o = stmt_mark_tail(sql->sa, lext, 0);
- s = stmt_reorder_project(sql->sa, stmt_reverse(sql->sa, o), s);
-
- /* now we have gid,cnt, blowup to full groupsizes */
- s = stmt_gen_group(sql->sa, lext, s);
- }
+ s = stmt_mirror(sql->sa, lext);
+ s = stmt_tdiff(sql->sa, s, lm);
+
+ /* first we find those missing in R */
+ next = stmt_project(sql->sa, s, lext);
+ ncnt = stmt_project(sql->sa, s, lcnt);
+ zero = stmt_const(sql->sa, s, stmt_atom_wrd(sql->sa, 0));
+
+ /* ext, lcount, rcount */
+ lext = stmt_project(sql->sa, lm, lext);
+ lcnt = stmt_project(sql->sa, lm, lcnt);
+ rcnt = stmt_project(sql->sa, rm, rcnt);
+
+ /* append those missing in L */
+ lext = stmt_append(sql->sa, lext, next);
+ lcnt = stmt_append(sql->sa, lcnt, ncnt);
+ rcnt = stmt_append(sql->sa, rcnt, zero);
+
+ min = sql_bind_func(sql->sa, sql->session->schema, "sql_sub", wrd, wrd,
F_FUNC);
+ s = stmt_binop(sql->sa, lcnt, rcnt, min); /* use count */
+
+ /* now we have gid,cnt, blowup to full groupsizes */
+ s = stmt_gen_group(sql->sa, lext, s);
/* project columns of left hand expression */
+ stmts = sa_list(sql->sa);
for (n = left->op4.lval->h; n; n = n->next) {
stmt *c1 = column(sql->sa, n->data);
char *rnme = NULL;
@@ -2051,30 +2043,7 @@ rel2bin_except( mvc *sql, sql_rel *rel,
list_append(stmts, c1);
}
sub = stmt_list(sql->sa, stmts);
-
- /* TODO put in sep function !!!, and add to all is_project(op) */
- /* except can be a projection too */
- if (rel->exps) {
- node *en;
- list *l = sa_list(sql->sa);
-
- for( en = rel->exps->h; en; en = en->next ) {
- sql_exp *exp = en->data;
- stmt *s = exp_bin(sql, exp, sub, NULL, NULL, NULL,
NULL, NULL);
-
- if (!s) {
- assert(0);
- return NULL;
- }
- s = stmt_rename(sql, rel, exp, s);
- list_append(l, s);
- }
- sub = stmt_list(sql->sa, l);
- }
-
- if (need_distinct(rel))
- sub = rel2bin_distinct(sql, sub, NULL);
- return sub;
+ return rel_rename(sql, rel, sub);
}
static stmt *
@@ -2084,6 +2053,7 @@ rel2bin_inter( mvc *sql, sql_rel *rel, l
list *stmts;
node *n, *m;
stmt *left = NULL, *right = NULL, *sub;
+ sql_subfunc *min;
stmt *lg = NULL, *rg = NULL;
stmt *lgrp = NULL, *rgrp = NULL;
@@ -2101,17 +2071,11 @@ rel2bin_inter( mvc *sql, sql_rel *rel, l
return NULL;
left = row2cols(sql, left);
- /* construct relation */
- stmts = sa_list(sql->sa);
/*
* The multi column intersect is handled using group by's and
* group size counts on both sides of the intersect. We then
- * return for each group of A with min(A.count,B.count),
+ * return for each group of L with min(L.count,R.count),
* number of rows.
- *
- * The problem with this approach is that the groups should
- * have equal group identifiers. So we take the union of all
- * columns before the group by.
*/
for (n = left->op4.lval->h; n; n = n->next) {
lg = stmt_group(sql->sa, column(sql->sa, n->data), lgrp, lext,
lcnt);
@@ -2131,6 +2095,11 @@ rel2bin_inter( mvc *sql, sql_rel *rel, l
_______________________________________________
checkin-list mailing list
[email protected]
https://www.monetdb.org/mailman/listinfo/checkin-list
