Changeset: 5e0839367a63 for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=5e0839367a63
Modified Files:
monetdb5/modules/mal/array.mx
sql/backends/monet5/sciql.c
sql/server/rel_schema.c
Branch: SciQL-2
Log Message:
some comments, printouts and renames to help me understand
diffs (truncated from 390 to 300 lines):
diff --git a/monetdb5/modules/mal/array.mx b/monetdb5/modules/mal/array.mx
--- a/monetdb5/modules/mal/array.mx
+++ b/monetdb5/modules/mal/array.mx
@@ -607,7 +607,7 @@ ARRAYfiller(Client cntxt, MalBlkPtr mb,
* - only dimension types SMALLINT (bte), TINYINY (sht), INT(EGER) (int) are
supported
* - dimensions must be ascending, i.e., start <= stop && step > 0
* - only step-size 1 is supported
- * - array must be stored "canonically", i.e., sorted (ascending) on first
dimension,
+ * - array must be stored ", p, (double)bRest[p]);canonically", i.e., sorted
(ascending) on first dimension,
* and each subsequent dimension sub-sorted (ascending) within each value of
its preceding dimension
*/
str
@@ -665,6 +665,8 @@ ARRAYtiles_@4_@1_@8(Client cntxt, MalBlk
arrsze *= dSize[i];
}
+//fprintf(stderr, "arrsze = %u\n", (unsigned int)arrsze);
+
/* check sanity of value BAT */
if (!BAThdense(bVal)) {
AGGR_CLEANUP();
@@ -675,6 +677,8 @@ ARRAYtiles_@4_@1_@8(Client cntxt, MalBlk
throw(MAL, "array.@4", "tail type of value BAT is not type @1");
}
arrcnt = BATcount(bVal);
+fprintf(stderr, "arrcnt = %u\n", (unsigned int)arrcnt);
+
if (arrcnt != arrsze) {
AGGR_CLEANUP();
throw(MAL, "array.@4", "count of value BAT ("BUNFMT") !=
product of dimension sizes ("BUNFMT")",
@@ -726,8 +730,9 @@ ARRAYtiles_@4_@1_@8(Client cntxt, MalBlk
/* access tails as arrays */
bDimsT[i] = (@8*) Tloc(bDims[i], BUNfirst(bDims[i]));
bOffsetsT[i] = (@8*) Tloc(bOffsets[i], BUNfirst(bOffsets[i]));
-
- /* be optimistic */
+for(p=0; p<BATcount(bOffsets[i]); p++)
+ fprintf(stderr, "%u: %d - %d\n", (unsigned int)i, (int)bDimsT[i][p],
(int)bOffsetsT[i][p]);
+ /* be optimistic, i.e. assume that BATs are ordered and
subordered (e.g. y subordered on x etc.) */
dMin[i] = bDimsT[i][0];
dMax[i] = bDimsT[i][arrcnt-1];
@@ -802,6 +807,7 @@ ARRAYtiles_@4_@1_@8(Client cntxt, MalBlk
offset[r] += mul * bOffsetsT[i][r];
mul *= dSize[i];
}
+fprintf(stderr, "offset: %u => %f\n", (unsigned int)r, (double)offset[r]);
}
/* For each anchor piont, compute all cells belong to this tile
(bVal.head
@@ -838,14 +844,15 @@ ARRAYtiles_@4_@1_@8(Client cntxt, MalBlk
}
if (!skip) {
@1 elm = bValT[vid];
-
if (elm != @1_nil) {
- @6;
+ @6; //@6 := sum += elm (if sum)
cnt++;
}
+fprintf(stderr, "elm: (p=%u, r=%u) => %d -> %f\n", (unsigned int)p, (unsigned
int)r, (int)elm, (double)@7);
}
}
- bResT[p] = (cnt ? @7 : @3_nil);
+ bResT[p] = (cnt ? @7 : @3_nil); //@7 := sum (if sum)
+fprintf(stderr, "\t%u: %f\n", (unsigned int)p, (double)bResT[p]);
nils |= !cnt;
}
diff --git a/sql/backends/monet5/sciql.c b/sql/backends/monet5/sciql.c
--- a/sql/backends/monet5/sciql.c
+++ b/sql/backends/monet5/sciql.c
@@ -23,8 +23,8 @@
#define MTRL_CLEANUP() \
{ \
- if (N) GDKfree(N); \
- if (M) GDKfree(M); \
+ if (idxRepetitionsPerGroup) GDKfree(idxRepetitionsPerGroup); \
+ if (groups) GDKfree(groups); \
if (bids) { \
for (i = 0; i < list_length(a->columns.set); i++) { \
if (bids[i] != 0) \
@@ -58,16 +58,18 @@ static str
return createException(type, fcn, "%s", msg);
}
-str
-SCIQLmaterialise(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
-{
+/* It is called in each update query involving an array. If the array is
materialised nothing happens.
+ * Otherwise, it creates all necessary BATs initialising them with the
appropriate values (the
+ * non-dimensional BAT is initialised with the default value). The query is
handled then as any other
+ * query */
+str SCIQLmaterialise(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci) {
mvc *sql = NULL;
str msg = getSQLContext(cntxt, mb, &sql, NULL);
- str sname = *(str*) getArgReference(stk, pci, 1);
- str aname = *(str*) getArgReference(stk, pci, 2);
+ str schema_name = *(str*) getArgReference(stk, pci, 1);
+ str array_name = *(str*) getArgReference(stk, pci, 2);
sql_schema *s = NULL;
sql_table *a = NULL;
- int i = 0, j = 0, *N = NULL, *M = NULL, *bids = NULL;
+ int i = 0, j = 0, *idxRepetitionsPerGroup = NULL, *groups = NULL, *bids
= NULL;
BUN cntall = 1; /* cntall must be initialised with 1! */
node *n = NULL;
@@ -82,50 +84,51 @@ SCIQLmaterialise(Client cntxt, MalBlkPtr
if (msg)
return msg;
- if (!(s = mvc_bind_schema(sql, sname)))
+ if (!(s = mvc_bind_schema(sql, schema_name)))
throw(MAL, "sciql.materialise", RUNTIME_OBJECT_MISSING);
- if (!(a = mvc_bind_table(sql, s, aname)))
+ if (!(a = mvc_bind_table(sql, s, array_name)))
throw(MAL, "sciql.materialise", RUNTIME_OBJECT_MISSING);
- if (a->materialised)
+ if (a->materialised) //If materialised then it is treated than any
other BAT (should this change? Does it make sense to have partially initialised
BATs?)
return MAL_SUCCEED;
- /* To compute N (the #times each value is repeated), multiply the size
of
- * dimensions defined after the current dimension. For the last
dimension,
- * its N is 1. To compute M (the #times each value group is repeated),
- * multiply the size of dimensions defined before the current dimension.
- * For the first dimension, its M is 1. */
- N = GDKmalloc(a->valence * sizeof(int)); /* #repeats of each value */
- M = GDKmalloc(a->valence * sizeof(int)); /* #repeats of each group of
values */
+ /* When having a single dimension, e.g. x then each value corresponds
to a different x_idx, x_idx=0, x_idx=1, etc.
+ * When having more dimensions, e.g. x and y then each value corresponds
to a different pair of idxs,
+ * e.g. (x_idx=0, y_idx=0), (x_idx=0, y_idx=1), (x_idx=0, y_idx=2) etc.
From the example is obvious that until all
+ * possible values of y_idx has been used the x_idx will be 0, when all
y_idx values has been used x_idx will increase
+ * and the y_idx values will start from the beggining. So, the
idxRepettitionsPerGroup for y is 1 (the y_idx is changed
+ * in each repetition) while the groups groups for y equals the number
of x_idx values (all y_idx values are repeated for
+ * each value of x_idx) */
+ idxRepetitionsPerGroup = GDKmalloc(a->valence * sizeof(int)); /*
#repeats of each value */
+ groups = GDKmalloc(a->valence * sizeof(int)); /* #repeats of each group
of values */
bids = GDKzalloc(list_length(a->columns.set) * sizeof(int)); /* BAT ids
for each column */
- if (!N || !M || !bids) {
+ if (!idxRepetitionsPerGroup || !groups || !bids) {
MTRL_CLEANUP();
throw(MAL, "sciql.materialise", MAL_MALLOC_FAIL);
}
for (i = 0; i < a->valence; i++)
- N[i] = M[i] = 1;
+ idxRepetitionsPerGroup[i] = groups[i] = 1;
for (n = a->columns.set->h, i = 0; n; n = n->next) {
sql_column *sc = n->data;
if (sc->dim) {
/* TODO: overflow check, see gdk_calc.c */
- lng cnt_l = sc->dim->step > 0 ? (sc->dim->stop -
sc->dim->strt + sc->dim->step - 1) / sc->dim->step :
- (sc->dim->strt - sc->dim->stop - sc->dim->step
- 1) / -sc->dim->step;
- int cnt_i = (int) cnt_l;
+ lng dimensionElementsNum_lng = 1+(sc->dim->step > 0 ?
(sc->dim->stop-1 - sc->dim->strt) / sc->dim->step : (sc->dim->strt - 1 -
sc->dim->stop) / -sc->dim->step);
+ int dimensionElementsNum_int = (int)
dimensionElementsNum_lng;
#ifndef NDEBUG
- int lim = GDK_int_max / cnt_i;
+ int lim = GDK_int_max / dimensionElementsNum_int;
#endif
- assert(cnt_l <= (lng) GDK_int_max);
+ assert(dimensionElementsNum_lng <= (lng) GDK_int_max);
for (j = 0; j < i; j++) {
- assert(N[j] <= lim);
- N[j] *= cnt_i;
+ assert(idxRepetitionsPerGroup[j] <= lim);
+ idxRepetitionsPerGroup[j] *=
dimensionElementsNum_int;
}
for (j = a->valence - 1; j > i; j--) {
- assert(M[j] <= lim);
- M[j] *= cnt_i;
+ assert(groups[j] <= lim);
+ groups[j] *= dimensionElementsNum_int;
}
- assert((BUN) cnt_i <= BUN_MAX / cntall);
- cntall *= cnt_i;
+ assert((BUN) dimensionElementsNum_int <= BUN_MAX /
cntall);
+ cntall *= dimensionElementsNum_int;
i++;
}
}
@@ -134,27 +137,27 @@ SCIQLmaterialise(Client cntxt, MalBlkPtr
for (n = a->columns.set->h, i = 0; n; n = n->next, i++){
sql_column *sc = n->data;
BAT *bn = NULL;
- int tpe = sc->type.type->localtype;
+ int tpe = sc->type.type->localtype; //the type of the column
if (sc->dim) {
switch (tpe) {
case TYPE_bte:
- ARRAYseries_bte(&bids[i], (bte
*)&sc->dim->strt, (bte *)&sc->dim->step, (bte *)&sc->dim->stop, &N[i], &M[i]);
+ ARRAYseries_bte(&bids[i], (bte
*)&sc->dim->strt, (bte *)&sc->dim->step, (bte *)&sc->dim->stop,
&idxRepetitionsPerGroup[i], &groups[i]);
break;
case TYPE_sht:
- ARRAYseries_sht(&bids[i], (sht
*)&sc->dim->strt, (sht *)&sc->dim->step, (sht *)&sc->dim->stop, &N[i], &M[i]);
+ ARRAYseries_sht(&bids[i], (sht
*)&sc->dim->strt, (sht *)&sc->dim->step, (sht *)&sc->dim->stop,
&idxRepetitionsPerGroup[i], &groups[i]);
break;
case TYPE_int:
- ARRAYseries_int(&bids[i], (int
*)&sc->dim->strt, (int *)&sc->dim->step, (int *)&sc->dim->stop, &N[i], &M[i]);
+ ARRAYseries_int(&bids[i], (int
*)&sc->dim->strt, (int *)&sc->dim->step, (int *)&sc->dim->stop,
&idxRepetitionsPerGroup[i], &groups[i]);
break;
case TYPE_lng:
- ARRAYseries_lng(&bids[i], (lng
*)&sc->dim->strt, (lng *)&sc->dim->step, (lng *)&sc->dim->stop, &N[i], &M[i]);
+ ARRAYseries_lng(&bids[i], (lng
*)&sc->dim->strt, (lng *)&sc->dim->step, (lng *)&sc->dim->stop,
&idxRepetitionsPerGroup[i], &groups[i]);
break;
case TYPE_flt:
- ARRAYseries_flt(&bids[i], (flt
*)&sc->dim->strt, (flt *)&sc->dim->step, (flt *)&sc->dim->stop, &N[i], &M[i]);
+ ARRAYseries_flt(&bids[i], (flt
*)&sc->dim->strt, (flt *)&sc->dim->step, (flt *)&sc->dim->stop,
&idxRepetitionsPerGroup[i], &groups[i]);
break;
case TYPE_dbl:
- ARRAYseries_dbl(&bids[i], (dbl
*)&sc->dim->strt, (dbl *)&sc->dim->step, (dbl *)&sc->dim->stop, &N[i], &M[i]);
+ ARRAYseries_dbl(&bids[i], (dbl
*)&sc->dim->strt, (dbl *)&sc->dim->step, (dbl *)&sc->dim->stop,
&idxRepetitionsPerGroup[i], &groups[i]);
break;
default:
MTRL_CLEANUP();
@@ -165,12 +168,13 @@ SCIQLmaterialise(Client cntxt, MalBlkPtr
MTRL_CLEANUP();
throw(MAL, "sciql.materialise", "Cannot access
descriptor");
}
- } else {
+ } else { //non-dimensional column
ValRecord src, dst;
int ret = 0;
src.vtype = TYPE_str;
- if (sc->def) {
+ if (sc->def) { //the column has a default value
+ //no mater what the default value is all values
set will have the same value
size_t l = strlen(sc->def);
if (l == 4 && (sc->def[0] == 'n' || sc->def[0]
== 'N') &&
(sc->def[1] == 'u' ||
sc->def[1] == 'U') &&
@@ -197,6 +201,7 @@ SCIQLmaterialise(Client cntxt, MalBlkPtr
return _rethrow(MAL, "sciql.materialise",
"string conversion failed");
}
+ /* Fill the BAT with the constant value */
switch(tpe) { /* TODO: check for overflow */
case TYPE_bte:
bn = BATconstant(tpe, &dst.val.btval, cntall);
@@ -239,11 +244,11 @@ SCIQLmaterialise(Client cntxt, MalBlkPtr
systable = find_sql_table(syss, "_tables");
sysarray = find_sql_table(syss, "_arrays");
/* find 'id' of this array in _tables */
- rid = table_funcs.column_find_row(tr, find_sql_column(systable,
"name"), aname, NULL);
- tid = *(sqlid*) table_funcs.column_find_value(tr,
find_sql_column(systable, "id"), rid);
+ rid = table_funcs.column_find_row(tr, find_sql_column(systable,
"name"), array_name, NULL); //find the row in _tables that is about the
processed array
+ tid = *(sqlid*) table_funcs.column_find_value(tr,
find_sql_column(systable, "id"), rid); //find the column in the above row that
is about id
/* update value in _arrays */
- rid = table_funcs.column_find_row(tr, find_sql_column(sysarray,
"table_id"), &tid, NULL);
- table_funcs.column_update_value(tr, find_sql_column(sysarray,
"materialised"), rid, &materialised);
+ rid = table_funcs.column_find_row(tr, find_sql_column(sysarray,
"table_id"), &tid, NULL); //find the row in arrays that is about the processed
array
+ table_funcs.column_update_value(tr, find_sql_column(sysarray,
"materialised"), rid, &materialised); //set the value in the column about
materialised
a->materialised = 1;
return MAL_SUCCEED;
diff --git a/sql/server/rel_schema.c b/sql/server/rel_schema.c
--- a/sql/server/rel_schema.c
+++ b/sql/server/rel_schema.c
@@ -891,74 +891,76 @@ rel_create_table(mvc *sql, sql_schema *s
mvc_create_table(sql, s, name, tt, 0,
SQL_DECLARED_TABLE, commit_action, -1);
dnode *n;
dlist *columns = table_elements_or_subquery->data.lval;
- list *drngs = sa_list(sql->sa);
+ list *dimensionsRanges = sa_list(sql->sa);
- for (n = columns->h; n; n = n->next) {
- symbol *sym = n->data.sym;
+ for (n = columns->h; n; n = n->next) { //for each column
+ symbol *sym = n->data.sym; //each column is represented
with a symbol
int res = table_element(sql, sym, s, t, 0);
if (res == SQL_ERR)
return NULL;
/* A dimension column? Add the range expressions to
rel_table */
- if (sym->token == SQL_COLUMN && sym->data.lval->cnt ==
4) {
- dnode *dn = sym->data.lval->h->next->next->next;
- sql_column *dc = t->columns.set->t->data;
- char *sqltpe = dc->type.type->sqlname;
+ if (sym->token == SQL_COLUMN && sym->data.lval->cnt ==
4 && sym->data.lval->h->next->next->next->data.sym->token == SQL_DIMENSION) {
+ dnode *dimensionNode =
sym->data.lval->h->next->next->next;
+ sql_column *dimensionColumn =
t->columns.set->t->data;
+ char *sqltpe =
dimensionColumn->type.type->sqlname;
int is_int_dc = isAnSQLIntType(sqltpe,
strlen(sqltpe));
- dlist *drng = NULL;
+ dlist *dimensionRange =
dimensionNode->data.sym->data.lval;
sql_exp *val_exp = NULL;
exp_kind ek = {type_value, card_value, TRUE};
sql_subtype *lngtpe = sql_bind_localtype("lng");
if (!isArray(t))
- return sql_error(sql, 02, "%s %s:
dimensions ('%s') not allowed in non-ARRAY\n", action, tt2string(tt),
dc->base.name);
+ return sql_error(sql, 02, "%s %s:
dimensions ('%s') not allowed in non-ARRAY\n", action, tt2string(tt),
dimensionColumn->base.name);
if (!isSupportedType(sqltpe))
return sql_error(sql, 02, "%s ARRAY:
dimension type '%s' not supported yet\n", action, sqltpe);
_______________________________________________
checkin-list mailing list
[email protected]
https://www.monetdb.org/mailman/listinfo/checkin-list
