/*
* The pack is an ordinary multi BAT insert. Oid synchronistion
* between pieces should be ensured by the code generators.
* The pack operation could be quite expensive, because it
* may create a really large BAT.
* The slice over a mat helps to avoid constructing intermediates
* that are subsequently reduced.
* Contrary to most operations, NIL arguments are skipped and
* do not produce RUNTIME_OBJECT_MISSING.
*/
static str
MATpackInternal(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr p)
{
int i, *ret = (int*) getArgReference(stk,p,0);
BAT *b, *bn;
BUN cap = 0;
int tt = TYPE_any;
(void) cntxt;
(void) mb;
for (i = 1; i < p->argc; i++) {
int bid = stk->stk[getArg(p,i)].val.ival;
b = BBPquickdesc(abs(bid),FALSE);
if (b && bid < 0)
b = BATmirror(b);
if( b ){
assert(BAThdense(b));
if (tt == TYPE_any){
tt = b->ttype;
}
if (!tt && tt != b->ttype)
tt = b->ttype;
cap += BATcount(b);
}
}
if (tt == TYPE_any){
*ret = 0;
return MAL_SUCCEED;
}
bn = BATnew(TYPE_void, tt, cap, TRANSIENT);
if (bn == NULL)
throw(MAL, "mat.pack", MAL_MALLOC_FAIL);
for (i = 1; i < p->argc; i++) {
b = BATdescriptor(stk->stk[getArg(p,i)].val.ival);
if( b ){
if (BATcount(bn) == 0)
BATseqbase(bn, b->H->seq);
if (BATcount(bn) == 0)
BATseqbase(BATmirror(bn), b->T->seq);
BATappend(bn,b,FALSE);
BBPunfix(b->batCacheid);
}
}
assert(!bn->H->nil || !bn->H->nonil);
assert(!bn->T->nil || !bn->T->nonil);
BATsettrivprop(bn);
BATderiveProps(bn,FALSE);
BBPkeepref(*ret = bn->batCacheid);
return MAL_SUCCEED;
}
gdk_return
BATmaterializeh(BAT *b)
{
int ht;
BUN cnt;
Heap head;
BUN p, q;
oid h, *x;
bte tshift;
BATcheck(b, "BATmaterialize", GDK_FAIL);
assert(!isVIEW(b));
ht = b->htype;
cnt = BATcapacity(b);
head = b->H->heap;
p = BUNfirst(b);
q = BUNlast(b);
assert(cnt >= q - p);
ALGODEBUG fprintf(stderr, "#BATmaterialize(%d);\n", (int) b->batCacheid);
if (!BAThdense(b) || ht != TYPE_void) {
/* no voids */
return GDK_SUCCEED;
}
ht = TYPE_oid;
/* cleanup possible ACC's */
HASHdestroy(b);
IMPSdestroy(b);
b->H->heap.filename = NULL;
if (HEAPalloc(&b->H->heap, cnt, sizeof(oid)) != GDK_SUCCEED) {
b->H->heap = head;
return GDK_FAIL;
}
/* point of no return */
b->htype = ht;
tshift = b->T->shift;
BATsetdims(b);
if (b->ttype) {
b->T->shift = tshift; /* restore in case it got changed */
b->T->width = 1 << tshift;
}
b->batDirty = TRUE;
b->batDirtydesc = TRUE;
b->H->heap.dirty = TRUE;
/* set the correct dense info */
b->hdense = TRUE;
/* So now generate [h..h+cnt-1] */
h = b->hseqbase;
x = (oid *) b->H->heap.base;
for (; p < q; p++)
*x++ = h++;
cnt = h - b->hseqbase;
BATsetcount(b, cnt);
/* cleanup the old heaps */
HEAPfree(&head, 0);
return GDK_SUCCEED;
}
/* BATsample implements sampling for void headed BATs */
BAT *
BATsample(BAT *b, BUN n)
{
BAT *bn;
BUN cnt, slen;
BUN rescnt;
struct oidtreenode *tree = NULL;
BATcheck(b, "BATsample", NULL);
assert(BAThdense(b));
ERRORcheck(n > BUN_MAX, "BATsample: sample size larger than BUN_MAX\n", NULL);
ALGODEBUG
fprintf(stderr, "#BATsample: sample " BUNFMT " elements.\n", n);
cnt = BATcount(b);
/* empty sample size */
if (n == 0) {
bn = BATnew(TYPE_void, TYPE_void, 0, TRANSIENT);
if (bn == NULL) {
GDKerror("BATsample: memory allocation error");
return NULL;
}
BATsetcount(bn, 0);
BATseqbase(bn, 0);
BATseqbase(BATmirror(bn), 0);
/* sample size is larger than the input BAT, return all oids */
} else if (cnt <= n) {
bn = BATnew(TYPE_void, TYPE_void, cnt, TRANSIENT);
if (bn == NULL) {
GDKerror("BATsample: memory allocation error");
return NULL;
}
BATsetcount(bn, cnt);
BATseqbase(bn, 0);
BATseqbase(BATmirror(bn), b->H->seq);
} else {
oid minoid = b->hseqbase;
oid maxoid = b->hseqbase + cnt;
/* if someone samples more than half of our tree, we
* do the antiset */
bit antiset = n > cnt / 2;
slen = n;
if (antiset)
n = cnt - n;
tree = GDKmalloc(n * sizeof(struct oidtreenode));
if (tree == NULL) {
GDKerror("#BATsample: memory allocation error");
return NULL;
}
bn = BATnew(TYPE_void, TYPE_oid, slen, TRANSIENT);
if (bn == NULL) {
GDKfree(tree);
GDKerror("#BATsample: memory allocation error");
return NULL;
}
/* while we do not have enough sample OIDs yet */
for (rescnt = 0; rescnt < n; rescnt++) {
oid candoid;
do {
/* generate a new random OID */
candoid = (oid) (minoid + DRAND * (maxoid - minoid));
/* if that candidate OID was already
* generated, try again */
} while (!OIDTreeMaybeInsert(tree, candoid, rescnt));
}
if (!antiset) {
OIDTreeToBAT(tree, bn);
} else {
OIDTreeToBATAntiset(tree, bn, minoid, maxoid);
}
GDKfree(tree);
BATsetcount(bn, slen);
bn->trevsorted = bn->batCount <= 1;
bn->tsorted = 1;
bn->tkey = 1;
bn->tdense = bn->batCount <= 1;
if (bn->batCount == 1)
bn->tseqbase = *(oid *) Tloc(bn, BUNfirst(bn));
bn->hdense = 1;
bn->hseqbase = 0;
bn->hkey = 1;
bn->hrevsorted = bn->batCount <= 1;
bn->hsorted = 1;
}
return bn;
}
static BUN
ALGjoinCost(Client cntxt, BAT *l, BAT *r, int flag)
{
BUN lc, rc;
BUN cost=0;
#if 0
BUN lsize,rsize;
BAT *lsample, *rsample, *j;
#endif
(void) flag;
(void) cntxt;
lc = BATcount(l);
rc = BATcount(r);
#if 0
/* The sampling method */
if(flag < 2 && ( lc > 100000 || rc > 100000)){
lsize= MIN(lc/100, (1<<SAMPLE_THRESHOLD_lOG)/3);
lsample= BATsample(l,lsize);
BBPreclaim(lsample);
rsize= MIN(rc/100, (1<<SAMPLE_THRESHOLD_lOG)/3);
rsample= BATsample(r,rsize);
BBPreclaim(rsample);
j= BATjoin(l,r, MAX(lsize,rsize));
lsize= BATcount(j);
BBPreclaim(j);
return lsize;
}
#endif
/* first use logical properties to estimate upper bound of result size */
if (l->tkey && r->hkey)
cost = MIN(lc,rc);
else
if (l->tkey)
cost = rc;
else
if (r->hkey)
cost = lc;
else
if (lc * rc >= BUN_MAX)
cost = BUN_MAX;
else
cost = lc * rc;
/* then use physical properties to rank costs */
if (BATtdense(l) && BAThdense(r))
/* densefetchjoin -> sequential access */
cost /= 7;
else
if (BATtordered(l) && BAThdense(r))
/* orderedfetchjoin > sequential access */
cost /= 6;
else
if (BATtdense(l) && BAThordered(r) && flag != 0 /* no leftjoin */)
/* (reversed-) orderedfetchjoin -> sequential access */
cost /= 6;
else
if (BAThdense(r) && rc <= SMALL_OPERAND)
/* fetchjoin with random access in L1 */
cost /= 5;
else
if (BATtdense(l) && lc <= SMALL_OPERAND && flag != 0 /* no leftjoin */)
/* (reversed-) fetchjoin with random access in L1 */
cost /= 5;
else
if (BATtordered(l) && BAThordered(r))
/* mergejoin > sequential access */
cost /= 4;
else
if (BAThordered(r) && rc <= SMALL_OPERAND)
/* binary-lookup-join with random access in L1 */
cost /= 3;
else
if (BATtordered(l) && lc <= SMALL_OPERAND && flag != 0 /* no leftjoin */)
/* (reversed-) binary-lookup-join with random access in L1 */
cost /= 3;
else
if ((BAThordered(r) && lc <= SMALL_OPERAND) || (BATtordered(l) && rc <= SMALL_OPERAND))
/* sortmergejoin with sorting in L1 */
cost /= 3;
else
if (rc <= SMALL_OPERAND)
/* hashjoin with hashtable in L1 */
cost /= 3;
else
if (lc <= SMALL_OPERAND && flag != 0 /* no leftjoin */)
/* (reversed-) hashjoin with hashtable in L1 */
cost /= 3;
else
if (BAThdense(r))
/* fetchjoin with random access beyond L1 */
cost /= 2;
else
if (BATtdense(l) && flag != 0 /* no leftjoin */)
/* (reversed-) fetchjoin with random access beyond L1 */
cost /= 2;
else
/* hashjoin with hashtable larger than L1 */
/* sortmergejoin with sorting beyond L1 */
cost /= 1;
ALGODEBUG
fprintf(stderr,"#batjoin cost ?"BUNFMT"\n",cost);
return cost;
}
static str
AGGRsubgroupedExt(bat *retval1, bat *retval2, bat *bid, bat *gid, bat *eid, bat *sid,
int skip_nils, int abort_on_error, int tp,
BAT *(*grpfunc1)(BAT *, BAT *, BAT *, BAT *, int, int, int),
gdk_return (*grpfunc2)(BAT **, BAT **, BAT *, BAT *, BAT *, BAT *, int, int, int),
BAT *(*quantilefunc)(BAT *, BAT *, BAT *, BAT *, int, double, int, int),
bat *quantile,
const char *malfunc)
{
BAT *b, *g, *e, *s, *bn = NULL, *cnts, *q = NULL;
double qvalue;
/* one of grpfunc1, grpfunc2 and quantilefunc is non-NULL and the others are */
assert((grpfunc1 && grpfunc2 == NULL && quantilefunc == NULL) ||
(grpfunc1 == NULL && grpfunc2 && quantilefunc == NULL) ||
(grpfunc1 == NULL && grpfunc2 == NULL && quantilefunc) );
/* if retval2 is non-NULL, we must have grpfunc2 */
assert(retval2 == NULL || grpfunc2 != NULL);
b = BATdescriptor(*bid);
g = gid ? BATdescriptor(*gid) : NULL;
e = eid ? BATdescriptor(*eid) : NULL;
q = quantile ? BATdescriptor(*quantile) : NULL;
if (b == NULL || (gid != NULL && g == NULL) || (eid != NULL && e == NULL)) {
if (b)
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
throw(MAL, malfunc, RUNTIME_OBJECT_MISSING);
}
if (tp == TYPE_any && (grpfunc1 == BATgroupmedian || quantilefunc == BATgroupquantile))
tp = b->ttype;
if (sid) {
s = BATdescriptor(*sid);
if (s == NULL) {
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
throw(MAL, malfunc, RUNTIME_OBJECT_MISSING);
}
} else {
if (!BAThdense(b)) {
/* XXX backward compatibility code: ignore non-dense head, but
* only if no candidate list */
s = BATmirror(BATmark(BATmirror(b), 0));
BBPreleaseref(b->batCacheid);
b = s;
}
s = NULL;
}
if (grpfunc1)
bn = (*grpfunc1)(b, g, e, s, tp, skip_nils, abort_on_error);
if (quantilefunc) {
assert(BATcount(q)>0);
assert(q->ttype == TYPE_dbl);
qvalue = ((const double *)Tloc(q, BUNfirst(q)))[0];
if (qvalue < 0|| qvalue > 1) {
char *s;
s = createException(MAL, malfunc, "quantile value of %f is not in range [0,1]", qvalue);
return s;
}
bn = (*quantilefunc)(b, g, e, s, tp, qvalue, skip_nils, abort_on_error);
}
if (grpfunc2 && (*grpfunc2)(&bn, retval2 ? &cnts : NULL, b, g, e, s, tp, skip_nils, abort_on_error) == GDK_FAIL)
bn = NULL;
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
if (s)
BBPreleaseref(s->batCacheid);
if (bn == NULL) {
char *errbuf = GDKerrbuf;
char *s;
if (errbuf && *errbuf) {
if (strncmp(errbuf, "!ERROR: ", 8) == 0)
errbuf += 8;
if (strchr(errbuf, '!') == errbuf + 5) {
s = createException(MAL, malfunc, "%s", errbuf);
} else if ((s = strchr(errbuf, ':')) != NULL && s[1] == ' ') {
s = createException(MAL, malfunc, "%s", s + 2);
} else {
s = createException(MAL, malfunc, "%s", errbuf);
}
*GDKerrbuf = 0;
return s;
}
throw(MAL, malfunc, OPERATION_FAILED);
}
*retval1 = bn->batCacheid;
BBPkeepref(bn->batCacheid);
if (retval2) {
*retval2 = cnts->batCacheid;
BBPkeepref(cnts->batCacheid);
}
return MAL_SUCCEED;
}
/*
* grouped aggregates
*/
static str
AGGRgrouped(bat *retval1, bat *retval2, BAT *b, BAT *g, BAT *e, int tp,
BAT *(*grpfunc1)(BAT *, BAT *, BAT *, BAT *, int, int, int),
gdk_return (*grpfunc2)(BAT **, BAT **, BAT *, BAT *, BAT *, BAT *, int, int, int),
BAT *(*quantilefunc)(BAT *, BAT *, BAT *, BAT *, int, double, int, int),
BAT *quantile,
int skip_nils,
const char *malfunc)
{
BAT *bn, *cnts = NULL, *t, *map;
double qvalue;
/* one of grpfunc1, grpfunc2 and quantilefunc is non-NULL and the others are */
assert((grpfunc1 != NULL && grpfunc2 == NULL && quantilefunc == NULL) ||
(grpfunc1 == NULL && grpfunc2 != NULL && quantilefunc == NULL) ||
(grpfunc1 == NULL && grpfunc2 == NULL && quantilefunc != NULL) );
/* if retval2 is non-NULL, we must have grpfunc2 */
assert(retval2 == NULL || grpfunc2 != NULL);
assert(quantile == NULL || quantilefunc != NULL);
if (b == NULL || g == NULL || e == NULL) {
if (b)
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
throw(MAL, malfunc, RUNTIME_OBJECT_MISSING);
}
if (tp == TYPE_any && (grpfunc1 == BATgroupmedian || quantilefunc == BATgroupquantile))
tp = b->ttype;
if (!BAThdense(b) || !BAThdense(g)) {
/* if b or g don't have a dense head, replace the head with a
* dense sequence */
t = BATjoin(BATmirror(b), g, MIN(BATcount(b), BATcount(g)));
BBPreleaseref(b->batCacheid);
BBPreleaseref(g->batCacheid);
b = BATmirror(BATmark(t, 0));
g = BATmirror(BATmark(BATmirror(t), 0));
BBPreleaseref(t->batCacheid);
}
if (b->hseqbase != g->hseqbase || BATcount(b) != BATcount(g)) {
/* b and g are not aligned: align them by creating a view on
* one or the other */
oid min; /* lowest common oid */
oid max; /* highest common oid */
min = b->hseqbase;
if (min < g->hseqbase)
min = g->hseqbase;
max = b->hseqbase + BATcount(b);
if (g->hseqbase + BATcount(g) < max)
max = g->hseqbase + BATcount(g);
if (b->hseqbase != min || b->hseqbase + BATcount(b) != max) {
if (min >= max)
min = max = b->hseqbase;
t = BATslice(b, BUNfirst(b) + (BUN) (min - b->hseqbase),
BUNfirst(b) + (BUN) (max - b->hseqbase));
BBPreleaseref(b->batCacheid);
b = t;
}
if (g->hseqbase != min || g->hseqbase + BATcount(g) != max) {
if (min >= max)
min = max = g->hseqbase;
t = BATslice(g, BUNfirst(g) + (BUN) (min - g->hseqbase),
BUNfirst(g) + (BUN) (max - g->hseqbase));
BBPreleaseref(g->batCacheid);
g = t;
}
}
if (!BAThdense(e)) {
/* if e doesn't have a dense head, renumber the group ids with
* a dense sequence at the cost of some left joins */
map = BATmark(e, 0); /* [gid,newgid(dense)] */
BBPreleaseref(e->batCacheid);
e = BATmirror(map); /* [newgid(dense),gid] */
t = BATleftjoin(g, map, BATcount(g)); /* [oid,newgid] */
BBPreleaseref(g->batCacheid);
g = t;
} else {
map = NULL;
}
if (grpfunc1)
bn = (*grpfunc1)(b, g, e, NULL, tp, skip_nils, 1);
if (quantilefunc) {
assert(BATcount(quantile)>0);
assert(quantile->ttype == TYPE_dbl);
qvalue = ((const double *)Tloc(quantile, BUNfirst(quantile)))[0];
if (qvalue < 0|| qvalue > 1) {
char *s;
s = createException(MAL, malfunc, "quantile value of %f is not in range [0,1]", qvalue);
return s;
}
bn = (*quantilefunc)(b, g, e, NULL, tp, qvalue, skip_nils, 1);
}
if (grpfunc2 && (*grpfunc2)(&bn, retval2 ? &cnts : NULL, b, g, e, NULL, tp, skip_nils, 1) == GDK_FAIL)
bn = NULL;
if (bn != NULL && (grpfunc1 == BATgroupmin || grpfunc1 == BATgroupmax)) {
t = BATproject(bn, b);
BBPreleaseref(bn->batCacheid);
bn = t;
}
BBPreleaseref(b->batCacheid);
BBPreleaseref(g->batCacheid);
if (map == NULL) /* if map!=NULL, e is mirror of map */
BBPreleaseref(e->batCacheid);
if (bn == NULL) {
char *errbuf = GDKerrbuf;
char *s;
if (map)
BBPreleaseref(map->batCacheid);
if (errbuf && *errbuf) {
if (strncmp(errbuf, "!ERROR: ", 8) == 0)
errbuf += 8;
if (strchr(errbuf, '!') == errbuf + 5) {
s = createException(MAL, malfunc, "%s", errbuf);
} else if ((s = strchr(errbuf, ':')) != NULL && s[1] == ' ') {
s = createException(MAL, malfunc, "%s", s + 2);
} else {
s = createException(MAL, malfunc, "%s", errbuf);
}
*GDKerrbuf = 0;
return s;
}
throw(MAL, malfunc, OPERATION_FAILED);
}
if (map) {
t = BATleftjoin(map, bn, BATcount(bn));
BBPreleaseref(bn->batCacheid);
bn = t;
if (cnts) {
t = BATleftjoin(map, cnts, BATcount(cnts));
BBPreleaseref(cnts->batCacheid);
cnts = t;
}
BBPreleaseref(map->batCacheid);
}
*retval1 = bn->batCacheid;
BBPkeepref(bn->batCacheid);
if (retval2) {
*retval2 = cnts->batCacheid;
BBPkeepref(cnts->batCacheid);
}
return MAL_SUCCEED;
}
static str
AGGRsubgrouped(bat *retval, bat *bid, bat *gid, bat *eid, bat *sid,
int skip_nils, int abort_on_error, int tp,
BAT *(*grpfunc)(BAT *, BAT *, BAT *, BAT *, int, int, int),
const char *malfunc)
{
BAT *b, *g, *e, *s, *bn;
b = BATdescriptor(*bid);
g = gid ? BATdescriptor(*gid) : NULL;
e = eid ? BATdescriptor(*eid) : NULL;
if (b == NULL || (gid != NULL && g == NULL) || (eid != NULL && e == NULL)) {
if (b)
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
throw(MAL, malfunc, RUNTIME_OBJECT_MISSING);
}
if (tp == TYPE_any && grpfunc == BATgroupmedian)
tp = b->ttype;
if (sid) {
s = BATdescriptor(*sid);
if (s == NULL) {
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
throw(MAL, malfunc, RUNTIME_OBJECT_MISSING);
}
} else {
if (!BAThdense(b)) {
/* XXX backward compatibility code: ignore non-dense head, but
* only if no candidate list */
s = BATmirror(BATmark(BATmirror(b), 0));
BBPreleaseref(b->batCacheid);
b = s;
}
s = NULL;
}
bn = (*grpfunc)(b, g, e, s, tp, skip_nils, abort_on_error);
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
if (s)
BBPreleaseref(s->batCacheid);
if (bn == NULL) {
char *errbuf = GDKerrbuf;
char *s;
if (errbuf && *errbuf) {
if (strncmp(errbuf, "!ERROR: ", 8) == 0)
errbuf += 8;
if (strchr(errbuf, '!') == errbuf + 5) {
s = createException(MAL, malfunc, "%s", errbuf);
} else if ((s = strchr(errbuf, ':')) != NULL && s[1] == ' ') {
s = createException(MAL, malfunc, "%s", s + 2);
} else {
s = createException(MAL, malfunc, "%s", errbuf);
}
*GDKerrbuf = 0;
return s;
}
throw(MAL, malfunc, OPERATION_FAILED);
}
*retval = bn->batCacheid;
BBPkeepref(bn->batCacheid);
return MAL_SUCCEED;
}
/*
* grouped aggregates
*/
static str
AGGRgrouped(bat *retval, BAT *b, BAT *g, BAT *e, int tp,
BAT *(*grpfunc)(BAT *, BAT *, BAT *, BAT *, int, int, int),
int skip_nils,
const char *malfunc)
{
BAT *bn, *t, *map;
if (b == NULL || g == NULL || e == NULL) {
if (b)
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
throw(MAL, malfunc, RUNTIME_OBJECT_MISSING);
}
if (tp == TYPE_any && grpfunc == BATgroupmedian)
tp = b->ttype;
if (!BAThdense(b) || !BAThdense(g)) {
/* if b or g don't have a dense head, replace the head with a
* dense sequence */
t = BATjoin(BATmirror(b), g, MIN(BATcount(b), BATcount(g)));
BBPreleaseref(b->batCacheid);
BBPreleaseref(g->batCacheid);
b = BATmirror(BATmark(t, 0));
g = BATmirror(BATmark(BATmirror(t), 0));
BBPreleaseref(t->batCacheid);
}
if (b->hseqbase != g->hseqbase || BATcount(b) != BATcount(g)) {
/* b and g are not aligned: align them by creating a view on
* one or the other */
oid min; /* lowest common oid */
oid max; /* highest common oid */
min = b->hseqbase;
if (min < g->hseqbase)
min = g->hseqbase;
max = b->hseqbase + BATcount(b);
if (g->hseqbase + BATcount(g) < max)
max = g->hseqbase + BATcount(g);
if (b->hseqbase != min || b->hseqbase + BATcount(b) != max) {
if (min >= max)
min = max = b->hseqbase;
t = BATslice(b, BUNfirst(b) + (BUN) (min - b->hseqbase),
BUNfirst(b) + (BUN) (max - b->hseqbase));
BBPreleaseref(b->batCacheid);
b = t;
}
if (g->hseqbase != min || g->hseqbase + BATcount(g) != max) {
if (min >= max)
min = max = g->hseqbase;
t = BATslice(g, BUNfirst(g) + (BUN) (min - g->hseqbase),
BUNfirst(g) + (BUN) (max - g->hseqbase));
BBPreleaseref(g->batCacheid);
g = t;
}
}
if (!BAThdense(e)) {
/* if e doesn't have a dense head, renumber the group ids with
* a dense sequence at the cost of some left joins */
map = BATmark(e, 0); /* [gid,newgid(dense)] */
BBPreleaseref(e->batCacheid);
e = BATmirror(map); /* [newgid(dense),gid] */
t = BATleftjoin(g, map, BATcount(g)); /* [oid,newgid] */
BBPreleaseref(g->batCacheid);
g = t;
} else {
map = NULL;
}
bn = (*grpfunc)(b, g, e, NULL, tp, skip_nils, 1);
if (bn != NULL && (grpfunc == BATgroupmin || grpfunc == BATgroupmax)) {
BAT *bnn = BATouterjoin(bn, b, BATcount(bn));
BBPreleaseref(bn->batCacheid);
bn = bnn;
}
BBPreleaseref(b->batCacheid);
BBPreleaseref(g->batCacheid);
if (map == NULL) /* if map!=NULL, e is mirror of map */
BBPreleaseref(e->batCacheid);
if (bn == NULL) {
char *errbuf = GDKerrbuf;
char *s;
if (map)
BBPreleaseref(map->batCacheid);
if (errbuf && *errbuf) {
if (strncmp(errbuf, "!ERROR: ", 8) == 0)
errbuf += 8;
if (strchr(errbuf, '!') == errbuf + 5) {
s = createException(MAL, malfunc, "%s", errbuf);
} else if ((s = strchr(errbuf, ':')) != NULL && s[1] == ' ') {
s = createException(MAL, malfunc, "%s", s + 2);
} else {
s = createException(MAL, malfunc, "%s", errbuf);
}
*GDKerrbuf = 0;
return s;
}
throw(MAL, malfunc, OPERATION_FAILED);
}
if (map) {
t = BATleftjoin(map, bn, BATcount(bn));
BBPreleaseref(map->batCacheid);
BBPreleaseref(bn->batCacheid);
bn = t;
}
*retval = bn->batCacheid;
BBPkeepref(bn->batCacheid);
return MAL_SUCCEED;
}
static gdk_return
CMDinfo(BAT **ret1, BAT **ret2, BAT *b)
{
BAT *bk, *bv;
const char *mode, *accessmode;
if (!(bk = BATnew(TYPE_void, TYPE_str, 128, TRANSIENT)))
return GDK_FAIL;
if (!(bv = BATnew(TYPE_void, TYPE_str, 128, TRANSIENT))) {
BBPreclaim(bk);
return GDK_FAIL;
}
BATseqbase(bk,0);
BATseqbase(bv,0);
*ret1 = bk;
*ret2 = bv;
if (b->batPersistence == PERSISTENT) {
mode = "persistent";
} else if (b->batPersistence == TRANSIENT) {
mode = "transient";
} else {
mode ="unknown";
}
switch (b->batRestricted) {
case BAT_READ:
accessmode = "read-only";
break;
case BAT_WRITE:
accessmode = "updatable";
break;
case BAT_APPEND:
accessmode = "append-only";
break;
default:
accessmode = "unknown";
}
BUNappend(bk, "batId", FALSE);
BUNappend(bv, BATgetId(b),FALSE);
BUNappend(bk, "batCacheid", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->batCacheid)),FALSE);
BUNappend(bk, "hparentid", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->H->heap.parentid)),FALSE);
BUNappend(bk, "tparentid", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->T->heap.parentid)),FALSE);
BUNappend(bk, "batSharecnt", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->batSharecnt)),FALSE);
BUNappend(bk, "batCount", FALSE);
BUNappend(bv, local_utoa((size_t)b->batCount),FALSE);
BUNappend(bk, "batCapacity", FALSE);
BUNappend(bv, local_utoa((size_t)b->batCapacity),FALSE);
BUNappend(bk, "head", FALSE);
BUNappend(bv, ATOMname(b->htype),FALSE);
BUNappend(bk, "tail", FALSE);
BUNappend(bv, ATOMname(b->ttype),FALSE);
BUNappend(bk, "batPersistence", FALSE);
BUNappend(bv, mode,FALSE);
BUNappend(bk, "batRestricted", FALSE);
BUNappend(bv, accessmode,FALSE);
BUNappend(bk, "batRefcnt", FALSE);
BUNappend(bv, local_itoa((ssize_t)(BBP_refs(b->batCacheid))),FALSE);
BUNappend(bk, "batLRefcnt", FALSE);
BUNappend(bv, local_itoa((ssize_t)(BBP_lrefs(b->batCacheid))),FALSE);
BUNappend(bk, "batDirty", FALSE);
BUNappend(bv, BATdirty(b) ? "dirty" : "clean",FALSE);
BUNappend(bk, "hsorted", FALSE);
BUNappend(bv, local_itoa((ssize_t)BAThordered(b)),FALSE);
BUNappend(bk, "hrevsorted", FALSE);
BUNappend(bv, local_itoa((ssize_t)BAThrevordered(b)),FALSE);
BUNappend(bk, "hident", FALSE);
BUNappend(bv, b->hident,FALSE);
BUNappend(bk, "hdense", FALSE);
BUNappend(bv, local_itoa((ssize_t)(BAThdense(b))),FALSE);
BUNappend(bk, "hseqbase", FALSE);
BUNappend(bv, oidtostr(b->hseqbase),FALSE);
BUNappend(bk, "hkey", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->hkey)),FALSE);
BUNappend(bk, "hvarsized", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->hvarsized)),FALSE);
BUNappend(bk, "halign", FALSE);
BUNappend(bv, local_utoa(b->halign),FALSE);
BUNappend(bk, "hnosorted", FALSE);
BUNappend(bv, local_utoa(b->H->nosorted),FALSE);
BUNappend(bk, "hnorevsorted", FALSE);
BUNappend(bv, local_utoa(b->H->norevsorted),FALSE);
BUNappend(bk, "hnodense", FALSE);
BUNappend(bv, local_utoa(b->H->nodense),FALSE);
BUNappend(bk, "hnokey[0]", FALSE);
BUNappend(bv, local_utoa(b->H->nokey[0]),FALSE);
BUNappend(bk, "hnokey[1]", FALSE);
BUNappend(bv, local_utoa(b->H->nokey[1]),FALSE);
BUNappend(bk, "hnonil", FALSE);
BUNappend(bv, local_utoa(b->H->nonil),FALSE);
BUNappend(bk, "hnil", FALSE);
BUNappend(bv, local_utoa(b->H->nil),FALSE);
BUNappend(bk, "tident", FALSE);
BUNappend(bv, b->tident,FALSE);
BUNappend(bk, "tdense", FALSE);
BUNappend(bv, local_itoa((ssize_t)(BATtdense(b))), FALSE);
BUNappend(bk, "tseqbase", FALSE);
BUNappend(bv, oidtostr(b->tseqbase), FALSE);
BUNappend(bk, "tsorted", FALSE);
BUNappend(bv, local_itoa((ssize_t)BATtordered(b)), FALSE);
BUNappend(bk, "trevsorted", FALSE);
BUNappend(bv, local_itoa((ssize_t)BATtrevordered(b)), FALSE);
BUNappend(bk, "tkey", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->tkey)), FALSE);
BUNappend(bk, "tvarsized", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->tvarsized)), FALSE);
BUNappend(bk, "talign", FALSE);
BUNappend(bv, local_utoa(b->talign), FALSE);
BUNappend(bk, "tnosorted", FALSE);
BUNappend(bv, local_utoa(b->T->nosorted), FALSE);
BUNappend(bk, "tnorevsorted", FALSE);
BUNappend(bv, local_utoa(b->T->norevsorted), FALSE);
BUNappend(bk, "tnodense", FALSE);
BUNappend(bv, local_utoa(b->T->nodense), FALSE);
BUNappend(bk, "tnokey[0]", FALSE);
BUNappend(bv, local_utoa(b->T->nokey[0]), FALSE);
BUNappend(bk, "tnokey[1]", FALSE);
BUNappend(bv, local_utoa(b->T->nokey[1]), FALSE);
BUNappend(bk, "tnonil", FALSE);
BUNappend(bv, local_utoa(b->T->nonil), FALSE);
BUNappend(bk, "tnil", FALSE);
BUNappend(bv, local_utoa(b->T->nil), FALSE);
BUNappend(bk, "batInserted", FALSE);
BUNappend(bv, local_utoa(b->batInserted), FALSE);
BUNappend(bk, "batDeleted", FALSE);
BUNappend(bv, local_utoa(b->batDeleted), FALSE);
BUNappend(bk, "batFirst", FALSE);
BUNappend(bv, local_utoa(b->batFirst), FALSE);
BUNappend(bk, "htop", FALSE);
BUNappend(bv, local_utoa(b->H->heap.free), FALSE);
BUNappend(bk, "ttop", FALSE);
BUNappend(bv, local_utoa(b->T->heap.free), FALSE);
BUNappend(bk, "batStamp", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->batStamp)), FALSE);
BUNappend(bk, "lastUsed", FALSE);
BUNappend(bv, local_itoa((ssize_t)(BBP_lastused(b->batCacheid))), FALSE);
BUNappend(bk, "curStamp", FALSE);
BUNappend(bv, local_itoa((ssize_t)(BBPcurstamp())), FALSE);
BUNappend(bk, "batCopiedtodisk", FALSE);
BUNappend(bv, local_itoa((ssize_t)(b->batCopiedtodisk)), FALSE);
BUNappend(bk, "batDirtydesc", FALSE);
BUNappend(bv, b->batDirtydesc ? "dirty" : "clean", FALSE);
BUNappend(bk, "H->heap.dirty", FALSE);
BUNappend(bv, b->H->heap.dirty ? "dirty" : "clean", FALSE);
BUNappend(bk, "T->heap.dirty", FALSE);
BUNappend(bv, b->T->heap.dirty ? "dirty" : "clean", FALSE);
infoHeap(bk, bv, &b->H->heap, "head.");
infoHeap(bk, bv, &b->T->heap, "tail.");
BUNappend(bk, "H->vheap->dirty", FALSE);
BUNappend(bv, (b->H->vheap && b->H->vheap->dirty) ? "dirty" : "clean", FALSE);
infoHeap(bk, bv, b->H->vheap, "hheap.");
BUNappend(bk, "T->vheap->dirty", FALSE);
BUNappend(bv, (b->T->vheap && b->T->vheap->dirty) ? "dirty" : "clean", FALSE);
infoHeap(bk, bv, b->T->vheap, "theap.");
/* dump index information */
if (b->H->hash) {
HASHinfo(bk, bv, b->H->hash, "hhash->");
}
if (b->T->hash) {
HASHinfo(bk, bv, b->T->hash, "thash->");
}
assert(BATcount(bk) == BATcount(bv));
return GDK_SUCCEED;
}
