Changeset: 07da947053f9 for MonetDB
URL: https://dev.monetdb.org/hg/MonetDB/rev/07da947053f9
Modified Files:
clients/Tests/exports.stable.out
gdk/gdk_logger.c
gdk/gdk_logger.h
gdk/gdk_logger_internals.h
sql/storage/store.c
Branch: group-commit
Log Message:
Merge with default.
diffs (truncated from 13311 to 300 lines):
diff --git a/.hgtags b/.hgtags
--- a/.hgtags
+++ b/.hgtags
@@ -761,3 +761,4 @@ 8c015afafb5903ea59b0e2cffac1138a0d82e007
cab90a348501b045e19cee5cebcc44f3800bd0a8 Jul2021_21
cab90a348501b045e19cee5cebcc44f3800bd0a8 Jul2021_SP5_release
5872f047d97c98d3a848514438b8f97fa446855d Jan2022_11
+025239a5a6f122042798c0f1132a2c6298514e06 Jan2022_13
diff --git a/MonetDB.spec b/MonetDB.spec
--- a/MonetDB.spec
+++ b/MonetDB.spec
@@ -848,6 +848,24 @@ fi
%endif
%changelog
+* Fri Apr 01 2022 Sjoerd Mullender <sjo...@acm.org> - 11.43.13-20220401
+- Rebuilt.
+- GH#7278: BUG when there is more than one field/filter in the having
+ clause
+
+* Fri Apr 1 2022 Sjoerd Mullender <sjo...@acm.org> - 11.43.13-20220401
+- gdk: Improved speed of BATappend to empty varsized bat: we now just copy
+ the heaps instead of inserting individual values.
+
+* Fri Apr 1 2022 Sjoerd Mullender <sjo...@acm.org> - 11.43.13-20220401
+- monetdb5: Improved parsing speed of blob values, especially on Windows.
+ On Windows, using the locale-aware functions isdigit and isxdigit is
+ comparatively very slow, so we avoid them.
+
+* Tue Mar 29 2022 Sjoerd Mullender <sjo...@acm.org> - 11.43.13-20220401
+- gdk: Improved speed of projection (BATproject) on varsized bats by sharing
+ the data heap (vheap).
+
* Fri Mar 25 2022 Sjoerd Mullender <sjo...@acm.org> - 11.43.11-20220325
- Rebuilt.
- GH#7252: Segmentation fault on second run
diff --git a/cmake/monetdb-versions.cmake b/cmake/monetdb-versions.cmake
--- a/cmake/monetdb-versions.cmake
+++ b/cmake/monetdb-versions.cmake
@@ -40,7 +40,7 @@ set(MONETDB_VERSION "${MONETDB_VERSION_M
# common/options and common/utils)
set(GDK_VERSION_MAJOR "25")
set(GDK_VERSION_MINOR "0")
-set(GDK_VERSION_PATCH "3")
+set(GDK_VERSION_PATCH "4")
set(GDK_VERSION
"${GDK_VERSION_MAJOR}.${GDK_VERSION_MINOR}.${GDK_VERSION_PATCH}")
# version of the MAPI library (subdirectory clients/mapilib)
@@ -52,7 +52,7 @@ set(MAPI_VERSION "${MAPI_VERSION_MAJOR}.
# version of the MONETDB5 library (subdirectory monetdb5, not including extras
or sql)
set(MONETDB5_VERSION_MAJOR "32")
set(MONETDB5_VERSION_MINOR "0")
-set(MONETDB5_VERSION_PATCH "4")
+set(MONETDB5_VERSION_PATCH "5")
set(MONETDB5_VERSION
"${MONETDB5_VERSION_MAJOR}.${MONETDB5_VERSION_MINOR}.${MONETDB5_VERSION_PATCH}")
# version of the MONETDBE library (subdirectory tools/monetdbe)
@@ -70,5 +70,5 @@ set(STREAM_VERSION "${STREAM_VERSION_MAJ
# version of the SQL library (subdirectory sql)
set(SQL_VERSION_MAJOR "12")
set(SQL_VERSION_MINOR "0")
-set(SQL_VERSION_PATCH "3")
+set(SQL_VERSION_PATCH "4")
set(SQL_VERSION
"${SQL_VERSION_MAJOR}.${SQL_VERSION_MINOR}.${SQL_VERSION_PATCH}")
diff --git a/debian/changelog b/debian/changelog
--- a/debian/changelog
+++ b/debian/changelog
@@ -1,3 +1,33 @@
+monetdb (11.43.13) unstable; urgency=low
+
+ * Rebuilt.
+ * GH#7278: BUG when there is more than one field/filter in the having
+ clause
+
+ -- Sjoerd Mullender <sjo...@acm.org> Fri, 01 Apr 2022 11:31:11 +0200
+
+monetdb (11.43.13) unstable; urgency=low
+
+ * gdk: Improved speed of BATappend to empty varsized bat: we now just copy
+ the heaps instead of inserting individual values.
+
+ -- Sjoerd Mullender <sjo...@acm.org> Fri, 1 Apr 2022 11:31:11 +0200
+
+monetdb (11.43.13) unstable; urgency=low
+
+ * monetdb5: Improved parsing speed of blob values, especially on Windows.
+ On Windows, using the locale-aware functions isdigit and isxdigit is
+ comparatively very slow, so we avoid them.
+
+ -- Sjoerd Mullender <sjo...@acm.org> Fri, 1 Apr 2022 11:31:11 +0200
+
+monetdb (11.43.13) unstable; urgency=low
+
+ * gdk: Improved speed of projection (BATproject) on varsized bats by sharing
+ the data heap (vheap).
+
+ -- Sjoerd Mullender <sjo...@acm.org> Tue, 29 Mar 2022 11:31:11 +0200
+
monetdb (11.43.11) unstable; urgency=low
* Rebuilt.
diff --git a/gdk/ChangeLog-Archive b/gdk/ChangeLog-Archive
--- a/gdk/ChangeLog-Archive
+++ b/gdk/ChangeLog-Archive
@@ -1,6 +1,14 @@
# DO NOT EDIT THIS FILE -- MAINTAINED AUTOMATICALLY
# This file contains past ChangeLog entries
+* Fri Apr 1 2022 Sjoerd Mullender <sjo...@acm.org> - 11.43.13-20220401
+- Improved speed of BATappend to empty varsized bat: we now just copy
+ the heaps instead of inserting individual values.
+
+* Tue Mar 29 2022 Sjoerd Mullender <sjo...@acm.org> - 11.43.13-20220401
+- Improved speed of projection (BATproject) on varsized bats by sharing
+ the data heap (vheap).
+
* Fri Mar 18 2022 Sjoerd Mullender <sjo...@acm.org> - 11.43.11-20220325
- Fixed a race condition which could cause a too large size being written
for a .theap file to the BBP.dir file after the correct size file had
diff --git a/gdk/gdk.h b/gdk/gdk.h
--- a/gdk/gdk.h
+++ b/gdk/gdk.h
@@ -774,22 +774,34 @@ typedef enum {
BAT_APPEND, /* only reads and appends allowed */
} restrict_t;
+/* batDirtydesc: should be set (true) if any of the following fields
+ * have changed since the bat was last saved: hseqbase, batRestricted,
+ * batTransient, batCount, and the theap properties tkey, tseqbase,
+ * tsorted, trevsorted, twidth, tshift, tnonil, tnil, tnokey, tnosorted,
+ * tnorevsorted, tminpos, tmaxpos, and tunique_est; in addition, the
+ * value should be set if the BBP field BBP_logical(bid) is changed.
+ *
+ * theaplock: this lock should be held when reading or writing any of
+ * the fields mentioned above for batDirtydesc, and also when reading or
+ * writing any of the following fields: batDirtydesc, theap, tvheap,
+ * batInserted, batCapacity. There is no need for the lock if the bat
+ * cannot possibly be modified concurrently, e.g. when it is new and not
+ * yet returned to the interpreter or during system initialization. */
typedef struct BAT {
/* static bat properties */
oid hseqbase; /* head seq base */
MT_Id creator_tid; /* which thread created it */
bat batCacheid; /* index into BBP */
+ role_t batRole; /* role of the bat */
/* dynamic bat properties */
- restrict_t batRestricted; /* access privileges */
- bool batTransient; /* should the BAT persist on disk? */
+ restrict_t batRestricted:2; /* access privileges */
bool
+ batTransient:1, /* should the BAT persist on disk? */
batCopiedtodisk:1, /* once written */
batDirtyflushed:1, /* was dirty before commit started? */
batDirtydesc:1; /* bat descriptor dirty marker */
- uint16_t /* adjacent bit fields are packed together (if they fit) */
- selcnt:10; /* how often used in equi select without hash */
- role_t batRole; /* role of the bat */
+ uint16_t selcnt; /* how often used in equi select without hash */
uint16_t unused; /* value=0 for now (sneakily used by mat.c) */
int batSharecnt; /* incoming view count */
@@ -947,6 +959,7 @@ typedef struct BATiter {
void *base;
Heap *vh;
BUN count;
+ BUN baseoff;
uint16_t width;
uint8_t shift;
int8_t type;
@@ -954,13 +967,24 @@ typedef struct BATiter {
BUN hfree, vhfree;
BUN minpos, maxpos;
double unique_est;
+ bool key:1,
+ nonil:1,
+ nil:1,
+ sorted:1,
+ revsorted:1,
+ hdirty:1,
+ vhdirty:1,
+ dirtydesc:1,
+ copiedtodisk:1,
+ transient:1;
+ restrict_t restricted:2;
+#ifndef NDEBUG
+ bool locked:1;
+#endif
union {
oid tvid;
bool tmsk;
};
-#ifndef NDEBUG
- bool locked;
-#endif
} BATiter;
static inline BATiter
@@ -973,6 +997,7 @@ bat_iterator_nolock(BAT *b)
.b = b,
.h = b->theap,
.base = b->theap->base ? b->theap->base + (b->tbaseoff
<< b->tshift) : NULL,
+ .baseoff = b->tbaseoff,
.vh = b->tvheap,
.count = b->batCount,
.width = b->twidth,
@@ -989,6 +1014,19 @@ bat_iterator_nolock(BAT *b)
.minpos = isview ? BUN_NONE : b->tminpos,
.maxpos = isview ? BUN_NONE : b->tmaxpos,
.unique_est = b->tunique_est,
+ .key = b->tkey,
+ .nonil = b->tnonil,
+ .nil = b->tnil,
+ .sorted = b->tsorted,
+ .revsorted = b->trevsorted,
+ /* only look at heap dirty flag if we own it */
+ .hdirty = b->theap->parentid == b->batCacheid &&
b->theap->dirty,
+ /* also, if there is no vheap, it's not dirty */
+ .vhdirty = b->tvheap && b->tvheap->parentid ==
b->batCacheid && b->tvheap->dirty,
+ .dirtydesc = b->batDirtydesc,
+ .copiedtodisk = b->batCopiedtodisk,
+ .transient = b->batTransient,
+ .restricted = b->batRestricted,
#ifndef NDEBUG
.locked = false,
#endif
@@ -1023,6 +1061,20 @@ bat_iterator(BAT *b)
return bi;
}
+/* return a copy of a BATiter instance; needs to be released with
+ * bat_iterator_end */
+static inline BATiter
+bat_iterator_copy(BATiter *bip)
+{
+ assert(bip);
+ assert(bip->locked);
+ if (bip->h)
+ HEAPincref(bip->h);
+ if (bip->vh)
+ HEAPincref(bip->vh);
+ return *bip;
+}
+
static inline void
bat_iterator_end(BATiter *bip)
{
@@ -1252,6 +1304,10 @@ gdk_export restrict_t BATgetaccess(BAT *
(b)->batDirtydesc || \
(b)->theap->dirty || \
((b)->tvheap != NULL && (b)->tvheap->dirty))
+#define BATdirtybi(bi) (!(bi).copiedtodisk || \
+ (bi).dirtydesc || \
+ (bi).hdirty || \
+ (bi).vhdirty)
#define BATdirtydata(b) (!(b)->batCopiedtodisk ||
\
(b)->theap->dirty || \
((b)->tvheap != NULL && (b)->tvheap->dirty))
@@ -1359,8 +1415,6 @@ gdk_export gdk_return BATsort(BAT **sort
gdk_export void GDKqsort(void *restrict h, void *restrict t, const void
*restrict base, size_t n, int hs, int ts, int tpe, bool reverse, bool nilslast);
-#define BATtordered(b) ((b)->tsorted)
-#define BATtrevordered(b) ((b)->trevsorted)
/* BAT is dense (i.e., BATtvoid() is true and tseqbase is not NIL) */
#define BATtdense(b) (!is_oid_nil((b)->tseqbase) && \
((b)->tvheap == NULL || (b)->tvheap->free == 0))
@@ -1368,7 +1422,8 @@ gdk_export void GDKqsort(void *restrict
#define BATtvoid(b) (BATtdense(b) || (b)->ttype==TYPE_void)
#define BATtkey(b) ((b)->tkey || BATtdense(b))
-/* set some properties that are trivial to deduce */
+/* set some properties that are trivial to deduce; called with theaplock
+ * held */
static inline void
BATsettrivprop(BAT *b)
{
diff --git a/gdk/gdk_aggr.c b/gdk/gdk_aggr.c
--- a/gdk/gdk_aggr.c
+++ b/gdk/gdk_aggr.c
@@ -961,8 +961,8 @@ BATgroupsum(BAT *b, BAT *g, BAT *e, BAT
gids = (const oid *) Tloc(g, 0);
BATiter bi = bat_iterator(b);
- nils = dosum(bi.base, b->tnonil, b->hseqbase, &ci,
- Tloc(bn, 0), ngrp, b->ttype, tp, gids, min, max,
+ nils = dosum(bi.base, bi.nonil, b->hseqbase, &ci,
+ Tloc(bn, 0), ngrp, bi.type, tp, gids, min, max,
skip_nils, abort_on_error, true, __func__, &algo);
bat_iterator_end(&bi);
@@ -1191,8 +1191,8 @@ BATsum(void *res, int tp, BAT *b, BAT *s
if (ci.ncand == 0)
return GDK_SUCCEED;
BATiter bi = bat_iterator(b);
- BUN nils = dosum(bi.base, b->tnonil, b->hseqbase, &ci,
- res, true, b->ttype, tp, &min, min, max,
+ BUN nils = dosum(bi.base, bi.nonil, b->hseqbase, &ci,
+ res, true, bi.type, tp, &min, min, max,
skip_nils, abort_on_error, nil_if_empty, __func__,
&algo);
_______________________________________________
checkin-list mailing list -- checkin-list@monetdb.org
To unsubscribe send an email to checkin-list-le...@monetdb.org
