X Start reworking BTree to provide keycov on visit callback

go/zodb/btree/btree.go.in

@@ -65,7 +65,7 @@ type BTree struct {
 
 // Entry is one BTree node entry.
 //
-// It contains key and child, who is either BTree or Bucket.
+// It contains key and child, which is either BTree or Bucket.
 //
 // Key limits child's keys - see BTree.Entryv for details.
 type Entry struct {
@@ -102,6 +102,10 @@ type BucketEntry struct {
 	value interface{}
 }
 
+
+const _KeyMin KEY = math.Min<Key>
+const _KeyMax KEY = math.Max<Key>
+
 // ---- access []entry ----
 
 // Key returns BTree entry key.
@@ -123,7 +127,7 @@ func (e *Entry) Child() Node { return e.child }
 // Children of all entries are guaranteed to be of the same kind - either all BTree, or all Bucket.
 //
 // The caller must not modify returned array.
-func (t *BTree) Entryv() []Entry {
+func (t *BTree) Entryv() /*readonly*/ []Entry {
 	return t.data
 }
 
@@ -168,36 +172,54 @@ func (t *BTree) Get(ctx context.Context, key KEY) (_ interface{}, _ bool, err er
 // VGet is like Get but also calls visit while traversing the tree.
 //
 // Visit is called with node being activated.
-func (t *BTree) VGet(ctx context.Context, key KEY, visit func(node Node)) (_ interface{}, _ bool, err error) {
+func (t *BTree) VGet(ctx context.Context, key KEY, visit func(node Node, keycov KeyRange)) (_ interface{}, _ bool, err error) {
 	defer xerr.Contextf(&err, "btree(%s): get %v", t.POid(), key)
 	err = t.PActivate(ctx)
 	if err != nil {
 		return nil, false, err
 	}
 
+	keycov := KeyRange{Lo: _KeyMin, Hi_: _KeyMax}
 	if visit != nil {
-		visit(t)
-	}
-
-	if len(t.data) == 0 {
-		// empty btree
-		t.PDeactivate()
-		return nil, false, nil
+		visit(t, keycov)
 	}
 
 	for {
+		l := len(t.data)
+		if l == 0 {
+			// empty btree (top, or in leaf)
+			t.PDeactivate()
+			return nil, false, nil
+		}
+
 		// search i: K(i) ≤ k < K(i+1)	; K(0) = -∞, K(len) = +∞
-		i := sort.Search(len(t.data), func(i int) bool {
+		i := sort.Search(l, func(i int) bool {
 			j := i + 1
 			if j == len(t.data) {
 				return true // [len].key = +∞
 			}
 			return key < t.data[j].key
 		})
-
-		// FIXME panic index out of range (empty T without children;
-		// logically incorrect, but Restructure generated it once)
+		// i < l
 		child := t.data[i].child
+
+		// shorten global keycov by local [lo,hi) for this child
+		lo := _KeyMin
+		if i > 0 {
+			lo = t.data[i].key
+		}
+		i++
+		hi_ := _KeyMax
+		if i < l {
+			hi_ = t.data[i].key
+		}
+		if hi_ != _KeyMax {
+			hi_--
+		}
+		keycov.Lo  = kmax(keycov.Lo, lo)
+		keycov.Hi_ = kmin(keycov.Hi_, hi_)
+
+
 		t.PDeactivate()
 		err = child.PActivate(ctx)
 		if err != nil {
@@ -207,7 +229,7 @@ func (t *BTree) VGet(ctx context.Context, key KEY, visit func(node Node)) (_ int
 		// XXX verify child keys are in valid range according to parent
 
 		if visit != nil {
-			visit(child)
+			visit(child, keycov)
 		}
 
 		switch child := child.(type) {
@@ -324,6 +346,7 @@ func (t *BTree) VMaxKey(ctx context.Context, visit func(node Node)) (_ KEY, _ bo
 	}
 
 	for {
+		// FIXME need to refresh l
 		child := t.data[l-1].child
 		t.PDeactivate()
 		err = child.PActivate(ctx)
@@ -591,7 +614,7 @@ func (bt *btreeState) PySetState(pystate interface{}) (err error) {
 	var kprev int64
 	var childrenKind int // 1 - BTree, 2 - Bucket
 	for i, idx := 0, 0; i < n; i++ {
-		key := int64(math.Min<Key>) // KEY(-∞)   (qualifies for ≤)
+		key := int64(_KeyMin) // KEY(-∞)   (qualifies for ≤)
 		if i > 0 {
 			// key[0] is unused and not saved
 			key, ok = t[idx].(int64) // XXX Xint
@@ -646,3 +669,66 @@ func init() {
 	zodb.RegisterClass("BTrees.BTree.BTree",  t(BTree{}),  t(btreeState{}))
 	zodb.RegisterClass("BTrees.BTree.Bucket", t(Bucket{}), t(bucketState{}))
 }
+
+
+// XXX place
+// KeyRange represents [lo,hi) key range.
+type KeyRange struct {
+	Lo  KEY
+	Hi_ KEY // NOTE _not_ hi) to avoid overflow at ∞;  hi = hi_ + 1
+}
+
+// Has returns whether key k belongs to the range.
+func (r *KeyRange) Has(k KEY) bool {
+	return (r.Lo <= k && k <= r.Hi_)
+}
+
+// Empty returns whether key range is empty.
+func (r *KeyRange) Empty() bool {
+	hi := r.Hi_
+	if hi == _KeyMax {
+		// [x,∞] cannot be empty because max x is ∞ and [∞,∞] has one element: ∞
+		return false
+	}
+	hi++ // no overflow
+	return r.Lo >= hi
+}
+
+func (r KeyRange) String() string {
+	var shi string
+	if r.Hi_ == _KeyMax {
+		shi = kstr(r.Hi_) // ∞
+	} else {
+		shi = fmt.Sprintf("%d", r.Hi_+1)
+	}
+	return fmt.Sprintf("[%s,%s)", kstr(r.Lo), shi)
+}
+
+
+// XXX place
+func kmin(a, b KEY) KEY {
+	if a < b {
+		return a
+	} else {
+		return b
+	}
+}
+
+func kmax(a, b KEY) KEY {
+	if a > b {
+		return a
+	} else {
+		return b
+	}
+}
+
+// kstr formats key as string.
+func kstr(k KEY) string {
+	if k == _KeyMin {
+		return "-∞"
+	}
+	if k == _KeyMax {
+		return "∞"
+	}
+	return fmt.Sprintf("%d", k)
+}

go/zodb/btree/gen-btree

@@ -28,6 +28,9 @@ out=$3
 kind=${KIND,,}	# IO -> io
 Key=${KEY^}
 
+KEYKIND=${KIND:0:1}	# IO -> I
+keykind=${KEYKIND,,}	# I -> i
+
 input=$(dirname $0)/btree.go.in
 
 echo "// Code generated by gen-btree; DO NOT EDIT." >$out
@@ -45,4 +48,10 @@ sed	\
 	-e "s/\bBucketEntry\b/${KIND}BucketEntry/g"	\
 	-e "s/\bbtreeState\b/${kind}btreeState/g"	\
 	-e "s/\bbucketState\b/${kind}bucketState/g"	\
+	-e "s/\b_KeyMin\b/_${KEYKIND}KeyMin/g"		\
+	-e "s/\b_KeyMax\b/_${KEYKIND}KeyMax/g"		\
+	-e "s/\bKeyRange\b/_${KEYKIND}KeyRange/g"	\
+	-e "s/\bkmin\b/${keykind}kmin/g"		\
+	-e "s/\bkmax\b/${keykind}kmax/g"		\
+	-e "s/\bkstr\b/${keykind}kstr/g"		\
 	$input >>$out

go/zodb/btree/ziobtree.go

@@ -67,7 +67,7 @@ type IOBTree struct {
 
 // IOEntry is one IOBTree node entry.
 //
-// It contains key and child, who is either IOBTree or IOBucket.
+// It contains key and child, which is either IOBTree or IOBucket.
 //
 // Key limits child's keys - see IOBTree.Entryv for details.
 type IOEntry struct {
@@ -104,6 +104,10 @@ type IOBucketEntry struct {
 	value interface{}
 }
 
+
+const _IKeyMin int32 = math.MinInt32
+const _IKeyMax int32 = math.MaxInt32
+
 // ---- access []entry ----
 
 // Key returns IOBTree entry key.
@@ -125,7 +129,7 @@ func (e *IOEntry) Child() IONode { return e.child }
 // Children of all entries are guaranteed to be of the same kind - either all IOBTree, or all IOBucket.
 //
 // The caller must not modify returned array.
-func (t *IOBTree) Entryv() []IOEntry {
+func (t *IOBTree) Entryv() /*readonly*/ []IOEntry {
 	return t.data
 }
 
@@ -170,36 +174,54 @@ func (t *IOBTree) Get(ctx context.Context, key int32) (_ interface{}, _ bool, er
 // VGet is like Get but also calls visit while traversing the tree.
 //
 // Visit is called with node being activated.
-func (t *IOBTree) VGet(ctx context.Context, key int32, visit func(node IONode)) (_ interface{}, _ bool, err error) {
+func (t *IOBTree) VGet(ctx context.Context, key int32, visit func(node IONode, keycov _IKeyRange)) (_ interface{}, _ bool, err error) {
 	defer xerr.Contextf(&err, "btree(%s): get %v", t.POid(), key)
 	err = t.PActivate(ctx)
 	if err != nil {
 		return nil, false, err
 	}
 
+	keycov := _IKeyRange{Lo: _IKeyMin, Hi_: _IKeyMax}
 	if visit != nil {
-		visit(t)
-	}
-
-	if len(t.data) == 0 {
-		// empty btree
-		t.PDeactivate()
-		return nil, false, nil
+		visit(t, keycov)
 	}
 
 	for {
+		l := len(t.data)
+		if l == 0 {
+			// empty btree (top, or in leaf)
+			t.PDeactivate()
+			return nil, false, nil
+		}
+
 		// search i: K(i) ≤ k < K(i+1)	; K(0) = -∞, K(len) = +∞
-		i := sort.Search(len(t.data), func(i int) bool {
+		i := sort.Search(l, func(i int) bool {
 			j := i + 1
 			if j == len(t.data) {
 				return true // [len].key = +∞
 			}
 			return key < t.data[j].key
 		})
-
-		// FIXME panic index out of range (empty T without children;
-		// logically incorrect, but Restructure generated it once)
+		// i < l
 		child := t.data[i].child
+
+		// shorten global keycov by local [lo,hi) for this child
+		lo := _IKeyMin
+		if i > 0 {
+			lo = t.data[i].key
+		}
+		i++
+		hi_ := _IKeyMax
+		if i < l {
+			hi_ = t.data[i].key
+		}
+		if hi_ != _IKeyMax {
+			hi_--
+		}
+		keycov.Lo  = ikmax(keycov.Lo, lo)
+		keycov.Hi_ = ikmin(keycov.Hi_, hi_)
+
+
 		t.PDeactivate()
 		err = child.PActivate(ctx)
 		if err != nil {
@@ -209,7 +231,7 @@ func (t *IOBTree) VGet(ctx context.Context, key int32, visit func(node IONode))
 		// XXX verify child keys are in valid range according to parent
 
 		if visit != nil {
-			visit(child)
+			visit(child, keycov)
 		}
 
 		switch child := child.(type) {
@@ -326,6 +348,7 @@ func (t *IOBTree) VMaxKey(ctx context.Context, visit func(node IONode)) (_ int32
 	}
 
 	for {
+		// FIXME need to refresh l
 		child := t.data[l-1].child
 		t.PDeactivate()
 		err = child.PActivate(ctx)
@@ -593,7 +616,7 @@ func (bt *iobtreeState) PySetState(pystate interface{}) (err error) {
 	var kprev int64
 	var childrenKind int // 1 - IOBTree, 2 - IOBucket
 	for i, idx := 0, 0; i < n; i++ {
-		key := int64(math.MinInt32) // int32(-∞)   (qualifies for ≤)
+		key := int64(_IKeyMin) // int32(-∞)   (qualifies for ≤)
 		if i > 0 {
 			// key[0] is unused and not saved
 			key, ok = t[idx].(int64) // XXX Xint
@@ -648,3 +671,66 @@ func init() {
 	zodb.RegisterClass("BTrees.IOBTree.IOBTree",  t(IOBTree{}),  t(iobtreeState{}))
 	zodb.RegisterClass("BTrees.IOBTree.IOBucket", t(IOBucket{}), t(iobucketState{}))
 }
+
+
+// XXX place
+// _IKeyRange represents [lo,hi) key range.
+type _IKeyRange struct {
+	Lo  int32
+	Hi_ int32 // NOTE _not_ hi) to avoid overflow at ∞;  hi = hi_ + 1
+}
+
+// Has returns whether key k belongs to the range.
+func (r *_IKeyRange) Has(k int32) bool {
+	return (r.Lo <= k && k <= r.Hi_)
+}
+
+// Empty returns whether key range is empty.
+func (r *_IKeyRange) Empty() bool {
+	hi := r.Hi_
+	if hi == _IKeyMax {
+		// [x,∞] cannot be empty because max x is ∞ and [∞,∞] has one element: ∞
+		return false
+	}
+	hi++ // no overflow
+	return r.Lo >= hi
+}
+
+func (r _IKeyRange) String() string {
+	var shi string
+	if r.Hi_ == _IKeyMax {
+		shi = ikstr(r.Hi_) // ∞
+	} else {
+		shi = fmt.Sprintf("%d", r.Hi_+1)
+	}
+	return fmt.Sprintf("[%s,%s)", ikstr(r.Lo), shi)
+}
+
+
+// XXX place
+func ikmin(a, b int32) int32 {
+	if a < b {
+		return a
+	} else {
+		return b
+	}
+}
+
+func ikmax(a, b int32) int32 {
+	if a > b {
+		return a
+	} else {
+		return b
+	}
+}
+
+// ikstr formats key as string.
+func ikstr(k int32) string {
+	if k == _IKeyMin {
+		return "-∞"
+	}
+	if k == _IKeyMax {
+		return "∞"
+	}
+	return fmt.Sprintf("%d", k)
+}

go/zodb/btree/zlobtree.go

@@ -67,7 +67,7 @@ type LOBTree struct {
 
 // LOEntry is one LOBTree node entry.
 //
-// It contains key and child, who is either LOBTree or LOBucket.
+// It contains key and child, which is either LOBTree or LOBucket.
 //
 // Key limits child's keys - see LOBTree.Entryv for details.
 type LOEntry struct {
@@ -104,6 +104,10 @@ type LOBucketEntry struct {
 	value interface{}
 }
 
+
+const _LKeyMin int64 = math.MinInt64
+const _LKeyMax int64 = math.MaxInt64
+
 // ---- access []entry ----
 
 // Key returns LOBTree entry key.
@@ -125,7 +129,7 @@ func (e *LOEntry) Child() LONode { return e.child }
 // Children of all entries are guaranteed to be of the same kind - either all LOBTree, or all LOBucket.
 //
 // The caller must not modify returned array.
-func (t *LOBTree) Entryv() []LOEntry {
+func (t *LOBTree) Entryv() /*readonly*/ []LOEntry {
 	return t.data
 }
 
@@ -170,36 +174,54 @@ func (t *LOBTree) Get(ctx context.Context, key int64) (_ interface{}, _ bool, er
 // VGet is like Get but also calls visit while traversing the tree.
 //
 // Visit is called with node being activated.
-func (t *LOBTree) VGet(ctx context.Context, key int64, visit func(node LONode)) (_ interface{}, _ bool, err error) {
+func (t *LOBTree) VGet(ctx context.Context, key int64, visit func(node LONode, keycov _LKeyRange)) (_ interface{}, _ bool, err error) {
 	defer xerr.Contextf(&err, "btree(%s): get %v", t.POid(), key)
 	err = t.PActivate(ctx)
 	if err != nil {
 		return nil, false, err
 	}
 
+	keycov := _LKeyRange{Lo: _LKeyMin, Hi_: _LKeyMax}
 	if visit != nil {
-		visit(t)
-	}
-
-	if len(t.data) == 0 {
-		// empty btree
-		t.PDeactivate()
-		return nil, false, nil
+		visit(t, keycov)
 	}
 
 	for {
+		l := len(t.data)
+		if l == 0 {
+			// empty btree (top, or in leaf)
+			t.PDeactivate()
+			return nil, false, nil
+		}
+
 		// search i: K(i) ≤ k < K(i+1)	; K(0) = -∞, K(len) = +∞
-		i := sort.Search(len(t.data), func(i int) bool {
+		i := sort.Search(l, func(i int) bool {
 			j := i + 1
 			if j == len(t.data) {
 				return true // [len].key = +∞
 			}
 			return key < t.data[j].key
 		})
-
-		// FIXME panic index out of range (empty T without children;
-		// logically incorrect, but Restructure generated it once)
+		// i < l
 		child := t.data[i].child
+
+		// shorten global keycov by local [lo,hi) for this child
+		lo := _LKeyMin
+		if i > 0 {
+			lo = t.data[i].key
+		}
+		i++
+		hi_ := _LKeyMax
+		if i < l {
+			hi_ = t.data[i].key
+		}
+		if hi_ != _LKeyMax {
+			hi_--
+		}
+		keycov.Lo  = lkmax(keycov.Lo, lo)
+		keycov.Hi_ = lkmin(keycov.Hi_, hi_)
+
+
 		t.PDeactivate()
 		err = child.PActivate(ctx)
 		if err != nil {
@@ -209,7 +231,7 @@ func (t *LOBTree) VGet(ctx context.Context, key int64, visit func(node LONode))
 		// XXX verify child keys are in valid range according to parent
 
 		if visit != nil {
-			visit(child)
+			visit(child, keycov)
 		}
 
 		switch child := child.(type) {
@@ -326,6 +348,7 @@ func (t *LOBTree) VMaxKey(ctx context.Context, visit func(node LONode)) (_ int64
 	}
 
 	for {
+		// FIXME need to refresh l
 		child := t.data[l-1].child
 		t.PDeactivate()
 		err = child.PActivate(ctx)
@@ -593,7 +616,7 @@ func (bt *lobtreeState) PySetState(pystate interface{}) (err error) {
 	var kprev int64
 	var childrenKind int // 1 - LOBTree, 2 - LOBucket
 	for i, idx := 0, 0; i < n; i++ {
-		key := int64(math.MinInt64) // int64(-∞)   (qualifies for ≤)
+		key := int64(_LKeyMin) // int64(-∞)   (qualifies for ≤)
 		if i > 0 {
 			// key[0] is unused and not saved
 			key, ok = t[idx].(int64) // XXX Xint
@@ -648,3 +671,66 @@ func init() {
 	zodb.RegisterClass("BTrees.LOBTree.LOBTree",  t(LOBTree{}),  t(lobtreeState{}))
 	zodb.RegisterClass("BTrees.LOBTree.LOBucket", t(LOBucket{}), t(lobucketState{}))
 }
+
+
+// XXX place
+// _LKeyRange represents [lo,hi) key range.
+type _LKeyRange struct {
+	Lo  int64
+	Hi_ int64 // NOTE _not_ hi) to avoid overflow at ∞;  hi = hi_ + 1
+}
+
+// Has returns whether key k belongs to the range.
+func (r *_LKeyRange) Has(k int64) bool {
+	return (r.Lo <= k && k <= r.Hi_)
+}
+
+// Empty returns whether key range is empty.
+func (r *_LKeyRange) Empty() bool {
+	hi := r.Hi_
+	if hi == _LKeyMax {
+		// [x,∞] cannot be empty because max x is ∞ and [∞,∞] has one element: ∞
+		return false
+	}
+	hi++ // no overflow
+	return r.Lo >= hi
+}
+
+func (r _LKeyRange) String() string {
+	var shi string
+	if r.Hi_ == _LKeyMax {
+		shi = lkstr(r.Hi_) // ∞
+	} else {
+		shi = fmt.Sprintf("%d", r.Hi_+1)
+	}
+	return fmt.Sprintf("[%s,%s)", lkstr(r.Lo), shi)
+}
+
+
+// XXX place
+func lkmin(a, b int64) int64 {
+	if a < b {
+		return a
+	} else {
+		return b
+	}
+}
+
+func lkmax(a, b int64) int64 {
+	if a > b {
+		return a
+	} else {
+		return b
+	}
+}
+
+// lkstr formats key as string.
+func lkstr(k int64) string {
+	if k == _LKeyMin {
+		return "-∞"
+	}
+	if k == _LKeyMax {
+		return "∞"
+	}
+	return fmt.Sprintf("%d", k)
+}