// few bytes over it, because the view measures the real sizes.
const nodeBytes = 512
-// Node is one line of the summary tree. Node (l, i) covers the 2^l messages
-// from i*2^l on. A node of level 0 holds one message. A node of a higher level
-// merges its two children, (l-1, 2i) and (l-1, 2i+1). Size is the UTF-8 byte
-// length of Text.
-type Node struct {
- Level int `json:"l"`
- Index int `json:"i"`
- Text string `json:"text"`
- Size int `json:"size"`
-}
-
-// nodeKey names the node (l, i).
-type nodeKey struct {
- level int
- index int
-}
-
// treeRecordBytes is the size of a record of index/tree.idx: the position of
// the node line, the node index, and the node level.
const treeRecordBytes = 32
// built again from the day files.
type Tree struct {
clock func() time.Time
- journal journal
+ journal Journal
dir string
- order *records
- levels map[int]*records
+ order *Records
+ levels map[int]*Records
failure error
}
+// Node is one line of the summary tree. Node (l, i) covers the 2^l messages
+// from i*2^l on. A node of level 0 holds one message. A node of a higher level
+// merges its two children, (l-1, 2i) and (l-1, 2i+1). Size is the UTF-8 byte
+// length of Text.
+type Node struct {
+ Level int `json:"l"`
+ Index int `json:"i"`
+ Text string `json:"text"`
+ Size int `json:"size"`
+}
+
+// NodeKey names the node (l, i).
+type NodeKey struct {
+ level int
+ index int
+}
+
// openTree opens the tree of the chat in root and indexes its new lines. A chat
// of total messages cannot have a node that covers a later message.
func openTree(ctx context.Context, root string, total int, clock func() time.Time) (*Tree, error) {
t := &Tree{
clock: clock,
- journal: journal{
+ journal: Journal{
dir: filepath.Join(root, "tree"),
},
dir: filepath.Join(root, "index"),
- levels: make(map[int]*records),
+ levels: make(map[int]*Records),
}
err := t.open(ctx, total)
func (t *Tree) open(ctx context.Context, total int) error {
var err error
+
t.order, err = openRecords(filepath.Join(t.dir, "tree.idx"), treeRecordBytes)
if err != nil {
- return err
+ return fmt.Errorf("open node order index: %w", err)
}
- start := position{}
+ start := Position{}
if t.order.count == 0 {
err = t.removeLevels()
if err != nil {
- return err
+ return fmt.Errorf("remove empty level indexes: %w", err)
}
} else {
record, _, err := t.order.Read(t.order.count - 1)
if err != nil {
- return err
+ return fmt.Errorf("read last node order: %w", err)
}
// A crash can stop an add between the two index writes, so the
// last node gets its level record again.
- node, last := decodeTreeRecord(record)
+ node, last := treeDecodeRecord(record)
err = t.writeLevel(node, last)
if err != nil {
- return err
+ return fmt.Errorf("restore last node level: %w", err)
}
- start = position{
+ start = Position{
day: last.day,
offset: last.offset + uint64(last.length),
}
}
- return t.journal.Scan(ctx, start, func(line []byte, place position) error {
+ err = t.journal.Scan(ctx, start, func(line []byte, place Position) error {
var node Node
+
err := json.Unmarshal(line, &node)
if err != nil {
- return err
+ return fmt.Errorf("decode tree node: %w", err)
}
err = t.check(node)
if err != nil {
- return err
+ return fmt.Errorf("validate scanned node: %w", err)
}
if nodeEnd(node.Level, node.Index) > total {
return fmt.Errorf("node (%d, %d) covers messages after the log", node.Level, node.Index)
}
- return t.addIndex(nodeKey{node.Level, node.Index}, place)
+ err = t.addIndex(NodeKey{node.Level, node.Index}, place)
+ if err != nil {
+ return fmt.Errorf("index scanned node: %w", err)
+ }
+
+ return nil
})
+ if err != nil {
+ return fmt.Errorf("scan node journal: %w", err)
+ }
+
+ return nil
+}
+
+// Close closes the files of the tree.
+func (t *Tree) Close() error {
+ err := t.journal.Close()
+ if t.order != nil {
+ err = errors.Join(err, t.order.Close())
+ t.order = nil
+ }
+
+ for level, index := range t.levels {
+ err = errors.Join(err, index.Close())
+
+ delete(t.levels, level)
+ }
+
+ return err
}
// removeLevels removes the level indexes, which are not valid without the order
}
// writeLevel writes the position of node to its level index.
-func (t *Tree) writeLevel(node nodeKey, place position) error {
+func (t *Tree) writeLevel(node NodeKey, place Position) error {
level, err := t.level(node.level)
if err != nil {
- return err
+ return fmt.Errorf("open node level index: %w", err)
}
record := make([]byte, positionBytes)
encodePosition(record, place)
- return level.Write(node.index, record)
-}
-
-// level returns the index of level, and opens it at first use.
-func (t *Tree) level(level int) (*records, error) {
- index, open := t.levels[level]
- if open {
- return index, nil
- }
-
- index, err := openRecords(filepath.Join(t.dir, fmt.Sprintf("tree-%d.idx", level)), positionBytes)
+ err = level.Write(node.index, record)
if err != nil {
- return nil, err
+ return fmt.Errorf("write node level record: %w", err)
}
- t.levels[level] = index
-
- return index, nil
+ return nil
}
// check makes sure that node is valid and can join the tree now.
built, err := t.Built(node.Level, node.Index)
if err != nil {
- return err
+ return fmt.Errorf("check target node: %w", err)
}
if built {
for _, index := range []int{2 * node.Index, 2*node.Index + 1} {
built, err = t.Built(node.Level-1, index)
if err != nil {
- return err
+ return fmt.Errorf("check child node: %w", err)
}
if !built {
}
// addIndex adds the node line at place to both indexes.
-func (t *Tree) addIndex(node nodeKey, place position) error {
+func (t *Tree) addIndex(node NodeKey, place Position) error {
record := make([]byte, treeRecordBytes)
encodePosition(record, place)
binary.LittleEndian.PutUint64(record[positionBytes:], uint64(node.index))
err := t.order.Append(record)
if err != nil {
- return err
+ return fmt.Errorf("append node order: %w", err)
}
- return t.writeLevel(node, place)
-}
-
-// Close closes the files of the tree.
-func (t *Tree) Close() error {
- err := t.journal.Close()
- if t.order != nil {
- err = errors.Join(err, t.order.Close())
- t.order = nil
- }
-
- for level, index := range t.levels {
- err = errors.Join(err, index.Close())
- delete(t.levels, level)
+ err = t.writeLevel(node, place)
+ if err != nil {
+ return fmt.Errorf("write indexed node level: %w", err)
}
- return err
+ return nil
}
-// Add writes a new node to the tree.
-func (t *Tree) Add(node Node) error {
- if t.failure != nil {
- return fmt.Errorf("tree needs a reopen after a failed write: %w", t.failure)
- }
-
- err := t.check(node)
- if err != nil {
- return err
+// level returns the index of level, and opens it at first use.
+func (t *Tree) level(level int) (*Records, error) {
+ index, open := t.levels[level]
+ if open {
+ return index, nil
}
- line, err := json.Marshal(node)
+ index, err := openRecords(filepath.Join(t.dir, fmt.Sprintf("tree-%d.idx", level)), positionBytes)
if err != nil {
- return fmt.Errorf("encode node: %w", err)
- }
-
- place, err := t.journal.Write(t.clock().Format(time.DateOnly), append(line, '\n'))
- if err == nil {
- err = t.addIndex(nodeKey{node.Level, node.Index}, place)
+ return nil, err
}
- if err != nil {
- t.failure = err
- return fmt.Errorf("write node (%d, %d): %w", node.Level, node.Index, err)
- }
+ t.levels[level] = index
- return nil
+ return index, nil
}
// Built tells if the node (l, i) is built.
}
// place returns the position of the node (l, i) if it is built.
-func (t *Tree) place(level, index int) (position, bool, error) {
+func (t *Tree) place(level, index int) (Position, bool, error) {
if level < 0 || level >= bits.UintSize-2 || index < 0 {
- return position{}, false, nil
+ return Position{}, false, nil
}
records, err := t.level(level)
if err != nil {
- return position{}, false, err
+ return Position{}, false, fmt.Errorf("open node level index: %w", err)
}
record, found, err := records.Read(index)
if err != nil || !found {
- return position{}, false, err
+ if err != nil {
+ return Position{}, false, fmt.Errorf("read node level record: %w", err)
+ }
+
+ return Position{}, false, nil
}
place := decodePosition(record)
return place, place.length > 0, nil
}
+// Add writes a new node to the tree.
+func (t *Tree) Add(node Node) error {
+ if t.failure != nil {
+ return fmt.Errorf("tree needs a reopen after a failed write: %w", t.failure)
+ }
+
+ err := t.check(node)
+ if err != nil {
+ return fmt.Errorf("validate node: %w", err)
+ }
+
+ line, err := json.Marshal(node)
+ if err != nil {
+ return fmt.Errorf("encode node: %w", err)
+ }
+
+ place, err := t.journal.Write(t.clock().Format(time.DateOnly), append(line, '\n'))
+ if err == nil {
+ err = t.addIndex(NodeKey{node.Level, node.Index}, place)
+ }
+
+ if err != nil {
+ t.failure = err
+ return fmt.Errorf("write node (%d, %d): %w", node.Level, node.Index, err)
+ }
+
+ return nil
+}
+
// Node reads the node (l, i) if it is built.
func (t *Tree) Node(level, index int) (Node, bool, error) {
place, built, err := t.place(level, index)
if err != nil || !built {
- return Node{}, false, err
+ if err != nil {
+ return Node{}, false, fmt.Errorf("find node position: %w", err)
+ }
+
+ return Node{}, false, nil
}
line, err := t.journal.Read(place)
}
var node Node
+
err = json.Unmarshal(line, &node)
if err != nil {
return Node{}, false, fmt.Errorf("decode node (%d, %d): %w", level, index, err)
return node, true, nil
}
-// Count returns the number of built nodes.
-func (t *Tree) Count() int {
- return t.order.count
-}
-
// Keys returns the keys of the built nodes, ordered by level and then by index.
-func (t *Tree) Keys() ([]nodeKey, error) {
- keys := make([]nodeKey, 0, t.order.count)
+func (t *Tree) Keys() ([]NodeKey, error) {
+ keys := make([]NodeKey, 0, t.order.count)
for n := range t.order.count {
record, _, err := t.order.Read(n)
if err != nil {
return nil, err
}
- key, _ := decodeTreeRecord(record)
+ key, _ := treeDecodeRecord(record)
keys = append(keys, key)
}
- slices.SortFunc(keys, func(a, b nodeKey) int {
+ slices.SortFunc(keys, func(a, b NodeKey) int {
return cmp.Or(cmp.Compare(a.level, b.level), cmp.Compare(a.index, b.index))
})
return keys, nil
}
-// leafSource returns the text that the leaf of message holds or summarizes.
-func leafSource(message Message) string {
- return message.Kind + ": " + message.Text
+// Count returns the number of built nodes.
+func (t *Tree) Count() int {
+ return t.order.count
+}
+
+// treeDecodeRecord reads the key and the position of a record of
+// index/tree.idx.
+func treeDecodeRecord(record []byte) (NodeKey, Position) {
+ key := NodeKey{
+ level: int(binary.LittleEndian.Uint32(record[positionBytes+8:])),
+ index: int(binary.LittleEndian.Uint64(record[positionBytes:])),
+ }
+
+ return key, decodePosition(record)
}
-// freeLeaf returns the leaf of message when its source fits in a node. Such a
-// leaf holds the source word for word and needs no model call.
+// nodeEnd returns the ID after the last message of node (l, i).
+func nodeEnd(level, index int) int {
+ return (index + 1) << level
+}
+
+// freeLeaf returns a text-only leaf when its source fits in a node. Images need
+// a model description even when their text fits.
func freeLeaf(message Message) (Node, bool) {
text := leafSource(message)
- if len(text) > nodeBytes {
+ if len(text) > nodeBytes || len(imageFiles(message.Text)) > 0 {
return Node{}, false
}
}, true
}
+// leafSource returns the text that the leaf of message holds or summarizes.
+func leafSource(message Message) string {
+ return message.Kind + ": " + message.Text
+}
+
// freeMerge returns the parent of two sibling nodes when both texts, joined by
// a newline, fit in a node. Such a parent needs no model call.
func freeMerge(left, right Node) (Node, bool) {
Size: len(text),
}, true
}
-
-// decodeTreeRecord reads the key and the position of a record of
-// index/tree.idx.
-func decodeTreeRecord(record []byte) (nodeKey, position) {
- key := nodeKey{
- level: int(binary.LittleEndian.Uint32(record[positionBytes+8:])),
- index: int(binary.LittleEndian.Uint64(record[positionBytes:])),
- }
-
- return key, decodePosition(record)
-}
-
-// nodeEnd returns the ID after the last message of node (l, i).
-func nodeEnd(level, index int) int {
- return (index + 1) << level
-}
projects / onechat / blobdiff