Name and order the Telegram bot code

[onechat] / tree.go

diff --git a/tree.go b/tree.go
index 30a40ae04aa1737ac782e10a62bf8ec5bf3b87dd..9fb033f9aacfb9f9c69c0e28bf73997958955290 100644 (file)
--- a/tree.go
+++ b/tree.go
@@ -3,7 +3,9 @@ package main
 import (
        "cmp"
        "context"
 import (
        "cmp"
        "context"

+       "encoding/binary"

        "encoding/json"
        "encoding/json"

+       "errors"

        "fmt"
        "math/bits"
        "os"
        "fmt"
        "math/bits"
        "os"


@@ -34,25 +36,39 @@ type nodeKey struct {
        index 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
+

 // Tree holds the built nodes of one chat. It writes one JSON line for each node
 // Tree holds the built nodes of one chat. It writes one JSON line for each node

-// to tree/YYYY-MM-DD.jsonl, named for the local day of the build, and loads all
-// lines again at start. Each node is built once and never again. A node of a
-// higher level is built only after both of its children. After a failed write,
-// the tree refuses all writes until it is opened again. The caller holds the
-// lock of the chat and serializes all calls.
+// to tree/YYYY-MM-DD.jsonl, named for the local day of the build. Each node is
+// built once and never again. A node of a higher level is built only after both
+// of its children. After a failed write, the tree refuses all writes until it
+// is opened again. The caller holds the lock of the chat and serializes all
+// calls.
+//
+// The tree keeps no node in memory. The index index/tree.idx lists the node
+// lines in file order, and index/tree-L.idx holds the position of node (L, i)
+// in record i, with a zero record for a node that is not built. At open the
+// tree indexes the lines after the last indexed one, so a missing index is
+// built again from the day files.

 type Tree struct {
        clock   func() time.Time
        journal journal
 type Tree struct {
        clock   func() time.Time
        journal journal

-       nodes   map[nodeKey]Node
+       dir     string
+       order   *records
+       levels  map[int]*records

        failure error
 }
 
        failure error
 }
 

-// openTree loads the nodes of the chat in root. A chat of total messages cannot
-// have a node that covers a later message.
+// 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) {
 func openTree(ctx context.Context, root string, total int, clock func() time.Time) (*Tree, error) {

-       err := os.MkdirAll(filepath.Join(root, "tree"), 0o700)
-       if err != nil {
-               return nil, fmt.Errorf("create node directory: %w", err)
+       for _, dir := range []string{"tree", "index"} {
+               err := os.MkdirAll(filepath.Join(root, dir), 0o700)
+               if err != nil {
+                       return nil, fmt.Errorf("create %s directory: %w", dir, err)
+               }

        }
 
        t := &Tree{
        }
 
        t := &Tree{


@@ -60,11 +76,59 @@ func openTree(ctx context.Context, root string, total int, clock func() time.Tim
                journal: journal{
                        dir: filepath.Join(root, "tree"),
                },
                journal: journal{
                        dir: filepath.Join(root, "tree"),
                },

-               nodes: make(map[nodeKey]Node),
+               dir:    filepath.Join(root, "index"),
+               levels: make(map[int]*records),
+       }
+
+       err := t.open(ctx, total)
+       if err != nil {
+               return nil, errors.Join(fmt.Errorf("index tree: %w", err), t.Close())
+       }
+
+       return t, nil
+}
+
+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
+       }
+
+       start := position{}
+       if t.order.count == 0 {
+               err = t.removeLevels()
+               if err != nil {
+                       return err
+               }
+       } else {
+               record, _, err := t.order.Read(t.order.count - 1)
+               if err != nil {
+                       return 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)
+               err = t.writeLevel(node, last)
+               if err != nil {
+                       return err
+               }
+
+               start = position{
+                       day:    last.day,
+                       offset: last.offset + uint64(last.length),
+               }

        }
 
        }
 

-       err = readLines(ctx, t.journal.dir, func(node Node) error {
-               err := t.check(node)
+       return t.journal.Scan(ctx, start, func(line []byte, place position) error {
+               var node Node
+               err := json.Unmarshal(line, &node)
+               if err != nil {
+                       return err
+               }
+
+               err = t.check(node)

                if err != nil {
                        return err
                }
                if err != nil {
                        return err
                }


@@ -73,15 +137,56 @@ func openTree(ctx context.Context, root string, total int, clock func() time.Tim
                        return fmt.Errorf("node (%d, %d) covers messages after the log", node.Level, node.Index)
                }
 
                        return fmt.Errorf("node (%d, %d) covers messages after the log", node.Level, node.Index)
                }
 

-               t.nodes[nodeKey{node.Level, node.Index}] = node
-
-               return nil
+               return t.addIndex(nodeKey{node.Level, node.Index}, place)

        })
        })

+}
+
+// removeLevels removes the level indexes, which are not valid without the order
+// index.
+func (t *Tree) removeLevels() error {
+       paths, err := filepath.Glob(filepath.Join(t.dir, "tree-*.idx"))

        if err != nil {
        if err != nil {

-               return nil, fmt.Errorf("load tree: %w", err)
+               return fmt.Errorf("list level indexes: %w", err)

        }
 
        }
 

-       return t, nil
+       for _, path := range paths {
+               err = os.Remove(path)
+               if err != nil {
+                       return fmt.Errorf("remove level index: %w", err)
+               }
+       }
+
+       return nil
+}
+
+// writeLevel writes the position of node to its level index.
+func (t *Tree) writeLevel(node nodeKey, place position) error {
+       level, err := t.level(node.level)
+       if err != nil {
+               return 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)
+       if err != nil {
+               return nil, err
+       }
+
+       t.levels[level] = index
+
+       return index, nil

 }
 
 // check makes sure that node is valid and can join the tree now.
 }
 
 // check makes sure that node is valid and can join the tree now.


@@ -94,20 +199,62 @@ func (t *Tree) check(node Node) error {
                return fmt.Errorf("node (%d, %d) has no valid text of its size", node.Level, node.Index)
        }
 
                return fmt.Errorf("node (%d, %d) has no valid text of its size", node.Level, node.Index)
        }
 

-       if t.Built(node.Level, node.Index) {
+       built, err := t.Built(node.Level, node.Index)
+       if err != nil {
+               return err
+       }
+
+       if built {

                return fmt.Errorf("node (%d, %d) is already built", node.Level, node.Index)
        }
 
                return fmt.Errorf("node (%d, %d) is already built", node.Level, node.Index)
        }
 

-       if node.Level > 0 && (!t.Built(node.Level-1, 2*node.Index) || !t.Built(node.Level-1, 2*node.Index+1)) {
-               return fmt.Errorf("node (%d, %d) has a child that is not built", node.Level, node.Index)
+       if node.Level == 0 {
+               return nil
+       }
+
+       for _, index := range []int{2 * node.Index, 2*node.Index + 1} {
+               built, err = t.Built(node.Level-1, index)
+               if err != nil {
+                       return err
+               }
+
+               if !built {
+                       return fmt.Errorf("node (%d, %d) has a child that is not built", node.Level, node.Index)
+               }

        }
 
        return nil
 }
 
        }
 
        return nil
 }
 

-// Close closes the open day file.
+// addIndex adds the node line at place to both indexes.
+func (t *Tree) addIndex(node nodeKey, place position) error {
+       record := make([]byte, treeRecordBytes)
+       encodePosition(record, place)
+       binary.LittleEndian.PutUint64(record[positionBytes:], uint64(node.index))
+       binary.LittleEndian.PutUint32(record[positionBytes+8:], uint32(node.level))
+
+       err := t.order.Append(record)
+       if err != nil {
+               return err
+       }
+
+       return t.writeLevel(node, place)
+}
+
+// Close closes the files of the tree.

 func (t *Tree) Close() error {
 func (t *Tree) Close() error {

-       return t.journal.Close()
+       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

 }
 
 // Add writes a new node to the tree.
 }
 
 // Add writes a new node to the tree.


@@ -126,41 +273,94 @@ func (t *Tree) Add(node Node) error {
                return fmt.Errorf("encode node: %w", err)
        }
 
                return fmt.Errorf("encode node: %w", err)
        }
 

-       err = t.journal.Write(t.clock().Format(time.DateOnly), append(line, '\n'))
+       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)
        }
 
        if err != nil {
                t.failure = err
                return fmt.Errorf("write node (%d, %d): %w", node.Level, node.Index, err)
        }
 

-       t.nodes[nodeKey{node.Level, node.Index}] = node
-

        return nil
 }
 
 // Built tells if the node (l, i) is built.
        return nil
 }
 
 // Built tells if the node (l, i) is built.

-func (t *Tree) Built(level, index int) bool {
-       _, built := t.nodes[nodeKey{level, index}]
-       return built
+func (t *Tree) Built(level, index int) (bool, error) {
+       _, built, err := t.place(level, index)
+       return built, err
+}
+
+// place returns the position of the node (l, i) if it is built.
+func (t *Tree) place(level, index int) (position, bool, error) {
+       if level < 0 || level >= bits.UintSize-2 || index < 0 {
+               return position{}, false, nil
+       }
+
+       records, err := t.level(level)
+       if err != nil {
+               return position{}, false, err
+       }
+
+       record, found, err := records.Read(index)
+       if err != nil || !found {
+               return position{}, false, err
+       }
+
+       place := decodePosition(record)
+
+       return place, place.length > 0, 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
+       }
+
+       line, err := t.journal.Read(place)
+       if err != nil {
+               return Node{}, false, fmt.Errorf("read node (%d, %d): %w", level, index, err)
+       }
+
+       var node Node
+       err = json.Unmarshal(line, &node)
+       if err != nil {
+               return Node{}, false, fmt.Errorf("decode node (%d, %d): %w", level, index, err)
+       }
+
+       if node.Level != level || node.Index != index {
+               return Node{}, false, fmt.Errorf("index of node (%d, %d) points to node (%d, %d)", level, index, node.Level, node.Index)
+       }
+
+       return node, true, nil

 }
 
 }
 

-// Node returns the node (l, i) if it is built.
-func (t *Tree) Node(level, index int) (Node, bool) {
-       node, built := t.nodes[nodeKey{level, index}]
-       return node, built
+// Count returns the number of built nodes.
+func (t *Tree) Count() int {
+       return t.order.count

 }
 
 }
 

-// Nodes returns the built nodes, ordered by level and then by index.
-func (t *Tree) Nodes() []Node {
-       nodes := make([]Node, 0, len(t.nodes))
-       for _, node := range t.nodes {
-               nodes = append(nodes, node)
+// 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)
+       for n := range t.order.count {
+               record, _, err := t.order.Read(n)
+               if err != nil {
+                       return nil, err
+               }
+
+               key, _ := decodeTreeRecord(record)
+               keys = append(keys, key)

        }
 
        }
 

-       slices.SortFunc(nodes, func(a, b Node) int {
-               return cmp.Or(cmp.Compare(a.Level, b.Level), cmp.Compare(a.Index, b.Index))
+       slices.SortFunc(keys, func(a, b nodeKey) int {
+               return cmp.Or(cmp.Compare(a.level, b.level), cmp.Compare(a.index, b.index))

        })
 
        })
 

-       return nodes
+       return keys, nil

 }
 
 // leafSource returns the text that the leaf of message holds or summarizes.
 }
 
 // leafSource returns the text that the leaf of message holds or summarizes.


@@ -201,6 +401,17 @@ func freeMerge(left, right Node) (Node, bool) {
        }, true
 }
 
        }, 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
 // nodeEnd returns the ID after the last message of node (l, i).
 func nodeEnd(level, index int) int {
        return (index + 1) << level