brain/kvbrain: hash tag in keys

brain/kvbrain/kvbrain.go

@@ -2,6 +2,8 @@ package kvbrain
 
 import (
 	"context"
+	"hash/fnv"
+	"io"
 	"time"
 
 	"github.com/dgraph-io/badger/v4"
@@ -48,9 +50,6 @@ func New(knowledge *badger.DB) *Brain {
 	}
 }
 
-// tagBytes is the number of bytes used to record tags in the KV database.
-const tagBytes = 8 // TODO(zeph): we should just use a hash instead
-
 // Order returns the number of elements in the prefix of a chain. It is
 // called once at the beginning of learning. The returned value must always
 // be at least 1.
@@ -83,3 +82,9 @@ func (br *Brain) ForgetDuring(ctx context.Context, tag string, since, before tim
 func (br *Brain) ForgetUserSince(ctx context.Context, user *userhash.Hash, since time.Time) error {
 	panic("not implemented") // TODO: Implement
 }
+
+func hashTag(tag string) uint64 {
+	h := fnv.New64a()
+	io.WriteString(h, tag)
+	return h.Sum64()
+}

brain/kvbrain/learn.go

@@ -3,6 +3,7 @@ package kvbrain
 import (
 	"bytes"
 	"context"
+	"encoding/binary"
 	"errors"
 	"fmt"
 	"slices"
@@ -33,8 +34,8 @@ func (br *Brain) Learn(ctx context.Context, meta *brain.MessageMeta, tuples []br
 	for i, t := range tuples {
 		b.Reset()
 		// Write the tag.
-		u := make([]byte, tagBytes)
-		copy(u, meta.Tag)
+		u := make([]byte, 8)
+		binary.LittleEndian.PutUint64(u, hashTag(meta.Tag))
 		b.Write(u)
 		// Write prefixes.
 		k := slices.IndexFunc(t.Prefix, func(s string) bool { return s != "" })

brain/kvbrain/learn_test.go

@@ -2,6 +2,7 @@ package kvbrain
 
 import (
 	"context"
+	"encoding/binary"
 	"testing"
 	"time"
 
@@ -14,8 +15,8 @@ import (
 
 func TestLearn(t *testing.T) {
 	mkey := func(tag, toks string, id uuid.UUID) string {
-		b := make([]byte, tagBytes, tagBytes+len(toks)+len(id))
-		copy(b, tag)
+		b := make([]byte, 8, 8+len(toks)+len(id))
+		binary.LittleEndian.PutUint64(b, hashTag(tag))
 		b = append(b, toks...)
 		b = append(b, id[:]...)
 		return string(b)