package main import ( "context" "errors" "fmt" "maps" "math/rand/v2" "os" "path/filepath" "runtime" "slices" "strings" "testing" "time" "github.com/stretchr/testify/require" ) const ( // replayMessages is the length of the synthetic log. replayMessages = 100000 // replayShortMessages is the length of the synthetic log with -short. // It is long enough for batches of the chat view. replayShortMessages = 5000 ) func TestMemoryRestartLoadsTheSameView(t *testing.T) { root := t.TempDir() clock := newTestClock() memory, err := openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) memory.chat.high = 200 memory.chat.low = 100 memory.cview.high = 80 memory.cview.low = 40 for n := range 100 { _, err = memory.Append("user", fmt.Sprint(n)) require.NoError(t, err) } view := memory.View() context := memory.Context(0, 100) saved, err := os.ReadFile(filepath.Join(root, "view.json")) require.NoError(t, err) require.Less(t, len(memory.chat.lines), 100) require.NoError(t, memory.Close()) memory, err = openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() require.Equal(t, view, memory.View()) require.Equal(t, context, memory.Context(0, 100)) loaded, err := os.ReadFile(filepath.Join(root, "view.json")) require.NoError(t, err) require.Equal(t, saved, loaded) } func TestMemoryOpensAfterACrashBeforeTheViewSave(t *testing.T) { root := t.TempDir() clock := newTestClock() memory, err := openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) _, err = memory.Append("user", "first") require.NoError(t, err) saved, err := os.ReadFile(filepath.Join(root, "view.json")) require.NoError(t, err) _, err = memory.Append("user", "second") require.NoError(t, err) view := memory.View() require.NoError(t, memory.Close()) // The log and the tree hold message 1, but the view does not. require.NoError(t, os.WriteFile(filepath.Join(root, "view.json"), saved, 0o600)) memory, err = openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) require.Equal(t, view, memory.View()) require.NoError(t, memory.Close()) } func TestMemoryKeepsTheCompactionViewCoarser(t *testing.T) { memory, err := openMemory(t.Context(), t.TempDir(), newTestClock().Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() memory.chat.high = 200 memory.chat.low = 100 memory.cview.high = 80 memory.cview.low = 40 for n := range 64 { _, err = memory.Append("user", fmt.Sprint(n%10)) require.NoError(t, err) coarse := 0 for _, line := range memory.chat.lines { outer := memory.cview.lines[coarse] for nodeEnd(outer.level, outer.index) <= line.index<summary failed level=0 index=0: fake failure\n", logs.String()) summarizer.SetFailing(false) _, err = memory.Append("user", "retry") require.NoError(t, err) require.NoError(t, memory.Settle(t.Context())) } func TestMemorySettleEndsWithItsContext(t *testing.T) { summarizer := newFakeSummarizer() summarizer.gate = make(chan struct{}) memory, err := openMemory(t.Context(), t.TempDir(), newTestClock().Now, summarizer, discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() _, err = memory.Append("echo", memoryLongText(0)) require.NoError(t, err) ctx, cancel := context.WithTimeout(t.Context(), 50*time.Millisecond) defer cancel() require.ErrorIs(t, memory.Settle(ctx), context.DeadlineExceeded) } func TestMemoryStartsALeafAfterFewerThanEightUnbuilt(t *testing.T) { summarizer := newFakeSummarizer() summarizer.SetFailing(true) memory, err := openMemory(t.Context(), t.TempDir(), newTestClock().Now, summarizer, discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() for n := range 9 { _, err = memory.Append("echo", memoryLongText(n)) require.NoError(t, err) } require.NoError(t, memory.Wait(t.Context())) for _, task := range summarizer.Tasks() { require.Less(t, task.Index, 8) } require.False(t, treeIsBuilt(t, memory.tree, 0, 8)) summarizer.SetFailing(false) _, err = memory.Append("user", "retry") require.NoError(t, err) require.NoError(t, memory.Wait(t.Context())) require.True(t, treeIsBuilt(t, memory.tree, 0, 8)) require.True(t, treeIsBuilt(t, memory.tree, 3, 0)) } func TestMemoryMergesAfterBothHalves(t *testing.T) { summarizer := newFakeSummarizer() memory, err := openMemory(t.Context(), t.TempDir(), newTestClock().Now, summarizer, discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() _, err = memory.Append("echo", memoryLongText(0)) require.NoError(t, err) require.NoError(t, memory.Wait(t.Context())) _, err = memory.Append("echo", memoryLongText(1)) require.NoError(t, err) require.NoError(t, memory.Wait(t.Context())) require.Len(t, summarizer.Tasks(), 3) left := treeReadNode(t, memory.tree, 0, 0) right := treeReadNode(t, memory.tree, 0, 1) merge := summarizer.Tasks()[2] require.Equal(t, []string{left.Text, right.Text}, merge.Input) require.Equal(t, "\n0+1|"+left.Text+"\n1+1|"+right.Text+"\n", merge.Context) require.True(t, treeIsBuilt(t, memory.tree, 1, 0)) } func TestMemoryRestartRestoresTheQueue(t *testing.T) { root := t.TempDir() clock := newTestClock() summarizer := newFakeSummarizer() summarizer.SetFailing(true) memory, err := openMemory(t.Context(), root, clock.Now, summarizer, discardLog()) require.NoError(t, err) for n := range 4 { _, err = memory.Append("echo", memoryLongText(n)) require.NoError(t, err) } require.NoError(t, memory.Wait(t.Context())) require.NoError(t, memory.Close()) memory, err = openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() require.NoError(t, memory.Wait(t.Context())) require.True(t, treeIsBuilt(t, memory.tree, 2, 0)) } func TestMemoryBuildsFreeParents(t *testing.T) { memory, err := openMemory(t.Context(), t.TempDir(), newTestClock().Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() for _, text := range []string{"a", "b", "c", "d"} { _, err = memory.Append("user", text) require.NoError(t, err) } node := treeReadNode(t, memory.tree, 2, 0) require.Equal(t, "user: a\nuser: b\nuser: c\nuser: d", node.Text) } func TestMemoryKeepsALongMessageOutOfTheView(t *testing.T) { summarizer := newFakeSummarizer() summarizer.SetFailing(true) memory, err := openMemory(t.Context(), t.TempDir(), newTestClock().Now, summarizer, discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() _, err = memory.Append("echo", memoryLongText(0)) require.NoError(t, err) require.NoError(t, memory.Wait(t.Context())) require.Equal(t, "\n0+1|"+viewPending+"\n", memory.View()) summarizer.SetFailing(false) _, err = memory.Append("user", "again") require.NoError(t, err) require.NoError(t, memory.Wait(t.Context())) task, err := memory.task(NodeKey{0, 0}) require.NoError(t, err) require.Equal(t, "\n0+1|"+fakeLine(task)+"\n1+1|user: again\n", memory.View()) } func TestMemoryCoversMessagesAfterTheSavedView(t *testing.T) { root := t.TempDir() clock := newTestClock() memory, err := openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) _, err = memory.Append("user", "a") require.NoError(t, err) _, err = memory.log.Append("reply", "b") require.NoError(t, err) require.NoError(t, memory.Close()) memory, err = openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() require.Equal(t, "\n0+1|user: a\n1+1|reply: b\n", memory.View()) } func TestMemoryRefusesToRebuildAMissingView(t *testing.T) { root := t.TempDir() clock := newTestClock() memory, err := openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) _, err = memory.Append("user", "a") require.NoError(t, err) require.NoError(t, memory.Close()) require.NoError(t, os.Remove(filepath.Join(root, "view.json"))) _, err = openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.ErrorIs(t, err, errMemoryNoView) } func TestQueuedAppendResumesWithoutADuplicate(t *testing.T) { dir := t.TempDir() memory, err := openMemory(t.Context(), dir, newTestClock().Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) var saved *int first, err := memory.AppendOnce("user", "Record the gate size.", nil, func(id int) error { saved = &id require.Equal(t, 0, memory.log.Len()) return nil }) require.NoError(t, err) require.NoError(t, memory.Close()) memory, err = openMemory(t.Context(), dir, newTestClock().Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() second, err := memory.AppendOnce("user", first.Text, saved, func(int) error { require.FailNow(t, "a recorded message must not be prepared again") return nil }) require.NoError(t, err) require.Equal(t, first, second) require.Equal(t, 1, memory.Len()) } func TestQueuedAppendRetriesAnUnwrittenPreparedMessage(t *testing.T) { memory, err := openMemory(t.Context(), t.TempDir(), newTestClock().Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() failure := errors.New("fake queue write failure") var saved *int _, err = memory.AppendOnce("work", "[Garden] Check the gate.", nil, func(id int) error { saved = &id return failure }) require.ErrorIs(t, err, failure) require.Equal(t, 0, memory.Len()) message, err := memory.AppendOnce("work", "[Garden] Check the gate.", saved, func(id int) error { require.Equal(t, *saved, id) return nil }) require.NoError(t, err) require.Equal(t, 0, message.ID) require.Equal(t, 1, memory.Len()) } func TestQueuedAppendRefusesAnInvalidReceipt(t *testing.T) { memory, err := openMemory(t.Context(), t.TempDir(), newTestClock().Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() _, err = memory.Append("user", "Keep the original.") require.NoError(t, err) testCases := []struct { name string id int cause string }{ { name: "receipt points to different text", id: 0, cause: "does not match", }, { name: "receipt skips a log position", id: 2, cause: "beyond the log", }, { name: "receipt has a negative position", id: -1, cause: "read queued message", }, } for _, tc := range testCases { t.Run(tc.name, func(t *testing.T) { _, err := memory.AppendOnce("user", "Other text.", &tc.id, func(int) error { require.FailNow(t, "an invalid receipt must not be prepared") return nil }) require.ErrorContains(t, err, tc.cause) require.Equal(t, 1, memory.Len()) }) } } // TestReplaySyntheticLog runs a synthetic log through the memory core with the // fake summarizer. After each message it waits for the builds, as a turn does. // It checks that each view stays within its sawtooth, that no view line holds a // whole long message or a part of an unbuilt one, and that a restart in the // middle and at the end loads the same views byte for byte. With -short, the // log is shorter, and the test still requires a batch of the chat view. func TestReplaySyntheticLog(t *testing.T) { start := time.Now() messages := replayMessages if testing.Short() { messages = replayShortMessages } root := t.TempDir() clock := newTestClock() random := rand.New(rand.NewPCG(1, 2)) memory, err := openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) defer func() { require.NoError(t, memory.Close()) }() batches := 0 low := viewChatHigh long := make([]bool, messages) for id := range messages { if id == messages/2 { memory = replayRequireRestart(t, memory, root, clock) } clock.now = clock.now.Add(time.Minute) kind, text := replayMessage(random, id) long[id] = len(kind)+len(": ")+len(text) > nodeBytes lines := replayViewLines(memory) _, err = memory.Append(kind, text) require.NoError(t, err) size, merged := replayRequireViews(t, memory, lines, long, id%1000 == 0) if merged { batches++ low = min(low, size) } require.NoError(t, memory.Wait(t.Context())) } memory = replayRequireRestart(t, memory, root, clock) require.Equal(t, messages, memory.log.Len()) require.Positive(t, batches) require.NoError(t, memory.Wait(t.Context())) t.Logf("%d messages, %d nodes, %d batches in %s", messages, memory.tree.Count(), batches, time.Since(start).Round(time.Second)) t.Logf("chat view: %d lines, %d bytes, lowest after a batch %d bytes", len(memory.chat.lines), memory.chat.Size(), low) t.Logf("compaction view: %d lines, %d bytes", len(memory.cview.lines), memory.cview.Size()) runtime.GC() var stats runtime.MemStats runtime.ReadMemStats(&stats) t.Logf("heap in use: %d KB", stats.HeapInuse/1024) } // memoryLongText returns a text that is too long for its own node. func memoryLongText(n int) string { return fmt.Sprintf("long %d ", n) + strings.Repeat("x", nodeBytes) } // replayRequireRestart closes memory, opens it again, and checks that the views // and view.json stay the same byte for byte. func replayRequireRestart(t *testing.T, memory *Memory, root string, clock *TestClock) *Memory { t.Helper() require.NoError(t, memory.Wait(t.Context())) view := memory.View() context := memory.Context(0, memory.log.Len()) saved, err := os.ReadFile(filepath.Join(root, "view.json")) require.NoError(t, err) require.NoError(t, memory.Close()) memory, err = openMemory(t.Context(), root, clock.Now, newFakeSummarizer(), discardLog()) require.NoError(t, err) require.NoError(t, memory.Wait(t.Context())) require.Equal(t, view, memory.View()) require.Equal(t, context, memory.Context(0, memory.log.Len())) loaded, err := os.ReadFile(filepath.Join(root, "view.json")) require.NoError(t, err) require.Equal(t, saved, loaded) return memory } // replayMessage returns the kind and the text of message id. Most messages are // tool calls and their output. Users write short lines. One output in a hundred // is over 30 KB. func replayMessage(random *rand.Rand, id int) (string, string) { roll := random.IntN(100) switch { case roll < 5: return "user", replayText(id, 10+random.IntN(200)) case roll < 10: return "reply", replayText(id, 100+random.IntN(1500)) case roll < 12: return "note", replayText(id, 50+random.IntN(600)) case roll < 13: return "work", replayText(id, 300+random.IntN(3000)) case roll < 56: return "tool", replayText(id, 60+random.IntN(400)) case roll < 57: return "echo", replayText(id, 30000+random.IntN(30000)) default: return "echo", replayText(id, 20+random.IntN(4000)) } } // replayViewLines returns the number of lines in the chat view of memory. func replayViewLines(memory *Memory) int { memory.mutex.Lock() defer memory.mutex.Unlock() return len(memory.chat.lines) } // replayRequireViews checks the views after a new message, and with full // compares them with the tree. Each view is at most its high limit, and a batch // merges the chat view to its low limit. The leaf of the new message can be // built after the append, so the sizes count it as the pending mark, as the // append did. It returns the size of the chat view and if a batch merged it, // from the number of lines before the message. func replayRequireViews(t *testing.T, memory *Memory, lines int, long []bool, full bool) (int, bool) { t.Helper() memory.mutex.Lock() defer memory.mutex.Unlock() size := replayAppendSize(memory.chat) require.LessOrEqual(t, size, viewChatHigh) require.LessOrEqual(t, replayAppendSize(memory.cview), viewCompactionHigh) merged := len(memory.chat.lines) < lines if merged { require.LessOrEqual(t, size, viewChatLow) } replayRequireOnlySummaries(t, memory, long, full) return size, merged } // replayText returns a text of about size bytes that starts with its message ID // and holds some non-ASCII characters. func replayText(id, size int) string { head := fmt.Sprintf("m%d ", id) words := "ação data line café " return head + strings.Repeat(words, max(1, (size-len(head))/len(words))) } // replayAppendSize returns the size of view with its last line as the pending // mark. func replayAppendSize(view *View) int { last := view.lines[len(view.lines)-1] return view.Size() - view.lineSize(last) + len(viewPending) } // replayRequireOnlySummaries checks each line of the chat view as the model // reads it. A built line holds its node, a long message has a summary of the // fake summarizer and not its text, and an unbuilt line holds only the pending // mark. With full, it also checks that the view keeps the node of each built // line. The caller holds the lock of memory. func replayRequireOnlySummaries(t *testing.T, memory *Memory, long []bool, full bool) { t.Helper() rendered := strings.Split(memory.chat.Render(), "\n") require.Len(t, rendered, len(memory.chat.lines)+2) for n, line := range memory.chat.lines { head := fmt.Sprintf("%d+%d|", line.index<