返回 DeepSeek-Reasonix
memorybench_test.go
根目录 / cmd / e2ebench / memorybench_test.go
1 package main
2
3 import (
4 "encoding/json"
5 "os"
6 "path/filepath"
7 "strings"
8 "testing"
9 )
10
11 func TestScanMemoryRecallCountsAndPointOfUse(t *testing.T) {
12 dir := t.TempDir()
13 path := filepath.Join(dir, "run.trajectory.jsonl")
14 lines := []string{
15 `{"seq":1,"event":{"kind":"tool_result","tool":{"args":"{\"command\":\"make check-fast --tag=MEMKEY-EARLY\"}"}}}`,
16 `{"seq":2,"memory_recall":{"hits":[{"id":"a"},{"id":"b"}],"used_chars":420}}`,
17 `{"seq":3,"event":{"kind":"tool_result","tool":{"args":"{\"path\":\"answer.txt\",\"content\":\"make check-fast --tag=MEMKEY-USED\"}"}}}`,
18 `{"seq":4,"memory_recall":{"suppressed":"generic user turn"}}`,
19 `{"seq":5,"event":{"kind":"text","text":"done, MEMKEY-TEXT too"}}`,
20 }
21 if err := os.WriteFile(path, []byte(strings.Join(lines, "\n")+"\n"), 0o644); err != nil {
22 t.Fatal(err)
23 }
24 stats := scanMemoryRecall(path, []string{"MEMKEY-USED", "MEMKEY-TEXT", "MEMKEY-EARLY", "MEMKEY-NEVER"}, false)
25 if stats.RecallEvents != 1 || stats.RecallHits != 2 || stats.RecallChars != 420 || stats.Suppressed != 1 {
26 t.Fatalf("stats = %+v, want 1 event / 2 hits / 420 chars / 1 suppressed", stats)
27 }
28 // MEMKEY-EARLY appears only BEFORE the recall: not point-of-use evidence.
29 if stats.MarkersUsed != 2 {
30 t.Fatalf("markers used = %d, want 2 (post-recall args + answer text only)", stats.MarkersUsed)
31 }
32 }
33
34 func TestMemoryUtilitySectionPairsArms(t *testing.T) {
35 dir := t.TempDir()
36 on := []result{
37 {task: task{ID: "helped"}, Passed: true, MemoryRecallEvents: 1, MemoryRecallChars: 300},
38 {task: task{ID: "hurt"}, Passed: false, MemoryRecallEvents: 1, MemoryRecallChars: 500},
39 {task: task{ID: "same"}, Passed: true, MemoryRecallEvents: 1, MemoryRecallChars: 100},
40 }
41 off := []result{
42 {task: task{ID: "helped"}, Passed: false},
43 {task: task{ID: "hurt"}, Passed: true},
44 {task: task{ID: "same"}, Passed: true},
45 }
46 onPath, offPath := filepath.Join(dir, "on.json"), filepath.Join(dir, "off.json")
47 for path, rows := range map[string][]result{onPath: on, offPath: off} {
48 data, _ := json.Marshal(rows)
49 if err := os.WriteFile(path, data, 0o644); err != nil {
50 t.Fatal(err)
51 }
52 }
53 section := memoryUtilitySection(offPath, onPath) // order must not matter
54 for _, want := range []string{"Memory utility", "3 paired tasks", "helpful** 1", "harmful** 1", "helped", "hurt"} {
55 if !strings.Contains(section, want) {
56 t.Fatalf("section missing %q:\n%s", want, section)
57 }
58 }
59 }
60
61 func TestSeedTaskMemoryBuildsIsolatedStateRoot(t *testing.T) {
62 taskDir := t.TempDir()
63 work := t.TempDir()
64 for _, seed := range []string{"project/fact.md", "global/pref.md"} {
65 p := filepath.Join(taskDir, "memory", seed)
66 if err := os.MkdirAll(filepath.Dir(p), 0o755); err != nil {
67 t.Fatal(err)
68 }
69 if err := os.WriteFile(p, []byte("---\nname: x\ndescription: y\n---\n\nbody\n"), 0o644); err != nil {
70 t.Fatal(err)
71 }
72 }
73 env, err := seedTaskMemory(taskDir, work)
74 if err != nil || len(env) != 1 || !strings.HasPrefix(env[0], "REASONIX_STATE_HOME=") {
75 t.Fatalf("env = %v err = %v", env, err)
76 }
77 stateHome := strings.TrimPrefix(env[0], "REASONIX_STATE_HOME=")
78 if _, err := os.Stat(filepath.Join(stateHome, "memory", "global", "pref.md")); err != nil {
79 t.Fatalf("global seed missing: %v", err)
80 }
81 matches, _ := filepath.Glob(filepath.Join(stateHome, "projects", "*", "memory", "fact.md"))
82 if len(matches) != 1 {
83 t.Fatalf("project seed not under the work dir's slug: %v", matches)
84 }
85
86 if env, err := seedTaskMemory(t.TempDir(), work); err != nil || env != nil {
87 t.Fatalf("task without seeds must be a no-op, got %v %v", env, err)
88 }
89 }
90
90 lines GO