Direkte DeepSeek- und Mistral-Anbindung mit versionierten Aktionsprofilen, Budgetreservierung, Tokenabrechnung und administrativen Qualitätsprüfungen ersetzen die Flowise-Abhängigkeit im regulären Betrieb. 840 geprüfte A1-Aufgaben mit geschützter Testauswertung und konkreten offenen Arbeitsaufträgen ergänzen den Kurs. Inhalts- und Sprachbewertung bleiben von Azure-Aussprachewerten getrennt; historische Aufgaben und Antworten bleiben erhalten. Echte Wort- und Phonemmessungen werden mit Herkunft, Aktualität und Nachweisdichte gebündelt. Go bestimmt Trainingsziele und Wiederholungen, das Modell formuliert dazu passende Nachsprechübungen. Profilansicht und Auswahlgründe werden angezeigt. Validierung: Go-Race-Tests mit lokaler PostgreSQL-Integration, go vet, Web-Build, Audit aller 840 Aufgaben und drei Browser-Smokes mit lokalen Modell-/Speech-Fixtures bestanden. Keine Produktionsänderungen oder Live-Aufrufe.
181 lines
6.4 KiB
Go
181 lines
6.4 KiB
Go
package services
|
|
|
|
import (
|
|
"context"
|
|
"encoding/json"
|
|
"fmt"
|
|
"net/http"
|
|
"net/http/httptest"
|
|
"regexp"
|
|
"strings"
|
|
"testing"
|
|
|
|
"speech-learning-api/internal/config"
|
|
"speech-learning-api/internal/domain"
|
|
)
|
|
|
|
func pronunciationFixture() domain.GeneratedExercise {
|
|
return domain.GeneratedExercise{ExerciseType: "repeat_sentence", TargetLanguage: "en-US", CEFRLevel: "A1", ReferenceText: "I eat an apple.", TTSText: "I eat an apple.", DisplayPrompt: "Sprich den Satz nach.", FocusPoints: []string{"apple"}, SuccessCriteria: []string{"Sprich alle Wörter deutlich."}, ReasonForChoice: "Allgemeine A1-Übung ohne belegte Schwächen."}
|
|
}
|
|
|
|
func TestPronunciationSchemaControlCharacterPattern(t *testing.T) {
|
|
a := NewAIService(nil, config.Config{})
|
|
var schema struct {
|
|
Properties map[string]struct {
|
|
Pattern string `json:"pattern"`
|
|
} `json:"properties"`
|
|
}
|
|
if err := json.Unmarshal([]byte(a.Schemas["next_exercise"]), &schema); err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
pattern, err := regexp.Compile(schema.Properties["reason_for_choice"].Pattern)
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
for _, text := range []string{"Kurze A1-Übung.", "cat /k/", "这是猫。"} {
|
|
if !pattern.MatchString(text) {
|
|
t.Fatalf("valid text rejected: %q", text)
|
|
}
|
|
}
|
|
for _, text := range []string{"", "Zeile\nZeile", "Tab\tWort", "Wort\x7f"} {
|
|
if pattern.MatchString(text) {
|
|
t.Fatalf("invalid text accepted: %q", text)
|
|
}
|
|
}
|
|
}
|
|
|
|
func TestPronunciationPostgresCorrectsSubstitutedTrainingWord(t *testing.T) {
|
|
for _, corrected := range []bool{false, true} {
|
|
t.Run(fmt.Sprint(corrected), func(t *testing.T) {
|
|
a, _, user := setupAI(t)
|
|
ctx := context.Background()
|
|
if _, err := a.Repo.SaveProfile(ctx, user, aiTestProfile("next_exercise")); err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
if err := a.Repo.SaveBudget(ctx, user, 10000000, 0); err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
calls := 0
|
|
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
|
calls++
|
|
var body struct {
|
|
Messages []struct {
|
|
Content string `json:"content"`
|
|
} `json:"messages"`
|
|
}
|
|
json.NewDecoder(r.Body).Decode(&body)
|
|
var message struct {
|
|
Input struct {
|
|
Words []string `json:"required_practice_words"`
|
|
Error string `json:"technical_error"`
|
|
} `json:"input"`
|
|
}
|
|
json.Unmarshal([]byte(body.Messages[1].Content), &message)
|
|
if strings.Join(message.Input.Words, ",") != "cat" {
|
|
t.Error("training target not sent")
|
|
}
|
|
if calls == 2 && !strings.Contains(message.Input.Error, `required practice word "cat"`) {
|
|
t.Error("missing focused correction", message.Input.Error)
|
|
}
|
|
ex := pronunciationFixture()
|
|
ex.ReferenceText = "The dog sits on the rug."
|
|
ex.FocusPoints = []string{"dog", "sits"}
|
|
if calls == 2 && corrected {
|
|
ex.ReferenceText = "My cat is very small."
|
|
ex.FocusPoints = []string{"cat"}
|
|
}
|
|
ex.TTSText = ex.ReferenceText
|
|
json.NewEncoder(w).Encode(aiCompletion(ex, true))
|
|
}))
|
|
defer server.Close()
|
|
a.URLs["mistral"] = server.URL
|
|
ex, err := a.NextExercise(ctx, domain.GeneratedExerciseRequest{UserID: user, TargetLanguage: "en-US", CEFRLevel: "A1", Progress: domain.ProgressSummary{RecentAttempts: []domain.Attempt{{TargetLanguage: "en-US", WordScores: []domain.WordScore{{Word: "cat", Accuracy: 29}}}}}})
|
|
if calls != 2 {
|
|
t.Fatalf("retry count %d", calls)
|
|
}
|
|
if corrected {
|
|
if err != nil || ex.Source != "ai" {
|
|
t.Fatal("corrected exercise rejected", err)
|
|
}
|
|
} else if err == nil || !strings.Contains(err.Error(), `required practice word "cat"`) {
|
|
t.Fatal("wrong training accepted", err)
|
|
}
|
|
})
|
|
}
|
|
}
|
|
|
|
func TestPronunciationReasonValidationExplainsExactViolation(t *testing.T) {
|
|
input := domain.GeneratedExerciseRequest{TargetLanguage: "en-US", CEFRLevel: "A1"}
|
|
for _, tc := range []struct{ reason, want string }{
|
|
{"", "reason_for_choice is required"},
|
|
{strings.Repeat("ü", 301), "reason_for_choice exceeds maximum length of 300 characters"},
|
|
{"Ein Satz.\nNoch ein Satz.", "reason_for_choice must be a single line"},
|
|
} {
|
|
ex := pronunciationFixture()
|
|
ex.ReasonForChoice = tc.reason
|
|
if got := generatedExerciseValidationError(ex, input); !strings.Contains(got, tc.want) {
|
|
t.Fatalf("wrong diagnosis: %s", got)
|
|
}
|
|
}
|
|
ex := pronunciationFixture()
|
|
ex.ReasonForChoice = strings.Repeat("ü", 300)
|
|
if got := generatedExerciseValidationError(ex, input); got != "" {
|
|
t.Fatal("valid Unicode boundary rejected", got)
|
|
}
|
|
}
|
|
|
|
func TestPronunciationPostgresRepairsExactFieldContract(t *testing.T) {
|
|
for _, nativeSchema := range []bool{false, true} {
|
|
t.Run(map[bool]string{false: "json_prompt", true: "native_schema"}[nativeSchema], func(t *testing.T) {
|
|
a, _, user := setupAI(t)
|
|
ctx := context.Background()
|
|
p := aiTestProfile("next_exercise")
|
|
p.JSONSchema = nativeSchema
|
|
if _, err := a.Repo.SaveProfile(ctx, user, p); err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
if err := a.Repo.SaveBudget(ctx, user, 10000000, 0); err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
calls := 0
|
|
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
|
calls++
|
|
var body struct {
|
|
Messages []struct {
|
|
Content string `json:"content"`
|
|
} `json:"messages"`
|
|
Format map[string]any `json:"response_format"`
|
|
}
|
|
if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
|
|
t.Error(err)
|
|
}
|
|
if nativeSchema {
|
|
if body.Format["type"] != "json_schema" {
|
|
t.Error("native contract not sent")
|
|
}
|
|
} else if !strings.Contains(body.Messages[0].Content, `"maxLength": 300`) {
|
|
t.Error("field limits absent from prompt")
|
|
}
|
|
if calls == 2 && !strings.Contains(body.Messages[1].Content, "reason_for_choice exceeds maximum length of 300 characters") {
|
|
t.Error("retry lost exact correction")
|
|
}
|
|
ex := pronunciationFixture()
|
|
if calls == 1 {
|
|
ex.ReasonForChoice = strings.Repeat("x", 301)
|
|
}
|
|
json.NewEncoder(w).Encode(aiCompletion(ex, true))
|
|
}))
|
|
defer server.Close()
|
|
a.URLs["mistral"] = server.URL
|
|
ex, err := a.NextExercise(ctx, domain.GeneratedExerciseRequest{UserID: user, TargetLanguage: "en-US", CEFRLevel: "A1", UILanguage: "de-DE"})
|
|
if err != nil || ex.Source != "ai" || calls != 2 {
|
|
t.Fatalf("correction failed: %+v %v calls=%d", ex, err, calls)
|
|
}
|
|
usage, err := a.Repo.Usage(ctx, domain.AIUsageFilter{Action: "next_exercise"})
|
|
if err != nil || usage.Total != 2 {
|
|
t.Fatalf("retry accounting lost: %v %v", usage, err)
|
|
}
|
|
})
|
|
}
|
|
}
|