Files
weatherreporter/internal/promptdebug/debug_writer_test.go

377 lines
14 KiB
Go

//go:build unix
package promptdebug
import (
"fmt"
"os"
"path/filepath"
"strings"
"sync"
"testing"
"time"
"gitea.maximumdirect.net/eric/weatherreporter/internal/promptexec"
"gitea.maximumdirect.net/eric/weatherreporter/internal/report"
"gitea.maximumdirect.net/eric/weatherreporter/internal/testutil"
)
func TestPromptDebugWriterWritesIsolatedArtifacts(t *testing.T) {
root := filepath.Join(t.TempDir(), "operator-debug")
writer, err := NewPromptDebugWriter(root)
if err != nil {
t.Fatalf("NewPromptDebugWriter() error = %v", err)
}
ref := promptDebugRef()
preparationDir, err := writer.WritePreparation(ref, promptDebugPreparationFixture(), &promptexec.PreparationDebug{
RenderedMessages: []promptexec.RenderedMessage{{Role: "system", Content: "Use the supplied weather facts."}},
StructuredSchema: []byte(`{"type":"object","required":["summary"]}`),
Endpoint: "https://direct:resolved-secret-value@llm.example.test/v1/chat?api_key=resolved-secret-value",
ParametersJSON: []byte(`{"temperature":0.2,"api_key":"resolved-secret-value"}`),
})
if err != nil {
t.Fatalf("WritePreparation() error = %v", err)
}
wantDirectory := filepath.Join(root, "daily", "2026-05-29", "run-123")
if preparationDir != wantDirectory {
t.Fatalf("WritePreparation() directory = %q, want %q", preparationDir, wantDirectory)
}
executionDir, err := writer.WriteExecution(ref, promptDebugExecutionFixture())
if err != nil {
t.Fatalf("WriteExecution() error = %v", err)
}
if executionDir != preparationDir {
t.Fatalf("WriteExecution() directory = %q, want %q", executionDir, preparationDir)
}
preparationData := readPromptDebugFile(t, filepath.Join(preparationDir, "preparation.json"))
for _, want := range []string{"weatherreporter.prompt_preparation_debug.v2", "Use the supplied weather facts.", `"type": "object"`, "https://llm.example.test", `"temperature": 0.2`} {
if !strings.Contains(string(preparationData), want) {
t.Fatalf("preparation debug artifact missing %q:\n%s", want, preparationData)
}
}
executionData := readPromptDebugFile(t, filepath.Join(executionDir, "execution.json"))
for _, want := range []string{"weatherreporter.prompt_execution_debug.v2", "Generated forecast prose.", "validation details", `"status": "passed"`} {
if !strings.Contains(string(executionData), want) {
t.Fatalf("execution debug artifact missing %q:\n%s", want, executionData)
}
}
for _, data := range [][]byte{preparationData, executionData} {
if strings.Contains(string(data), "credential") || strings.Contains(string(data), "resolved-secret-value") || strings.Contains(string(data), "dataPackagePath") {
t.Fatalf("debug artifact contains credentials:\n%s", data)
}
}
assertPromptDebugMode(t, root, debugDirectoryMode)
assertPromptDebugMode(t, preparationDir, debugDirectoryMode)
assertPromptDebugMode(t, filepath.Join(preparationDir, "preparation.json"), debugFileMode)
assertPromptDebugMode(t, filepath.Join(executionDir, "execution.json"), debugFileMode)
}
func TestPromptDebugWriterProjectsProviderConfigurationSafely(t *testing.T) {
const marker = "private-debug-marker"
tests := []struct {
name string
endpoint string
}{
{
name: "user info signed path and access key query",
endpoint: "https://operator:" + marker + "@llm.example.test/signed/" + marker + "?access-key=" + marker + "&Signature=" + marker,
},
{
name: "case and separator query aliases",
endpoint: "https://llm.example.test/" + marker + "?X-Amz-Signature=" + marker + "&AUTH=" + marker + "&session_cookie=" + marker,
},
}
parameters := []byte(`{
"temperature": 0.2,
"max_tokens": 400,
"top_p": 0.9,
"timeout_seconds": 30,
"service_tier": "flex",
"reasoning_effort": "high",
"access-key": "private-debug-marker",
"signature": "private-debug-marker",
"auth": "private-debug-marker",
"cookie": "private-debug-marker",
"nested": {"X Api Key": "private-debug-marker"},
"extra_params": {"authorization": "private-debug-marker"}
}`)
for index, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
writer, err := NewPromptDebugWriter(filepath.Join(t.TempDir(), "debug"))
if err != nil {
t.Fatalf("NewPromptDebugWriter() error = %v", err)
}
ref := promptDebugRef()
ref.RunID = fmt.Sprintf("run-%d", index)
directory, err := writer.WritePreparation(ref, promptDebugPreparationFixture(), &promptexec.PreparationDebug{
Endpoint: tt.endpoint,
ParametersJSON: parameters,
})
if err != nil {
t.Fatalf("WritePreparation() error = %v", err)
}
data := readPromptDebugFile(t, filepath.Join(directory, "preparation.json"))
text := string(data)
if strings.Contains(text, marker) || strings.Contains(text, "signed") || strings.Contains(text, "access-key") || strings.Contains(text, "signature") || strings.Contains(text, "cookie") {
t.Fatalf("preparation debug artifact leaked provider configuration:\n%s", data)
}
for _, want := range []string{"https://llm.example.test", `"temperature": 0.2`, `"max_tokens": 400`, `"top_p": 0.9`, `"timeout_seconds": 30`, `"service_tier": "flex"`, `"reasoning_effort": "high"`} {
if !strings.Contains(text, want) {
t.Fatalf("preparation debug artifact missing safe value %q:\n%s", want, data)
}
}
})
}
}
func TestPromptDebugWriterDoesNotProjectParameterValuesInErrors(t *testing.T) {
const marker = "private-debug-marker"
writer, err := NewPromptDebugWriter(filepath.Join(t.TempDir(), "debug"))
if err != nil {
t.Fatalf("NewPromptDebugWriter() error = %v", err)
}
_, err = writer.WritePreparation(promptDebugRef(), promptDebugPreparationFixture(), &promptexec.PreparationDebug{
ParametersJSON: []byte(`{"access-key":"` + marker),
})
if err == nil || strings.Contains(err.Error(), marker) {
t.Fatalf("WritePreparation() error = %v, want safe parameter projection error", err)
}
}
func TestPromptDebugWriterAtomicallyReplacesArtifacts(t *testing.T) {
writer, err := NewPromptDebugWriter(filepath.Join(t.TempDir(), "debug"))
if err != nil {
t.Fatalf("NewPromptDebugWriter() error = %v", err)
}
ref := promptDebugRef()
first := promptDebugPreparationFixture()
first.PromptHash = "old-hash"
if _, err := writer.WritePreparation(ref, first, nil); err != nil {
t.Fatalf("first WritePreparation() error = %v", err)
}
second := promptDebugPreparationFixture()
second.PromptHash = "new-hash"
directory, err := writer.WritePreparation(ref, second, nil)
if err != nil {
t.Fatalf("second WritePreparation() error = %v", err)
}
data := readPromptDebugFile(t, filepath.Join(directory, "preparation.json"))
if strings.Contains(string(data), "old-hash") || !strings.Contains(string(data), "new-hash") {
t.Fatalf("replacement artifact = %s", data)
}
temporary, err := filepath.Glob(filepath.Join(directory, ".preparation.json.*.tmp"))
if err != nil || len(temporary) != 0 {
t.Fatalf("temporary files = %v, %v", temporary, err)
}
}
func TestPromptDebugWriterCreatesSharedMissingAncestorsConcurrently(t *testing.T) {
root := filepath.Join(t.TempDir(), "debug")
writer, err := NewPromptDebugWriter(root)
if err != nil {
t.Fatalf("NewPromptDebugWriter() error = %v", err)
}
const writerCount = 8
start := make(chan struct{})
type writeResult struct {
directory string
err error
}
results := make(chan writeResult, writerCount)
var writers sync.WaitGroup
for index := 0; index < writerCount; index++ {
writers.Add(1)
go func(index int) {
defer writers.Done()
<-start
directory, err := writer.WritePreparation(PromptDebugRef{
ReportID: report.Daily, ValidDate: "2026-05-29", RunID: fmt.Sprintf("run-%02d", index),
}, promptDebugPreparationFixture(), nil)
results <- writeResult{directory: directory, err: err}
}(index)
}
close(start)
finished := make(chan struct{})
go func() {
writers.Wait()
close(finished)
}()
select {
case <-finished:
case <-time.After(5 * time.Second):
t.Fatal("concurrent prompt debug writes did not finish")
}
close(results)
directories := map[string]struct{}{}
for result := range results {
if result.err != nil {
t.Fatalf("WritePreparation() error = %v", result.err)
}
if _, duplicate := directories[result.directory]; duplicate {
t.Fatalf("duplicate debug directory %q", result.directory)
}
directories[result.directory] = struct{}{}
if _, err := os.Stat(filepath.Join(result.directory, "preparation.json")); err != nil {
t.Fatalf("preparation artifact %q: %v", result.directory, err)
}
}
if len(directories) != writerCount {
t.Fatalf("debug directories = %#v", directories)
}
assertPromptDebugMode(t, root, debugDirectoryMode)
assertPromptDebugMode(t, filepath.Join(root, "daily"), debugDirectoryMode)
assertPromptDebugMode(t, filepath.Join(root, "daily", "2026-05-29"), debugDirectoryMode)
for directory := range directories {
assertPromptDebugMode(t, directory, debugDirectoryMode)
assertPromptDebugMode(t, filepath.Join(directory, "preparation.json"), debugFileMode)
}
}
func TestPromptDebugWriterKeepsWritesAnchoredToOpenedRoot(t *testing.T) {
parent := t.TempDir()
root := filepath.Join(parent, "debug")
writer, err := NewPromptDebugWriter(root)
if err != nil {
t.Fatalf("NewPromptDebugWriter() error = %v", err)
}
defer writer.Close()
anchoredRoot := filepath.Join(parent, "anchored-debug")
outside := filepath.Join(parent, "outside")
if err := os.Mkdir(outside, debugDirectoryMode); err != nil {
t.Fatalf("create outside directory: %v", err)
}
if err := os.Rename(root, anchoredRoot); err != nil {
t.Fatalf("replace opened root: %v", err)
}
testutil.RequireSymlink(t, outside, root)
ref := promptDebugRef()
ref.RunID = "run-anchored"
if _, err := writer.WritePreparation(ref, promptDebugPreparationFixture(), nil); err != nil {
t.Fatalf("WritePreparation() error = %v", err)
}
anchoredArtifact := filepath.Join(anchoredRoot, "daily", "2026-05-29", "run-anchored", "preparation.json")
if _, err := os.Stat(anchoredArtifact); err != nil {
t.Fatalf("anchored preparation artifact: %v", err)
}
entries, err := os.ReadDir(outside)
if err != nil {
t.Fatalf("read outside directory: %v", err)
}
if len(entries) != 0 {
t.Fatalf("outside directory received debug artifacts: %#v", entries)
}
}
func TestPromptDebugWriterDisabledDoesNotAccessFilesystem(t *testing.T) {
writer, err := NewPromptDebugWriter("")
if err != nil {
t.Fatalf("NewPromptDebugWriter() error = %v", err)
}
if writer.Enabled() {
t.Fatal("disabled writer reports enabled")
}
directory, err := writer.WritePreparation(PromptDebugRef{ReportID: report.ID("../unsafe")}, promptDebugPreparationFixture(), nil)
if err != nil || directory != "" {
t.Fatalf("disabled WritePreparation() = %q, %v", directory, err)
}
}
func TestPromptDebugWriterRejectsUnsafeRootsAndReferences(t *testing.T) {
root := t.TempDir()
if _, err := NewPromptDebugWriter("relative-debug"); err == nil {
t.Fatal("NewPromptDebugWriter(relative) error = nil")
}
nonDirectory := filepath.Join(root, "not-a-directory")
if err := os.WriteFile(nonDirectory, []byte("x"), 0o600); err != nil {
t.Fatalf("write non-directory root: %v", err)
}
if _, err := NewPromptDebugWriter(nonDirectory); err == nil {
t.Fatal("NewPromptDebugWriter(file) error = nil")
}
target := filepath.Join(root, "target")
link := filepath.Join(root, "root-link")
if err := os.Mkdir(target, debugDirectoryMode); err != nil {
t.Fatalf("create root symlink target: %v", err)
}
testutil.RequireSymlink(t, target, link)
if _, err := NewPromptDebugWriter(link); err == nil {
t.Fatal("NewPromptDebugWriter(symlink) error = nil")
}
writer, err := NewPromptDebugWriter(filepath.Join(root, "debug"))
if err != nil {
t.Fatalf("NewPromptDebugWriter() error = %v", err)
}
for _, ref := range []PromptDebugRef{
{ReportID: report.ID("../daily"), ValidDate: "2026-05-29", RunID: "run-123"},
{ReportID: report.Daily, ValidDate: "2026/05/29", RunID: "run-123"},
{ReportID: report.Daily, ValidDate: "2026-05-29", RunID: "/run-123"},
{ReportID: report.Daily, ValidDate: "not-a-date", RunID: "run-123"},
} {
if _, err := writer.WritePreparation(ref, promptDebugPreparationFixture(), nil); err == nil {
t.Fatalf("WritePreparation(%#v) error = nil", ref)
}
}
outside := filepath.Join(root, "outside")
if err := os.Mkdir(outside, debugDirectoryMode); err != nil {
t.Fatalf("create symlink component target: %v", err)
}
testutil.RequireSymlink(t, outside, filepath.Join(root, "debug", "daily"))
if _, err := writer.WritePreparation(promptDebugRef(), promptDebugPreparationFixture(), nil); err == nil {
t.Fatal("WritePreparation(symlink component) error = nil")
}
}
func promptDebugRef() PromptDebugRef {
return PromptDebugRef{ReportID: report.Daily, ValidDate: "2026-05-29", RunID: "run-123"}
}
func promptDebugPreparationFixture() promptexec.Preparation {
startedAt := time.Date(2026, 5, 29, 12, 0, 0, 0, time.UTC)
return promptexec.Preparation{
PromptID: "weather.daily", PromptVersion: "v1", PromptHash: "prompt-hash", RenderedPromptHash: "rendered-hash",
InputHashes: map[string]string{"data_package": "input-hash"}, ProfileID: "local", BackendID: "local", ModelName: "weather-model",
Output: promptexec.OutputContract{Format: "json_schema", ValidationMode: "strict", SchemaPath: "schemas/daily.json"},
StartedAt: startedAt, EndedAt: startedAt.Add(time.Second), Duration: time.Second,
}
}
func promptDebugExecutionFixture() promptexec.Execution {
startedAt := time.Date(2026, 5, 29, 12, 0, 0, 0, time.UTC)
return promptexec.Execution{
RunID: "run-123", PromptID: "weather.daily", PromptVersion: "v1", PromptHash: "prompt-hash", RenderedPromptHash: "rendered-hash",
InputHashes: map[string]string{"data_package": "input-hash"}, ProfileID: "local", BackendID: "local", ModelName: "weather-model",
GeneratedHash: "generated-hash", Usage: promptexec.TokenUsage{PromptTokens: 10, CompletionTokens: 5, TotalTokens: 15},
StartedAt: startedAt, EndedAt: startedAt.Add(time.Second), Duration: time.Second,
Validation: promptexec.NewValidation(promptexec.ValidationPassed, "strict", "schemas/daily.json", []string{"validation details"}),
RawOutput: []byte("Generated forecast prose."),
Debug: &promptexec.ExecutionDebug{ValidationDiagnostics: []string{"validation details"}},
}
}
func readPromptDebugFile(t *testing.T, path string) []byte {
t.Helper()
data, err := os.ReadFile(path)
if err != nil {
t.Fatalf("read %q: %v", path, err)
}
return data
}
func assertPromptDebugMode(t *testing.T, path string, want os.FileMode) {
t.Helper()
info, err := os.Stat(path)
if err != nil {
t.Fatalf("stat %q: %v", path, err)
}
if got := info.Mode().Perm(); got != want {
t.Fatalf("mode for %q = %#o, want %#o", path, got, want)
}
}