a06f331eb8
CI / benchmark (push) Has been skipped
install-script / posix-syntax (push) Successful in 6m1s
CI / build-onnx (push) Failing after 6m43s
init-smoke / dry-run (push) Failing after 15m57s
security / govulncheck (push) Has been cancelled
security / trivy-fs (push) Has been cancelled
CI / test (1.26, ubuntu-latest) (push) Has been cancelled
Scorecard supply-chain security / Scorecard analysis (push) Has been cancelled
CI / test (1.26, macos-latest) (push) Has been cancelled
CI / build-windows (push) Has been cancelled
CI / lint (push) Has been cancelled
install-script / powershell-syntax (push) Has been cancelled
install-script / install (macos-14) (push) Has been cancelled
install-script / install (ubuntu-latest) (push) Has been cancelled
290 lines
8.6 KiB
Go
290 lines
8.6 KiB
Go
package claudecli
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
"os"
|
|
"os/exec"
|
|
"path/filepath"
|
|
"strings"
|
|
"testing"
|
|
"time"
|
|
|
|
"github.com/zzet/gortex/internal/llm"
|
|
)
|
|
|
|
// fakeClaude writes a tiny shell script that impersonates the
|
|
// `claude` CLI: it echoes the script-baked stdout payload, optionally
|
|
// captures the args + stdin into sidecar files for assertions, and
|
|
// exits with the script-baked exit code. The returned path is the
|
|
// absolute path to the script.
|
|
//
|
|
// A shell-script fake is the smallest portable surface that exercises
|
|
// every code path the provider has — arg construction, stdin
|
|
// piping, exit-status handling, stderr capture — without pulling in
|
|
// the full TestHelperProcess re-exec dance.
|
|
func fakeClaude(t *testing.T, dir string, opts fakeOpts) string {
|
|
t.Helper()
|
|
if dir == "" {
|
|
dir = t.TempDir()
|
|
}
|
|
script := filepath.Join(dir, "fake-claude.sh")
|
|
|
|
argsLog := filepath.Join(dir, "args.txt")
|
|
stdinLog := filepath.Join(dir, "stdin.txt")
|
|
stderrPayload := strings.ReplaceAll(opts.stderr, "'", "'\\''")
|
|
stdoutPayload := strings.ReplaceAll(opts.stdout, "'", "'\\''")
|
|
|
|
body := "#!/bin/sh\n" +
|
|
"set -e\n" +
|
|
"printf '%s\\n' \"$@\" > '" + argsLog + "'\n" +
|
|
"cat > '" + stdinLog + "'\n"
|
|
if opts.sleep > 0 {
|
|
body += fmt.Sprintf("sleep %d\n", int(opts.sleep.Seconds()+1))
|
|
}
|
|
if stderrPayload != "" {
|
|
body += "printf '%s' '" + stderrPayload + "' >&2\n"
|
|
}
|
|
if stdoutPayload != "" {
|
|
body += "printf '%s' '" + stdoutPayload + "'\n"
|
|
}
|
|
if opts.exitCode != 0 {
|
|
body += fmt.Sprintf("exit %d\n", opts.exitCode)
|
|
}
|
|
|
|
if err := os.WriteFile(script, []byte(body), 0o700); err != nil {
|
|
t.Fatalf("write fake script: %v", err)
|
|
}
|
|
return script
|
|
}
|
|
|
|
type fakeOpts struct {
|
|
stdout string
|
|
stderr string
|
|
exitCode int
|
|
sleep time.Duration
|
|
}
|
|
|
|
func readSidecar(t *testing.T, scriptPath, name string) string {
|
|
t.Helper()
|
|
b, err := os.ReadFile(filepath.Join(filepath.Dir(scriptPath), name))
|
|
if err != nil {
|
|
t.Fatalf("read %s: %v", name, err)
|
|
}
|
|
return string(b)
|
|
}
|
|
|
|
func TestNew_BinaryNotFound(t *testing.T) {
|
|
if _, err := New(llm.ClaudeCLIConfig{Binary: "claude-nonexistent-zzzzz"}); err == nil {
|
|
t.Fatal("expected error when binary is not on PATH")
|
|
}
|
|
}
|
|
|
|
func TestNew_DefaultsBinary(t *testing.T) {
|
|
dir := t.TempDir()
|
|
bin := filepath.Join(dir, "claude")
|
|
if err := os.WriteFile(bin, []byte("#!/bin/sh\nexit 0\n"), 0o700); err != nil {
|
|
t.Fatalf("write fake binary: %v", err)
|
|
}
|
|
// Stash a fake `claude` on PATH so the default binary name resolves.
|
|
t.Setenv("PATH", dir+string(os.PathListSeparator)+os.Getenv("PATH"))
|
|
|
|
p, err := New(llm.ClaudeCLIConfig{})
|
|
if err != nil {
|
|
t.Fatalf("New: %v", err)
|
|
}
|
|
defer p.Close()
|
|
if p.Name() != "claudecli" {
|
|
t.Errorf("Name()=%q want claudecli", p.Name())
|
|
}
|
|
}
|
|
|
|
func TestComplete_FreeformSuccess(t *testing.T) {
|
|
script := fakeClaude(t, "", fakeOpts{stdout: "hello world"})
|
|
|
|
p, err := New(llm.ClaudeCLIConfig{Binary: script})
|
|
if err != nil {
|
|
t.Fatalf("New: %v", err)
|
|
}
|
|
defer p.Close()
|
|
|
|
resp, err := p.Complete(context.Background(), llm.CompletionRequest{
|
|
Messages: []llm.Message{
|
|
{Role: llm.RoleSystem, Content: "be terse"},
|
|
{Role: llm.RoleUser, Content: "hi"},
|
|
},
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("Complete: %v", err)
|
|
}
|
|
if resp.Text != "hello world" {
|
|
t.Errorf("text=%q want hello world", resp.Text)
|
|
}
|
|
|
|
gotArgs := readSidecar(t, script, "args.txt")
|
|
for _, want := range []string{"--print", "--output-format", "text", "--append-system-prompt", "be terse"} {
|
|
if !strings.Contains(gotArgs, want) {
|
|
t.Errorf("args missing %q\nargs=\n%s", want, gotArgs)
|
|
}
|
|
}
|
|
stdin := readSidecar(t, script, "stdin.txt")
|
|
if !strings.Contains(stdin, "User: hi") {
|
|
t.Errorf("stdin missing user turn:\n%s", stdin)
|
|
}
|
|
if strings.Contains(stdin, "be terse") {
|
|
t.Error("system content must travel via --append-system-prompt, not stdin")
|
|
}
|
|
}
|
|
|
|
func TestComplete_PassesModel(t *testing.T) {
|
|
script := fakeClaude(t, "", fakeOpts{stdout: "ok"})
|
|
|
|
p, _ := New(llm.ClaudeCLIConfig{Binary: script, Model: "sonnet"})
|
|
defer p.Close()
|
|
|
|
if _, err := p.Complete(context.Background(), llm.CompletionRequest{
|
|
Messages: []llm.Message{{Role: llm.RoleUser, Content: "hi"}},
|
|
}); err != nil {
|
|
t.Fatalf("Complete: %v", err)
|
|
}
|
|
args := readSidecar(t, script, "args.txt")
|
|
if !strings.Contains(args, "--model\nsonnet") && !strings.Contains(args, "--model sonnet") {
|
|
t.Errorf("args missing --model sonnet:\n%s", args)
|
|
}
|
|
}
|
|
|
|
func TestComplete_StructuredExtractsJSON(t *testing.T) {
|
|
wrapped := "Sure, here you go:\n```json\n{\"terms\":[\"bcrypt\",\"argon2\"]}\n```\n"
|
|
script := fakeClaude(t, "", fakeOpts{stdout: wrapped})
|
|
|
|
p, _ := New(llm.ClaudeCLIConfig{Binary: script})
|
|
defer p.Close()
|
|
|
|
resp, err := p.Complete(context.Background(), llm.CompletionRequest{
|
|
Messages: []llm.Message{{Role: llm.RoleUser, Content: "expand 'password hashing'"}},
|
|
Shape: llm.ShapeExpandTerms,
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("Complete: %v", err)
|
|
}
|
|
if resp.Text != `{"terms":["bcrypt","argon2"]}` {
|
|
t.Errorf("text=%q want the unwrapped JSON object", resp.Text)
|
|
}
|
|
|
|
args := readSidecar(t, script, "args.txt")
|
|
if !strings.Contains(args, "JSON Schema") {
|
|
t.Errorf("structured request must inject a JSON Schema rider; args=\n%s", args)
|
|
}
|
|
}
|
|
|
|
func TestComplete_StructuredNoJSONErrors(t *testing.T) {
|
|
script := fakeClaude(t, "", fakeOpts{stdout: "I cannot help with that."})
|
|
|
|
p, _ := New(llm.ClaudeCLIConfig{Binary: script})
|
|
defer p.Close()
|
|
|
|
if _, err := p.Complete(context.Background(), llm.CompletionRequest{
|
|
Messages: []llm.Message{{Role: llm.RoleUser, Content: "x"}},
|
|
Shape: llm.ShapeExpandTerms,
|
|
}); err == nil {
|
|
t.Fatal("expected error when structured response carried no JSON")
|
|
}
|
|
}
|
|
|
|
func TestComplete_NonZeroExit(t *testing.T) {
|
|
script := fakeClaude(t, "", fakeOpts{exitCode: 2, stderr: "auth required"})
|
|
|
|
p, _ := New(llm.ClaudeCLIConfig{Binary: script})
|
|
defer p.Close()
|
|
|
|
_, err := p.Complete(context.Background(), llm.CompletionRequest{
|
|
Messages: []llm.Message{{Role: llm.RoleUser, Content: "hi"}},
|
|
})
|
|
if err == nil {
|
|
t.Fatal("expected error for non-zero exit")
|
|
}
|
|
if !strings.Contains(err.Error(), "auth required") {
|
|
t.Errorf("error should include stderr snippet; got: %v", err)
|
|
}
|
|
}
|
|
|
|
func TestComplete_EmptyResponseErrors(t *testing.T) {
|
|
script := fakeClaude(t, "", fakeOpts{stdout: ""})
|
|
|
|
p, _ := New(llm.ClaudeCLIConfig{Binary: script})
|
|
defer p.Close()
|
|
|
|
if _, err := p.Complete(context.Background(), llm.CompletionRequest{
|
|
Messages: []llm.Message{{Role: llm.RoleUser, Content: "hi"}},
|
|
}); err == nil {
|
|
t.Fatal("expected error for empty response")
|
|
}
|
|
}
|
|
|
|
func TestComplete_ContextCancellation(t *testing.T) {
|
|
script := fakeClaude(t, "", fakeOpts{stdout: "late", sleep: 2 * time.Second})
|
|
|
|
p, _ := New(llm.ClaudeCLIConfig{Binary: script, TimeoutSeconds: 1})
|
|
defer p.Close()
|
|
|
|
if _, err := p.Complete(context.Background(), llm.CompletionRequest{
|
|
Messages: []llm.Message{{Role: llm.RoleUser, Content: "hi"}},
|
|
}); err == nil {
|
|
t.Fatal("expected timeout error")
|
|
}
|
|
}
|
|
|
|
func TestComplete_ExtraArgsForwarded(t *testing.T) {
|
|
script := fakeClaude(t, "", fakeOpts{stdout: "ok"})
|
|
|
|
p, _ := New(llm.ClaudeCLIConfig{Binary: script, Args: []string{"--allowed-tools", ""}})
|
|
defer p.Close()
|
|
|
|
if _, err := p.Complete(context.Background(), llm.CompletionRequest{
|
|
Messages: []llm.Message{{Role: llm.RoleUser, Content: "hi"}},
|
|
}); err != nil {
|
|
t.Fatalf("Complete: %v", err)
|
|
}
|
|
args := readSidecar(t, script, "args.txt")
|
|
if !strings.Contains(args, "--allowed-tools") {
|
|
t.Errorf("args missing --allowed-tools:\n%s", args)
|
|
}
|
|
}
|
|
|
|
func TestFlatten_Roles(t *testing.T) {
|
|
sys, prompt := flatten([]llm.Message{
|
|
{Role: llm.RoleSystem, Content: "rule 1"},
|
|
{Role: llm.RoleSystem, Content: "rule 2"},
|
|
{Role: llm.RoleUser, Content: "q1"},
|
|
{Role: llm.RoleAssistant, Content: "a1"},
|
|
{Role: llm.RoleTool, Content: "[1,2,3]", ToolName: "search_symbols"},
|
|
{Role: llm.RoleUser, Content: "q2"},
|
|
})
|
|
if sys != "rule 1\n\nrule 2" {
|
|
t.Errorf("system=%q", sys)
|
|
}
|
|
for _, want := range []string{"User: q1", "Assistant: a1", "Tool result (search_symbols)", "User: q2"} {
|
|
if !strings.Contains(prompt, want) {
|
|
t.Errorf("prompt missing %q\nprompt=\n%s", want, prompt)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Sanity check: the test helper itself can run a child process. If
|
|
// this fails the environment doesn't support `/bin/sh` and every
|
|
// other subprocess test in this file will skip cleanly.
|
|
func TestHelper_FakeScriptIsExecutable(t *testing.T) {
|
|
script := fakeClaude(t, "", fakeOpts{stdout: "alive"})
|
|
cmd := exec.Command(script, "ping")
|
|
cmd.Stdin = strings.NewReader("")
|
|
out, err := cmd.Output()
|
|
if err != nil {
|
|
// On a CI runner without /bin/sh, surface as skip instead of
|
|
// fail so we don't paint subprocess CI red.
|
|
t.Skipf("cannot exec fake script: %v", err)
|
|
}
|
|
if !strings.Contains(string(out), "alive") {
|
|
t.Errorf("output=%q", out)
|
|
}
|
|
}
|