fix: retry interrupted serial drain (#33)

internal/modem/session.go

@@ -8,6 +8,7 @@ import (
 	"io"
 	"strings"
 	"sync"
+	"syscall"
 	"time"
 )
 
@@ -151,7 +152,7 @@ func (session *Session) executeLocked(ctx context.Context, command string) (Resp
 		session.poisonLocked()
 		return response, fmt.Errorf("write %s: %w", command, err)
 	}
-	if err := session.transport.Drain(); err != nil {
+	if err := drainTransport(ctx, session.transport); err != nil {
 		session.poisonLocked()
 		return response, fmt.Errorf("drain %s: %w", command, err)
 	}
@@ -178,7 +179,7 @@ func (session *Session) executePromptLocked(
 		session.poisonLocked()
 		return response, fmt.Errorf("write %s: %w", command, err)
 	}
-	if err := session.transport.Drain(); err != nil {
+	if err := drainTransport(ctx, session.transport); err != nil {
 		session.poisonLocked()
 		return response, fmt.Errorf("drain %s: %w", command, err)
 	}
@@ -203,7 +204,7 @@ func (session *Session) executePromptLocked(
 		response.Duration = time.Since(started)
 		return response, fmt.Errorf("terminate %s payload: %w", command, err)
 	}
-	if err := session.transport.Drain(); err != nil {
+	if err := drainTransport(ctx, session.transport); err != nil {
 		session.poisonLocked()
 		response.Duration = time.Since(started)
 		return response, fmt.Errorf("drain %s payload: %w", command, err)
@@ -211,6 +212,21 @@ func (session *Session) executePromptLocked(
 	return session.readFinalLocked(ctx, started, command, string(payload), response)
 }
 
+// drainTransport retries tcdrain/TCSBRK when the kernel interrupts it with a
+// signal. go.bug.st/serial already retries EINTR for Read, but its Linux
+// Drain implementation currently returns the transient error directly.
+func drainTransport(ctx context.Context, transport Transport) error {
+	for {
+		err := transport.Drain()
+		if !errors.Is(err, syscall.EINTR) {
+			return err
+		}
+		if err := ctx.Err(); err != nil {
+			return err
+		}
+	}
+}
+
 func (session *Session) readFinalLocked(
 	ctx context.Context,
 	started time.Time,

internal/modem/session_test.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"io"
 	"sync"
+	"syscall"
 	"testing"
 	"time"
 )
@@ -27,6 +28,8 @@ type transcriptTransport struct {
 	unexpected    error
 	writePartial  bool
 	writeEvents   chan string
+	drainErrors   []error
+	drainCount    int
 }
 
 func (transport *transcriptTransport) Write(payload []byte) (int, error) {
@@ -114,7 +117,17 @@ func (transport *transcriptTransport) Read(buffer []byte) (int, error) {
 	return 0, nil
 }
 
-func (transport *transcriptTransport) Drain() error { return nil }
+func (transport *transcriptTransport) Drain() error {
+	transport.mu.Lock()
+	defer transport.mu.Unlock()
+	transport.drainCount++
+	if len(transport.drainErrors) == 0 {
+		return nil
+	}
+	err := transport.drainErrors[0]
+	transport.drainErrors = transport.drainErrors[1:]
+	return err
+}
 
 func (transport *transcriptTransport) ResetInputBuffer() error {
 	transport.mu.Lock()
@@ -138,6 +151,31 @@ func (transport *transcriptTransport) Close() error {
 	return nil
 }
 
+func TestSessionRetriesInterruptedDrain(t *testing.T) {
+	transport := &transcriptTransport{
+		steps: []transportStep{{
+			write:  "AT+CSQ\r",
+			chunks: []string{"\r\nAT+CSQ\r\n+CSQ: 24,99\r\nOK\r\n"},
+		}},
+		drainErrors: []error{syscall.EINTR},
+	}
+	session, err := NewSession(transport, SessionOptions{})
+	if err != nil {
+		t.Fatalf("NewSession() error = %v", err)
+	}
+
+	response, err := session.Execute(context.Background(), "AT+CSQ")
+	if err != nil {
+		t.Fatalf("Execute() error = %v", err)
+	}
+	if response.Final != "OK" {
+		t.Fatalf("response final = %q", response.Final)
+	}
+	if transport.drainCount != 2 {
+		t.Fatalf("Drain() calls = %d, want 2", transport.drainCount)
+	}
+}
+
 func TestSessionSeparatesInterleavedURCs(t *testing.T) {
 	transport := &transcriptTransport{steps: []transportStep{{
 		write: "AT+CSQ\r",