Retain successful response read failures

internal/llm/response_read_error_test.go

@@ -0,0 +1,48 @@
+package llm
+
+import (
+	"errors"
+	"io"
+	"strings"
+	"testing"
+)
+
+type responseBytesAndError struct {
+	data []byte
+	err  error
+}
+
+func (r *responseBytesAndError) Read(p []byte) (int, error) {
+	if len(r.data) == 0 {
+		return 0, r.err
+	}
+	n := copy(p, r.data)
+	r.data = r.data[n:]
+	if len(r.data) == 0 {
+		return n, r.err
+	}
+	return n, nil
+}
+
+func TestSuccessfulResponseReadFailuresRetainCause(t *testing.T) {
+	cause := errors.New("private reader diagnostic")
+	for _, test := range []struct {
+		name, prefix string
+		together     bool
+	}{
+		{"partial", `{"choices":[`, false},
+		{"complete then failure", successfulProviderResponse, false},
+		{"bytes and error together", successfulProviderResponse, true},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			var reader io.Reader = io.MultiReader(strings.NewReader(test.prefix), failingReader{err: cause})
+			if test.together {
+				reader = &responseBytesAndError{data: []byte(test.prefix), err: cause}
+			}
+			_, err := decodeOpenAIChatResponse(reader)
+			if !errors.Is(err, ErrTransport) || !errors.Is(err, cause) || errors.Is(err, ErrMalformedResponse) || strings.Contains(err.Error(), cause.Error()) {
+				t.Fatalf("read failure = %v", err)
+			}
+		})
+	}
+}