Changing pybind11::str to only hold PyUnicodeObject (NOT also bytes).

The corresponding behavior changes are captured by changes in the tests. A significant effort was made to keep the test diffs minimal but also comprehensive and easy to read.

Note: Unlike PR #2256 (dropped), this PR only changes exactly one behavior. The two other behavior changes discussed under PR #2256 are avoided here (1. disabling implicit decoding from bytes to unicode; 2. list_caster behavior change). Based on this PR, those can be easily implemented if and when desired.

Author: rwgk

include/pybind11/cast.h

@@ -1631,6 +1631,14 @@ struct pyobject_caster {
 
     template <typename T = type, enable_if_t<std::is_base_of<object, T>::value, int> = 0>
     bool load(handle src, bool /* convert */) {
+#ifndef PYBIND11_DISABLE_IMPLICIT_STR_FROM_BYTES
+        if (std::is_same<T, str>::value && isinstance<bytes>(src)) {
+            PyObject *str_from_bytes = PyUnicode_FromEncodedObject(src.ptr(), "utf-8", nullptr);
+            if (!str_from_bytes) throw error_already_set();
+            value = reinterpret_steal<type>(str_from_bytes);
+            return true;
+        }
+#endif
         if (!isinstance<type>(src))
             return false;
         value = reinterpret_borrow<type>(src);

include/pybind11/pytypes.h

@@ -752,8 +752,6 @@ inline bool PyIterable_Check(PyObject *obj) {
 inline bool PyNone_Check(PyObject *o) { return o == Py_None; }
 inline bool PyEllipsis_Check(PyObject *o) { return o == Py_Ellipsis; }
 
-inline bool PyUnicode_Check_Permissive(PyObject *o) { return PyUnicode_Check(o) || PYBIND11_BYTES_CHECK(o); }
-
 inline bool PyStaticMethod_Check(PyObject *o) { return o->ob_type == &PyStaticMethod_Type; }
 
 class kwargs_proxy : public handle {
@@ -899,7 +897,7 @@ class bytes;
 
 class str : public object {
 public:
-    PYBIND11_OBJECT_CVT(str, object, detail::PyUnicode_Check_Permissive, raw_str)
+    PYBIND11_OBJECT_CVT(str, object, PyUnicode_Check, raw_str)
 
     str(const char *c, size_t n)
         : object(PyUnicode_FromStringAndSize(c, (ssize_t) n), stolen_t{}) {

include/pybind11/stl.h

@@ -144,7 +144,7 @@ template <typename Type, typename Value> struct list_caster {
     using value_conv = make_caster<Value>;
 
     bool load(handle src, bool convert) {
-        if (!isinstance<sequence>(src) || isinstance<str>(src))
+        if (!isinstance<sequence>(src) || isinstance<bytes>(src) || isinstance<str>(src))
             return false;
         auto s = reinterpret_borrow<sequence>(src);
         value.clear();

tests/test_pytypes.py

@@ -190,7 +190,7 @@ def test_constructors():
     """C++ default and converting constructors are equivalent to type calls in Python"""
     types = [bytes, str, bool, int, float, tuple, list, dict, set]
     expected = {t.__name__: t() for t in types}
-    if str is bytes:  # Python 2.
+    if pytest.PY2:
         # Note that bytes.__name__ == 'str' in Python 2.
         # pybind11::str is unicode even under Python 2.
         expected["bytes"] = bytes()
@@ -211,7 +211,7 @@ def test_constructors():
     }
     inputs = {k.__name__: v for k, v in data.items()}
     expected = {k.__name__: k(v) for k, v in data.items()}
-    if str is bytes:  # Similar to the above. See comments above.
+    if pytest.PY2:  # Similar to the above. See comments above.
         inputs["bytes"] = b'41'
         inputs["str"] = 42
         expected["bytes"] = b'41'
@@ -254,13 +254,20 @@ def test_pybind11_str_raw_str():
     valid_orig = u"DZ"
     valid_utf8 = valid_orig.encode("utf-8")
     valid_cvt = cvt(valid_utf8)
-    assert type(valid_cvt) == bytes  # Probably surprising.
-    assert valid_cvt == b'\xc7\xb1'
+    assert type(valid_cvt) is unicode if pytest.PY2 else str  # noqa: F821
+    if pytest.PY2:
+        assert valid_cvt == valid_orig
+    else:
+        assert valid_cvt == u"b'\\xc7\\xb1'"
 
     malformed_utf8 = b'\x80'
-    malformed_cvt = cvt(malformed_utf8)
-    assert type(malformed_cvt) == bytes  # Probably surprising.
-    assert malformed_cvt == b'\x80'
+    if pytest.PY2:
+        with pytest.raises(UnicodeDecodeError):
+            cvt(malformed_utf8)
+    else:
+        malformed_cvt = cvt(malformed_utf8)
+        assert type(malformed_cvt) is unicode if pytest.PY2 else str  # noqa: F821
+        assert malformed_cvt == u"b'\\x80'"
 
 
 def test_implicit_casting():
@@ -397,19 +404,20 @@ def test_isinstance_string_types():
     assert not m.isinstance_pybind11_bytes(u"")
 
     assert m.isinstance_pybind11_str(u"")
-    assert m.isinstance_pybind11_str(b"")  # Probably surprising.
+    assert not m.isinstance_pybind11_str(b"")
 
 
 def test_pass_bytes_or_unicode_to_string_types():
     assert m.pass_to_pybind11_bytes(b"Bytes") == 5
     with pytest.raises(TypeError):
         m.pass_to_pybind11_bytes(u"Str")  # NO implicit encode
 
-    assert m.pass_to_pybind11_str(b"Bytes") == 5
+    assert m.pass_to_pybind11_str(b"Bytes") == 5  # implicit decode
     assert m.pass_to_pybind11_str(u"Str") == 3
 
     assert m.pass_to_std_string(b"Bytes") == 5
     assert m.pass_to_std_string(u"Str") == 3
 
     malformed_utf8 = b"\x80"
-    assert m.pass_to_pybind11_str(malformed_utf8) == 1  # NO decoding error
+    with pytest.raises(UnicodeDecodeError):
+        m.pass_to_pybind11_str(malformed_utf8)