Add a new directive 'str_is_str=True' that keeps unprefixed string li…

…terals and the 'str' builtin type unchanged even when 'language_level=3' is enabled.

See #2565.

CHANGES.rst

@@ -34,6 +34,11 @@ Features added
 * ``cython.inline()`` supports a direct ``language_level`` keyword argument that
   was previously only available via a directive.
 
+* A new directive ``str_is_str=True`` was added that keeps unprefixed string
+  literals as type 'str' in both Py2 and Py3, and the builtin 'str' type unchanged
+  even when ``language_level=3``  is enabled.  This is meant to help user code to
+  migrate to Python 3 semantics without making support for Python 2.x difficult.
+
 * In CPython 3.6 and later, looking up globals in the module dict is almost
   as fast as looking up C globals.
   (Github issue #2313)

Cython/Compiler/Main.py

@@ -94,9 +94,18 @@ def __init__(self, include_directories, compiler_directives, cpp=False,
 
         if language_level is not None:
             self.set_language_level(language_level)
+        if self.compiler_directives.get('str_is_str') is not None:
+            self.set_str_is_str(self.compiler_directives['str_is_str'])
 
         self.gdb_debug_outputwriter = None
 
+    def set_str_is_str(self, str_is_str):
+        from .Future import unicode_literals
+        if str_is_str:
+            self.future_directives.discard(unicode_literals)
+        else:
+            self.future_directives.add(unicode_literals)
+
     def set_language_level(self, level):
         self.language_level = level
         if level >= 3:

Cython/Compiler/Options.py

@@ -198,6 +198,7 @@ def get_directive_defaults():
     'iterable_coroutine': False,  # Make async coroutines backwards compatible with the old asyncio yield-from syntax.
     'c_string_type': 'bytes',
     'c_string_encoding': '',
+    'str_is_str': None,  # fall back to 'language_level == 2'
     'type_version_tag': True,  # enables Py_TPFLAGS_HAVE_VERSION_TAG on extension types
     'unraisable_tracebacks': True,
     'old_style_globals': False,
@@ -313,6 +314,7 @@ def normalise_encoding_name(option_name, encoding):
     'freelist': int,
     'c_string_type': one_of('bytes', 'bytearray', 'str', 'unicode'),
     'c_string_encoding': normalise_encoding_name,
+    'str_is_str': bool,
 }
 
 for key, val in _directive_defaults.items():
@@ -347,6 +349,7 @@ def normalise_encoding_name(option_name, encoding):
     # Avoid scope-specific to/from_py_functions for c_string.
     'c_string_type': ('module',),
     'c_string_encoding': ('module',),
+    'str_is_str': ('module',),
     'type_version_tag': ('module', 'cclass'),
     'language_level': ('module',),
     # globals() could conceivably be controlled at a finer granularity,

Cython/Compiler/Parsing.py

@@ -3652,6 +3652,9 @@ def p_compiler_directive_comments(s):
             if 'language_level' in new_directives:
                 # Make sure we apply the language level already to the first token that follows the comments.
                 s.context.set_language_level(new_directives['language_level'])
+            if 'str_is_str' in new_directives:
+                # Make sure we apply 'str_is_str' directive already to the first token that follows the comments.
+                s.context.set_str_is_str(new_directives['str_is_str'])
 
             result.update(new_directives)
 

Cython/Compiler/Symtab.py

@@ -21,6 +21,7 @@
 from .TypeSlots import (
     pyfunction_signature, pymethod_signature, richcmp_special_methods,
     get_special_method_signature, get_property_accessor_signature)
+from . import Future
 
 from . import Code
 
@@ -1002,10 +1003,12 @@ def __init__(self):
             cname, type = definition
             self.declare_var(name, type, None, cname)
 
-    def lookup(self, name, language_level=None):
-        # 'language_level' is passed by ModuleScope
-        if language_level == 3:
-            if name == 'str':
+    def lookup(self, name, language_level=None, str_is_str=None):
+        # 'language_level' and 'str_is_str' are passed by ModuleScope
+        if name == 'str':
+            if str_is_str is None:
+                str_is_str = language_level in (None, 2)
+            if not str_is_str:
                 name = 'unicode'
         return Scope.lookup(self, name)
 
@@ -1174,15 +1177,18 @@ def qualifying_scope(self):
     def global_scope(self):
         return self
 
-    def lookup(self, name, language_level=None):
+    def lookup(self, name, language_level=None, str_is_str=None):
         entry = self.lookup_here(name)
         if entry is not None:
             return entry
 
         if language_level is None:
             language_level = self.context.language_level if self.context is not None else 3
+        if str_is_str is None:
+            str_is_str = language_level == 2 or (
+                self.context is not None and Future.unicode_literals not in self.context.future_directives)
 
-        return self.outer_scope.lookup(name, language_level=language_level)
+        return self.outer_scope.lookup(name, language_level=language_level, str_is_str=str_is_str)
 
     def declare_tuple_type(self, pos, components):
         components = tuple(components)

tests/run/cython3_no_unicode_literals.pyx

@@ -0,0 +1,101 @@
+# cython: language_level=3, binding=True, str_is_str=True
+# mode: run
+# tag: python3, str_is_str
+
+print(end='')  # test that language_level 3 applies immediately at the module start, for the first token.
+
+__doc__ = """
+>>> items = sorted(locals_function(1).items())
+>>> for item in items:
+...     print('%s = %r' % item)
+a = 1
+b = 2
+x = 'abc'
+"""
+
+def locals_function(a, b=2):
+    x = 'abc'
+    return locals()
+
+
+### true division
+
+def truediv(x):
+    """
+    >>> truediv(4)
+    2.0
+    >>> truediv(3)
+    1.5
+    """
+    return x / 2
+
+
+def truediv_int(int x):
+    """
+    >>> truediv_int(4)
+    2.0
+    >>> truediv_int(3)
+    1.5
+    """
+    return x / 2
+
+
+### Py3 feature tests
+
+def print_function(*args):
+    """
+    >>> print_function(1,2,3)
+    1 2 3
+    """
+    print(*args) # this isn't valid Py2 syntax
+
+
+str_string = "abcdefg"
+
+def no_unicode_literals():
+    """
+    >>> print( no_unicode_literals() )
+    True
+    abcdefg
+    """
+    print(isinstance(str_string, str) or type(str_string))
+    return str_string
+
+
+def str_type_is_str():
+    """
+    >>> str_type, s = str_type_is_str()
+    >>> isinstance(s, type(str_string)) or (s, str_type)
+    True
+    >>> isinstance(s, str_type) or (s, str_type)
+    True
+    >>> isinstance(str_string, str_type) or str_type
+    True
+    """
+    cdef str s = 'abc'
+    return str, s
+
+
+def annotation_syntax(a: "test new test", b : "other" = 2, *args: "ARGS", **kwargs: "KWARGS") -> "ret":
+    """
+    >>> annotation_syntax(1)
+    3
+    >>> annotation_syntax(1,3)
+    4
+
+    >>> len(annotation_syntax.__annotations__)
+    5
+    >>> annotation_syntax.__annotations__['a']
+    'test new test'
+    >>> annotation_syntax.__annotations__['b']
+    'other'
+    >>> annotation_syntax.__annotations__['args']
+    'ARGS'
+    >>> annotation_syntax.__annotations__['kwargs']
+    'KWARGS'
+    >>> annotation_syntax.__annotations__['return']
+    'ret'
+    """
+    result : int = a + b
+
+    return result