make str values in compile time DEF env turn into actually usable Unicode...

make str values in compile time DEF env turn into actually usable Unicode strings when compiling under Python 3

CHANGES.rst

@@ -27,6 +27,11 @@ Bugs fixed
   strict C compiler options and prevents non-const strings literals
   with the same content from being incorrectly merged.
 
+* Compile time evaluated ``str`` expressions (``DEF``) now behave in a
+  more useful way by turning into Unicode strings when compiling under
+  Python 3.  This allows using them as intermediate values in expressions.
+  Previously, they always evaluated to bytes objects.
+
 * ``isinf()`` declarations in ``libc/math.pxd`` and ``numpy/math.pxd`` now
   reflect the actual tristate ``int`` return value instead of using ``bint``.
 

Cython/Compiler/ExprNodes.py

@@ -13,7 +13,8 @@ cython.declare(error=object, warning=object, warn_once=object, InternalError=obj
                unicode_type=object, str_type=object, bytes_type=object, type_type=object,
                Builtin=object, Symtab=object, Utils=object, find_coercion_error=object,
                debug_disposal_code=object, debug_temp_alloc=object, debug_coercion=object,
-               bytearray_type=object, slice_type=object, _py_int_types=object)
+               bytearray_type=object, slice_type=object, _py_int_types=object,
+               IS_PYTHON3=cython.bint)
 
 import sys
 import copy
@@ -54,8 +55,10 @@ else:
 
 
 if sys.version_info[0] >= 3:
+    IS_PYTHON3 = True
     _py_int_types = int
 else:
+    IS_PYTHON3 = False
     _py_int_types = (int, long)
 
 
@@ -1529,7 +1532,13 @@ class StringNode(PyConstNode):
         return self.result_code
 
     def compile_time_value(self, env):
-        return self.value.byteencode()
+        if not IS_PYTHON3:
+            # use plain str/bytes object in Py2
+            return self.value.byteencode()
+        # in Py3, always return a Unicode string
+        if self.unicode_value is not None:
+            return self.unicode_value
+        return self.value.decode('iso8859-1')
 
 
 class IdentifierStringNode(StringNode):

tests/run/ct_DEF.pyx

@@ -42,7 +42,7 @@ DEF INT_TUPLE1 = TUPLE[:2]
 DEF INT_TUPLE2 = TUPLE[1:4:2]
 DEF ELLIPSIS = ...
 DEF EXPRESSION = int(float(2*2)) + int(str(2)) + int(max(1,2,3)) + sum([TWO, FIVE])
-DEF UNICODE_EXPRESSION = unicode(b'abc'.decode(u'utf8')).encode(u'ascii').decode(u'latin1')
+DEF UNICODE_EXPRESSION = unicode(b'abc'.decode('utf8')).encode('ascii').decode('latin1')
 
 
 def c():