bpo-40521: Make empty Unicode string per interpreter (GH-21096)

Each interpreter now has its own empty Unicode string singleton.

Author: vstinner

Include/internal/pycore_interp.h

@@ -71,6 +71,8 @@ struct _Py_bytes_state {
 };
 
 struct _Py_unicode_state {
+    // The empty Unicode object is a singleton to improve performance.
+    PyObject *empty;
     struct _Py_unicode_fs_codec fs_codec;
 };
 

Include/internal/pycore_pylifecycle.h

@@ -31,7 +31,7 @@ PyAPI_FUNC(int) _Py_IsLocaleCoercionTarget(const char *ctype_loc);
 
 /* Various one-time initializers */
 
-extern PyStatus _PyUnicode_Init(void);
+extern PyStatus _PyUnicode_Init(PyThreadState *tstate);
 extern int _PyStructSequence_Init(void);
 extern int _PyLong_Init(PyThreadState *tstate);
 extern PyStatus _PyFaulthandler_Init(int enable);

Misc/NEWS.d/next/Core and Builtins/2020-05-20-01-17-34.bpo-40521.wvAehI.rst

@@ -2,7 +2,7 @@ Each interpreter now its has own free lists, singletons and caches:
 
 * Free lists: float, tuple, list, dict, frame, context,
   asynchronous generator, MemoryError.
-* Singletons: empty tuple, empty bytes string,
+* Singletons: empty tuple, empty bytes string, empty Unicode string,
   single byte character.
 * Slice cache.
 

Objects/stringlib/asciilib.h

@@ -11,7 +11,6 @@
 #define STRINGLIB_CHAR           Py_UCS1
 #define STRINGLIB_TYPE_NAME      "unicode"
 #define STRINGLIB_PARSE_CODE     "U"
-#define STRINGLIB_GET_EMPTY()    unicode_empty
 #define STRINGLIB_ISSPACE        Py_UNICODE_ISSPACE
 #define STRINGLIB_ISLINEBREAK    BLOOM_LINEBREAK
 #define STRINGLIB_ISDECIMAL      Py_UNICODE_ISDECIMAL

Objects/stringlib/partition.h

@@ -1,9 +1,14 @@
 /* stringlib: partition implementation */
 
 #ifndef STRINGLIB_FASTSEARCH_H
-#error must include "stringlib/fastsearch.h" before including this module
+#  error must include "stringlib/fastsearch.h" before including this module
 #endif
 
+#if !STRINGLIB_MUTABLE && !defined(STRINGLIB_GET_EMPTY)
+#  error "STRINGLIB_GET_EMPTY must be defined if STRINGLIB_MUTABLE is zero"
+#endif
+
+
 Py_LOCAL_INLINE(PyObject*)
 STRINGLIB(partition)(PyObject* str_obj,
                     const STRINGLIB_CHAR* str, Py_ssize_t str_len,

Objects/stringlib/stringdefs.h

@@ -1,10 +1,6 @@
 #ifndef STRINGLIB_STRINGDEFS_H
 #define STRINGLIB_STRINGDEFS_H
 
-#ifndef STRINGLIB_GET_EMPTY
-#  error "STRINGLIB_GET_EMPTY macro must be defined"
-#endif
-
 /* this is sort of a hack.  there's at least one place (formatting
    floats) where some stringlib code takes a different path if it's
    compiled as unicode. */

Objects/stringlib/ucs1lib.h

@@ -11,7 +11,6 @@
 #define STRINGLIB_CHAR           Py_UCS1
 #define STRINGLIB_TYPE_NAME      "unicode"
 #define STRINGLIB_PARSE_CODE     "U"
-#define STRINGLIB_GET_EMPTY()    unicode_empty
 #define STRINGLIB_ISSPACE        Py_UNICODE_ISSPACE
 #define STRINGLIB_ISLINEBREAK    BLOOM_LINEBREAK
 #define STRINGLIB_ISDECIMAL      Py_UNICODE_ISDECIMAL

Objects/stringlib/ucs2lib.h

@@ -11,7 +11,6 @@
 #define STRINGLIB_CHAR           Py_UCS2
 #define STRINGLIB_TYPE_NAME      "unicode"
 #define STRINGLIB_PARSE_CODE     "U"
-#define STRINGLIB_GET_EMPTY()    unicode_empty
 #define STRINGLIB_ISSPACE        Py_UNICODE_ISSPACE
 #define STRINGLIB_ISLINEBREAK    BLOOM_LINEBREAK
 #define STRINGLIB_ISDECIMAL      Py_UNICODE_ISDECIMAL

Objects/stringlib/ucs4lib.h

@@ -11,7 +11,6 @@
 #define STRINGLIB_CHAR           Py_UCS4
 #define STRINGLIB_TYPE_NAME      "unicode"
 #define STRINGLIB_PARSE_CODE     "U"
-#define STRINGLIB_GET_EMPTY()    unicode_empty
 #define STRINGLIB_ISSPACE        Py_UNICODE_ISSPACE
 #define STRINGLIB_ISLINEBREAK    BLOOM_LINEBREAK
 #define STRINGLIB_ISDECIMAL      Py_UNICODE_ISDECIMAL

Objects/stringlib/unicodedefs.h

@@ -13,7 +13,6 @@
 #define STRINGLIB_CHAR           Py_UNICODE
 #define STRINGLIB_TYPE_NAME      "unicode"
 #define STRINGLIB_PARSE_CODE     "U"
-#define STRINGLIB_GET_EMPTY()    unicode_empty
 #define STRINGLIB_ISSPACE        Py_UNICODE_ISSPACE
 #define STRINGLIB_ISLINEBREAK    BLOOM_LINEBREAK
 #define STRINGLIB_ISDECIMAL      Py_UNICODE_ISDECIMAL