python · vstinner · Jun 25, 2020 · Jun 25, 2020
diff --git a/Include/internal/pycore_interp.h b/Include/internal/pycore_interp.h
@@ -66,13 +66,13 @@ struct _Py_unicode_fs_codec {
 };
 
 struct _Py_bytes_state {
+    PyObject *empty_string;
     PyBytesObject *characters[256];
-    PyBytesObject *empty_string;
 };
 
 struct _Py_unicode_state {
     // The empty Unicode object is a singleton to improve performance.
-    PyObject *empty;
+    PyObject *empty_string;
     /* Single character Unicode strings in the Latin-1 range are being
        shared as well. */
     PyObject *latin1[256];

diff --git a/Include/internal/pycore_pylifecycle.h b/Include/internal/pycore_pylifecycle.h
@@ -32,6 +32,7 @@ PyAPI_FUNC(int) _Py_IsLocaleCoercionTarget(const char *ctype_loc);
 /* Various one-time initializers */
 
 extern PyStatus _PyUnicode_Init(PyThreadState *tstate);
+extern PyStatus _PyBytes_Init(PyThreadState *tstate);
 extern int _PyStructSequence_Init(void);
 extern int _PyLong_Init(PyThreadState *tstate);
 extern PyStatus _PyTuple_Init(PyThreadState *tstate);

diff --git a/Objects/bytesobject.c b/Objects/bytesobject.c
@@ -4,8 +4,9 @@
 
 #include "Python.h"
 #include "pycore_abstract.h"      // _PyIndex_Check()
-#include "pycore_bytes_methods.h"
-#include "pycore_object.h"
+#include "pycore_bytes_methods.h" // _Py_bytes_startswith()
+#include "pycore_initconfig.h"    // _PyStatus_OK()
+#include "pycore_object.h"        // _PyObject_GC_TRACK
 #include "pycore_pymem.h"         // PYMEM_CLEANBYTE
 
 #include "pystrhex.h"
@@ -41,6 +42,44 @@ get_bytes_state(void)
 }
 
 
+// Return a borrowed reference to the empty bytes string singleton.
+static inline PyObject* bytes_get_empty(void)
+{
+    struct _Py_bytes_state *state = get_bytes_state();
+    // bytes_get_empty() must not be called before _PyBytes_Init()
+    // or after _PyBytes_Fini()
+    assert(state->empty_string != NULL);
+    return state->empty_string;
+}
+
+
+// Return a strong reference to the empty bytes string singleton.
+static inline PyObject* bytes_new_empty(void)
+{
+    PyObject *empty = bytes_get_empty();
+    Py_INCREF(empty);
+    return (PyObject *)empty;
+}
+
+
+static int
+bytes_create_empty_string_singleton(struct _Py_bytes_state *state)
+{
+    // Create the empty bytes string singleton
+    PyBytesObject *op = (PyBytesObject *)PyObject_Malloc(PyBytesObject_SIZE);
+    if (op == NULL) {
+        return -1;
+    }
+    _PyObject_InitVar((PyVarObject*)op, &PyBytes_Type, 0);
+    op->ob_shash = -1;
+    op->ob_sval[0] = '\0';
+
+    assert(state->empty_string == NULL);
+    state->empty_string = (PyObject *)op;
+    return 0;
+}
+
+
 /*
    For PyBytes_FromString(), the parameter `str' points to a null-terminated
    string containing exactly `size' bytes.
@@ -70,12 +109,7 @@ _PyBytes_FromSize(Py_ssize_t size, int use_calloc)
     assert(size >= 0);
 
     if (size == 0) {
-        struct _Py_bytes_state *state = get_bytes_state();
-        op = state->empty_string;
-        if (op != NULL) {
-            Py_INCREF(op);
-            return (PyObject *)op;
-        }
+        return bytes_new_empty();
     }
 
     if ((size_t)size > (size_t)PY_SSIZE_T_MAX - PyBytesObject_SIZE) {
@@ -94,13 +128,8 @@ _PyBytes_FromSize(Py_ssize_t size, int use_calloc)
     }
     _PyObject_InitVar((PyVarObject*)op, &PyBytes_Type, size);
     op->ob_shash = -1;
-    if (!use_calloc)
+    if (!use_calloc) {
         op->ob_sval[size] = '\0';
-    /* empty byte string singleton */
-    if (size == 0) {
-        struct _Py_bytes_state *state = get_bytes_state();
-        Py_INCREF(op);
-        state->empty_string = op;
     }
     return (PyObject *) op;
 }
@@ -122,6 +151,9 @@ PyBytes_FromStringAndSize(const char *str, Py_ssize_t size)
             return (PyObject *)op;
         }
     }
+    if (size == 0) {
+        return bytes_new_empty();
+    }
 
     op = (PyBytesObject *)_PyBytes_FromSize(size, 0);
     if (op == NULL)
@@ -155,11 +187,7 @@ PyBytes_FromString(const char *str)
 
     struct _Py_bytes_state *state = get_bytes_state();
     if (size == 0) {
-        op = state->empty_string;
-        if (op != NULL) {
-            Py_INCREF(op);
-            return (PyObject *)op;
-        }
+        return bytes_new_empty();
     }
     else if (size == 1) {
         op = state->characters[*str & UCHAR_MAX];
@@ -178,11 +206,8 @@ PyBytes_FromString(const char *str)
     op->ob_shash = -1;
     memcpy(op->ob_sval, str, size+1);
     /* share short strings */
-    if (size == 0) {
-        Py_INCREF(op);
-        state->empty_string = op;
-    }
-    else if (size == 1) {
+    if (size == 1) {
+        assert(state->characters[*str & UCHAR_MAX] == NULL);
         Py_INCREF(op);
         state->characters[*str & UCHAR_MAX] = op;
     }
@@ -1272,7 +1297,7 @@ PyBytes_AsStringAndSize(PyObject *obj,
 /* -------------------------------------------------------------------- */
 /* Methods */
 
-#define STRINGLIB_GET_EMPTY() get_bytes_state()->empty_string
+#define STRINGLIB_GET_EMPTY() bytes_get_empty()
 
 #include "stringlib/stringdefs.h"
 
@@ -3053,9 +3078,9 @@ _PyBytes_Resize(PyObject **pv, Py_ssize_t newsize)
         goto error;
     }
     if (newsize == 0) {
-        *pv = _PyBytes_FromSize(0, 0);
+        *pv = bytes_new_empty();
         Py_DECREF(v);
-        return (*pv == NULL) ? -1 : 0;
+        return 0;
     }
     /* XXX UNREF/NEWREF interface should be more symmetrical */
 #ifdef Py_REF_DEBUG
@@ -3084,6 +3109,18 @@ _PyBytes_Resize(PyObject **pv, Py_ssize_t newsize)
     return -1;
 }
 
+
+PyStatus
+_PyBytes_Init(PyThreadState *tstate)
+{
+    struct _Py_bytes_state *state = &tstate->interp->bytes;
+    if (bytes_create_empty_string_singleton(state) < 0) {
+        return _PyStatus_NO_MEMORY();
+    }
+    return _PyStatus_OK();
+}
+
+
 void
 _PyBytes_Fini(PyThreadState *tstate)
 {

diff --git a/Objects/unicodeobject.c b/Objects/unicodeobject.c
@@ -41,16 +41,15 @@ OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 #define PY_SSIZE_T_CLEAN
 #include "Python.h"
 #include "pycore_abstract.h"       // _PyIndex_Check()
-#include "pycore_bytes_methods.h"
-#include "pycore_fileutils.h"
-#include "pycore_initconfig.h"
+#include "pycore_bytes_methods.h"  // _Py_bytes_lower()
+#include "pycore_initconfig.h"     // _PyStatus_OK()
 #include "pycore_interp.h"         // PyInterpreterState.fs_codec
-#include "pycore_object.h"
-#include "pycore_pathconfig.h"
-#include "pycore_pylifecycle.h"
+#include "pycore_object.h"         // _PyObject_GC_TRACK()
+#include "pycore_pathconfig.h"     // _Py_DumpPathConfig()
+#include "pycore_pylifecycle.h"    // _Py_SetFileSystemEncoding()
 #include "pycore_pystate.h"        // _PyInterpreterState_GET()
-#include "ucnhash.h"
-#include "stringlib/eq.h"
+#include "ucnhash.h"               // _PyUnicode_Name_CAPI
+#include "stringlib/eq.h"          // unicode_eq()
 
 #ifdef MS_WINDOWS
 #include <windows.h>
@@ -236,10 +235,12 @@ static inline PyObject* unicode_get_empty(void)
     struct _Py_unicode_state *state = get_unicode_state();
     // unicode_get_empty() must not be called before _PyUnicode_Init()
     // or after _PyUnicode_Fini()
-    assert(state->empty != NULL);
-    return state->empty;
+    assert(state->empty_string != NULL);
+    return state->empty_string;
 }
 
+
+// Return a strong reference to the empty string singleton.
 static inline PyObject* unicode_new_empty(void)
 {
     PyObject *empty = unicode_get_empty();
@@ -1385,6 +1386,26 @@ _PyUnicode_Dump(PyObject *op)
 }
 #endif
 
+static int
+unicode_create_empty_string_singleton(struct _Py_unicode_state *state)
+{
+    // Use size=1 rather than size=0, so PyUnicode_New(0, maxchar) can be
+    // optimized to always use state->empty_string without having to check if
+    // it is NULL or not.
+    PyObject *empty = PyUnicode_New(1, 0);
+    if (empty == NULL) {
+        return -1;
+    }
+    PyUnicode_1BYTE_DATA(empty)[0] = 0;
+    _PyUnicode_LENGTH(empty) = 0;
+    assert(_PyUnicode_CheckConsistency(empty, 1));
+
+    assert(state->empty_string == NULL);
+    state->empty_string = empty;
+    return 0;
+}
+
+
 PyObject *
 PyUnicode_New(Py_ssize_t size, Py_UCS4 maxchar)
 {
@@ -1972,7 +1993,7 @@ static int
 unicode_is_singleton(PyObject *unicode)
 {
     struct _Py_unicode_state *state = get_unicode_state();
-    if (unicode == state->empty) {
+    if (unicode == state->empty_string) {
         return 1;
     }
     PyASCIIObject *ascii = (PyASCIIObject *)unicode;
@@ -15542,20 +15563,10 @@ _PyUnicode_Init(PyThreadState *tstate)
         0x2029, /* PARAGRAPH SEPARATOR */
     };
 
-    // Use size=1 rather than size=0, so PyUnicode_New(0, maxchar) can be
-    // optimized to always use state->empty without having to check if it is
-    // NULL or not.
-    PyObject *empty = PyUnicode_New(1, 0);
-    if (empty == NULL) {
+    struct _Py_unicode_state *state = &tstate->interp->unicode;
+    if (unicode_create_empty_string_singleton(state) < 0) {
         return _PyStatus_NO_MEMORY();
     }
-    PyUnicode_1BYTE_DATA(empty)[0] = 0;
-    _PyUnicode_LENGTH(empty) = 0;
-    assert(_PyUnicode_CheckConsistency(empty, 1));
-
-    struct _Py_unicode_state *state = &tstate->interp->unicode;
-    assert(state->empty == NULL);
-    state->empty = empty;
 
     if (_Py_IsMainInterpreter(tstate)) {
         /* initialize the linebreak bloom filter */
@@ -16223,7 +16234,7 @@ _PyUnicode_Fini(PyThreadState *tstate)
 #endif /* __INSURE__ */
     }
 
-    Py_CLEAR(state->empty);
+    Py_CLEAR(state->empty_string);
 
     for (Py_ssize_t i = 0; i < 256; i++) {
         Py_CLEAR(state->latin1[i]);

diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c
@@ -607,6 +607,11 @@ pycore_init_types(PyThreadState *tstate)
         return status;
     }
 
+    status = _PyBytes_Init(tstate);
+    if (_PyStatus_EXCEPTION(status)) {
+        return status;
+    }
+
     status = _PyExc_Init(tstate);
     if (_PyStatus_EXCEPTION(status)) {
         return status;