Add efficient primitives for str.strip().

mypyc/lib-rt/CPy.h

@@ -717,13 +717,27 @@ static inline char CPyDict_CheckSize(PyObject *dict, CPyTagged size) {
 
 // Str operations
 
+// Macros for strip type. These values are copied from CPython.
+#define LEFTSTRIP  0
+#define RIGHTSTRIP 1
+#define BOTHSTRIP  2
 
 PyObject *CPyStr_Build(Py_ssize_t len, ...);
 PyObject *CPyStr_GetItem(PyObject *str, CPyTagged index);
 CPyTagged CPyStr_Find(PyObject *str, PyObject *substr, CPyTagged start, int direction);
 CPyTagged CPyStr_FindWithEnd(PyObject *str, PyObject *substr, CPyTagged start, CPyTagged end, int direction);
 PyObject *CPyStr_Split(PyObject *str, PyObject *sep, CPyTagged max_split);
 PyObject *CPyStr_RSplit(PyObject *str, PyObject *sep, CPyTagged max_split);
+PyObject *_CPyStr_Strip(PyObject *self, int strip_type, PyObject *sep);
+static inline PyObject *CPyStr_Strip(PyObject *self, PyObject *sep) {
+    return _CPyStr_Strip(self, BOTHSTRIP, sep);
+}
+static inline PyObject *CPyStr_LStrip(PyObject *self, PyObject *sep) {
+    return _CPyStr_Strip(self, LEFTSTRIP, sep);
+}
+static inline PyObject *CPyStr_RStrip(PyObject *self, PyObject *sep) {
+    return _CPyStr_Strip(self, RIGHTSTRIP, sep);
+}
 PyObject *CPyStr_Replace(PyObject *str, PyObject *old_substr, PyObject *new_substr, CPyTagged max_replace);
 PyObject *CPyStr_Append(PyObject *o1, PyObject *o2);
 PyObject *CPyStr_GetSlice(PyObject *obj, CPyTagged start, CPyTagged end);

mypyc/lib-rt/str_ops.c

@@ -5,6 +5,58 @@
 #include <Python.h>
 #include "CPy.h"
 
+// Copied from cpython.git:Objects/unicodeobject.c.
+#define BLOOM_MASK unsigned long
+#define BLOOM(mask, ch)     ((mask &  (1UL << ((ch) & (BLOOM_WIDTH - 1)))))
+#if LONG_BIT >= 128
+#define BLOOM_WIDTH 128
+#elif LONG_BIT >= 64
+#define BLOOM_WIDTH 64
+#elif LONG_BIT >= 32
+#define BLOOM_WIDTH 32
+#else
+#error "LONG_BIT is smaller than 32"
+#endif
+
+// Copied from cpython.git:Objects/unicodeobject.c. This is needed for str.strip("...").
+static inline BLOOM_MASK
+make_bloom_mask(int kind, const void* ptr, Py_ssize_t len)
+{
+#define BLOOM_UPDATE(TYPE, MASK, PTR, LEN)             \
+    do {                                               \
+        TYPE *data = (TYPE *)PTR;                      \
+        TYPE *end = data + LEN;                        \
+        Py_UCS4 ch;                                    \
+        for (; data != end; data++) {                  \
+            ch = *data;                                \
+            MASK |= (1UL << (ch & (BLOOM_WIDTH - 1))); \
+        }                                              \
+        break;                                         \
+    } while (0)
+
+    /* calculate simple bloom-style bitmask for a given unicode string */
+
+    BLOOM_MASK mask;
+
+    mask = 0;
+    switch (kind) {
+    case PyUnicode_1BYTE_KIND:
+        BLOOM_UPDATE(Py_UCS1, mask, ptr, len);
+        break;
+    case PyUnicode_2BYTE_KIND:
+        BLOOM_UPDATE(Py_UCS2, mask, ptr, len);
+        break;
+    case PyUnicode_4BYTE_KIND:
+        BLOOM_UPDATE(Py_UCS4, mask, ptr, len);
+        break;
+    default:
+        Py_UNREACHABLE();
+    }
+    return mask;
+
+#undef BLOOM_UPDATE
+}
+
 PyObject *CPyStr_GetItem(PyObject *str, CPyTagged index) {
     if (PyUnicode_READY(str) != -1) {
         if (CPyTagged_CheckShort(index)) {
@@ -174,6 +226,116 @@ PyObject *CPyStr_RSplit(PyObject *str, PyObject *sep, CPyTagged max_split) {
     return PyUnicode_RSplit(str, sep, temp_max_split);
 }
 
+// This function has been copied from _PyUnicode_XStrip in cpython.git:Objects/unicodeobject.c.
+static PyObject *_PyStr_XStrip(PyObject *self, int striptype, PyObject *sepobj) {
+    const void *data;
+    int kind;
+    Py_ssize_t i, j, len;
+    BLOOM_MASK sepmask;
+    Py_ssize_t seplen;
+
+    kind = PyUnicode_KIND(self);
+    data = PyUnicode_DATA(self);
+    len = PyUnicode_GET_LENGTH(self);
+    seplen = PyUnicode_GET_LENGTH(sepobj);
+    sepmask = make_bloom_mask(PyUnicode_KIND(sepobj),
+                              PyUnicode_DATA(sepobj),
+                              seplen);
+
+    i = 0;
+    if (striptype != RIGHTSTRIP) {
+        while (i < len) {
+            Py_UCS4 ch = PyUnicode_READ(kind, data, i);
+            if (!BLOOM(sepmask, ch))
+                break;
+            if (PyUnicode_FindChar(sepobj, ch, 0, seplen, 1) < 0)
+                break;
+            i++;
+        }
+    }
+
+    j = len;
+    if (striptype != LEFTSTRIP) {
+        j--;
+        while (j >= i) {
+            Py_UCS4 ch = PyUnicode_READ(kind, data, j);
+            if (!BLOOM(sepmask, ch))
+                break;
+            if (PyUnicode_FindChar(sepobj, ch, 0, seplen, 1) < 0)
+                break;
+            j--;
+        }
+
+        j++;
+    }
+
+    return PyUnicode_Substring(self, i, j);
+}
+
+// Copied from do_strip function in cpython.git/Objects/unicodeobject.c.
+PyObject *_CPyStr_Strip(PyObject *self, int strip_type, PyObject *sep) {
+    if (sep == NULL || sep == Py_None) {
+        Py_ssize_t len, i, j;
+
+        len = PyUnicode_GET_LENGTH(self);
+
+        if (PyUnicode_IS_ASCII(self)) {
+            const Py_UCS1 *data = PyUnicode_1BYTE_DATA(self);
+
+            i = 0;
+            if (strip_type != RIGHTSTRIP) {
+                while (i < len) {
+                    Py_UCS1 ch = data[i];
+                    if (!_Py_ascii_whitespace[ch])
+                        break;
+                    i++;
+                }
+            }
+
+            j = len;
+            if (strip_type != LEFTSTRIP) {
+                j--;
+                while (j >= i) {
+                    Py_UCS1 ch = data[j];
+                    if (!_Py_ascii_whitespace[ch])
+                        break;
+                    j--;
+                }
+                j++;
+            }
+        }
+        else {
+            int kind = PyUnicode_KIND(self);
+            const void *data = PyUnicode_DATA(self);
+
+            i = 0;
+            if (strip_type != RIGHTSTRIP) {
+                while (i < len) {
+                    Py_UCS4 ch = PyUnicode_READ(kind, data, i);
+                    if (!Py_UNICODE_ISSPACE(ch))
+                        break;
+                    i++;
+                }
+            }
+
+            j = len;
+            if (strip_type != LEFTSTRIP) {
+                j--;
+                while (j >= i) {
+                    Py_UCS4 ch = PyUnicode_READ(kind, data, j);
+                    if (!Py_UNICODE_ISSPACE(ch))
+                        break;
+                    j--;
+                }
+                j++;
+            }
+        }
+
+        return PyUnicode_Substring(self, i, j);
+    }
+    return _PyStr_XStrip(self, strip_type, sep);
+}
+
 PyObject *CPyStr_Replace(PyObject *str, PyObject *old_substr,
                          PyObject *new_substr, CPyTagged max_replace) {
     Py_ssize_t temp_max_replace = CPyTagged_AsSsize_t(max_replace);

mypyc/primitives/str_ops.py

@@ -135,6 +135,26 @@
     var_arg_type=str_rprimitive,
 )
 
+# str.strip, str.lstrip, str.rstrip
+# Order of iteration matters. It should correspond with LEFTSTRIP, RIGHTSTRIP and BOTHSTRIP macros defined in CPy.h.
+for strip_prefix in ["l", "r", ""]:
+    method_op(
+        name=f"{strip_prefix}strip",
+        arg_types=[str_rprimitive, str_rprimitive],
+        return_type=str_rprimitive,
+        c_function_name=f"CPyStr_{strip_prefix.upper()}Strip",
+        error_kind=ERR_NEVER,
+    )
+    method_op(
+        name=f"{strip_prefix}strip",
+        arg_types=[str_rprimitive],
+        return_type=str_rprimitive,
+        c_function_name=f"CPyStr_{strip_prefix.upper()}Strip",
+        # This 0 below is implicitly treated as NULL in C.
+        extra_int_constants=[(0, c_int_rprimitive)],
+        error_kind=ERR_NEVER,
+    )
+
 # str.startswith(str)
 method_op(
     name="startswith",

mypyc/test-data/fixtures/ir.py

@@ -107,7 +107,9 @@ def rfind(self, sub: str, start: Optional[int] = None, end: Optional[int] = None
     def split(self, sep: Optional[str] = None, maxsplit: int = -1) -> List[str]: pass
     def rsplit(self, sep: Optional[str] = None, maxsplit: int = -1) -> List[str]: pass
     def splitlines(self, keepends: bool = False) -> List[str]: ...
-    def strip (self, item: str) -> str: pass
+    def strip (self, item: Optional[str] = None) -> str: pass
+    def lstrip(self, item: Optional[str] = None) -> str: pass
+    def rstrip(self, item: Optional[str] = None) -> str: pass
     def join(self, x: Iterable[str]) -> str: pass
     def format(self, *args: Any, **kwargs: Any) -> str: ...
     def upper(self) -> str: ...

mypyc/test-data/irbuild-str.test

@@ -481,3 +481,26 @@ L0:
     keep_alive x
     r6 = unbox(int, r5)
     return r6
+
+[case testStrip]
+def do_strip(s: str) -> None:
+    s.lstrip("x")
+    s.strip("y")
+    s.rstrip("z")
+    s.lstrip()
+    s.strip()
+    s.rstrip()
+[out]
+def do_strip(s):
+    s, r0, r1, r2, r3, r4, r5, r6, r7, r8 :: str
+L0:
+    r0 = 'x'
+    r1 = CPyStr_LStrip(s, r0)
+    r2 = 'y'
+    r3 = CPyStr_Strip(s, r2)
+    r4 = 'z'
+    r5 = CPyStr_RStrip(s, r4)
+    r6 = CPyStr_LStrip(s, 0)
+    r7 = CPyStr_Strip(s, 0)
+    r8 = CPyStr_RStrip(s, 0)
+    return 1

mypyc/test-data/run-strings.test

@@ -774,3 +774,16 @@ def test_surrogate() -> None:
     assert ord(f()) == 0xd800
     assert ord("\udfff") == 0xdfff
     assert repr("foobar\x00\xab\ud912\U00012345") == r"'foobar\x00«\ud912𒍅'"
+
+[case testStrip]
+# This is a negative test. strip variants without args does not use efficient primitives.
+def test_all_strips_default() -> None:
+    s = " a1\t"
+    assert s.lstrip() == "a1\t"
+    assert s.strip() == "a1"
+    assert s.rstrip() == " a1"
+def test_all_strips() -> None:
+    s = "xxb2yy"
+    assert s.lstrip("xy") == "b2yy"
+    assert s.strip("xy") == "b2"
+    assert s.rstrip("xy") == "xxb2"