[mypyc] Implement bytes.endswith (#20447)

Rounding out #20387 and implementing `bytes.endswith`. Simple benchmark
shows a ~6.4x improvement.

Tested with the following benchmark code:

```
import time


def bench(suffix: bytes, a: list[bytes], n: int) -> int:
    i = 0
    for x in range(n):
        for b in a:
            if b.endswith(suffix):
                i += 1
    return i


a = [b"foo", b"barasdfsf", b"foobar", b"ab", b"asrtert", b"sertyeryt"]
n = 5 * 1000 * 1000
suffix = b"foo"

bench(suffix, a, n)

t0 = time.time()
bench(suffix, a, n)
td = time.time() - t0
print(f"{td}s")
```

Output:

```
$ python bench.py         
0.9002199172973633s
$ python -c "import bench"
0.13828086853027344s
```

Author: esarp

mypyc/lib-rt/CPy.h

@@ -790,7 +790,7 @@ PyObject *CPyBytes_Join(PyObject *sep, PyObject *iter);
 CPyTagged CPyBytes_Ord(PyObject *obj);
 PyObject *CPyBytes_Multiply(PyObject *bytes, CPyTagged count);
 int CPyBytes_Startswith(PyObject *self, PyObject *subobj);
-
+int CPyBytes_Endswith(PyObject *self, PyObject *subobj);
 int CPyBytes_Compare(PyObject *left, PyObject *right);
 
 

mypyc/lib-rt/bytes_ops.c

@@ -183,3 +183,36 @@ int CPyBytes_Startswith(PyObject *self, PyObject *subobj) {
     }
     return ret;
 }
+
+int CPyBytes_Endswith(PyObject *self, PyObject *subobj) {
+    if (PyBytes_CheckExact(self) && PyBytes_CheckExact(subobj)) {
+        if (self == subobj) {
+            return 1;
+        }
+
+        Py_ssize_t subobj_len = PyBytes_GET_SIZE(subobj);
+        if (subobj_len == 0) {
+            return 1;
+        }
+
+        Py_ssize_t self_len = PyBytes_GET_SIZE(self);
+        if (subobj_len > self_len) {
+            return 0;
+        }
+
+        const char *self_buf = PyBytes_AS_STRING(self);
+        const char *subobj_buf = PyBytes_AS_STRING(subobj);
+
+        return memcmp(self_buf + (self_len - subobj_len), subobj_buf, (size_t)subobj_len) == 0 ? 1 : 0;
+    }
+    PyObject *result = PyObject_CallMethodOneArg(self, mypyc_interned_str.endswith, subobj);
+    if (result == NULL) {
+        return 2;
+    }
+    int ret = PyObject_IsTrue(result);
+    Py_DECREF(result);
+    if (ret < 0) {
+        return 2;
+    }
+    return ret;
+}

mypyc/lib-rt/static_data.c

@@ -52,6 +52,7 @@ intern_strings(void) {
     INTERN_STRING(close_, "close");
     INTERN_STRING(copy, "copy");
     INTERN_STRING(dispatch_cache, "dispatch_cache");
+    INTERN_STRING(endswith, "endswith");
     INTERN_STRING(get_type_hints, "get_type_hints");
     INTERN_STRING(keys, "keys");
     INTERN_STRING(items, "items");

mypyc/lib-rt/static_data.h

@@ -44,6 +44,7 @@ typedef struct mypyc_interned_str_struct {
     PyObject *close_;
     PyObject *copy;
     PyObject *dispatch_cache;
+    PyObject *endswith;
     PyObject *get_type_hints;
     PyObject *keys;
     PyObject *items;

mypyc/primitives/bytes_ops.py

@@ -133,6 +133,16 @@
     error_kind=ERR_MAGIC,
 )
 
+# bytes.endswith(bytes)
+method_op(
+    name="endswith",
+    arg_types=[bytes_rprimitive, bytes_rprimitive],
+    return_type=c_int_rprimitive,
+    c_function_name="CPyBytes_Endswith",
+    truncated_type=bool_rprimitive,
+    error_kind=ERR_MAGIC,
+)
+
 # Join bytes objects and return a new bytes.
 # The first argument is the total number of the following bytes.
 bytes_build_op = custom_op(

mypyc/test-data/fixtures/ir.py

@@ -180,6 +180,7 @@ def join(self, x: Iterable[object]) -> bytes: ...
     def decode(self, encoding: str=..., errors: str=...) -> str: ...
     def translate(self, t: bytes | bytearray) -> bytes: ...
     def startswith(self, t: bytes | bytearray) -> bool: ...
+    def endswith(self, t: bytes | bytearray) -> bool: ...
     def __iter__(self) -> Iterator[int]: ...
 
 class bytearray:
@@ -197,6 +198,7 @@ def __getitem__(self, i: int) -> int: ...
     def __getitem__(self, i: slice) -> bytearray: ...
     def decode(self, x: str = ..., y: str = ...) -> str: ...
     def startswith(self, t: bytes) -> bool: ...
+    def endswith(self, t: bytes) -> bool: ...
 
 class bool(int):
     def __init__(self, o: object = ...) -> None: ...

mypyc/test-data/irbuild-bytes.test

@@ -266,6 +266,19 @@ L0:
     r1 = truncate r0: i32 to builtins.bool
     return r1
 
+[case testBytesEndsWith]
+def f(a: bytes, b: bytes) -> bool:
+    return a.endswith(b)
+[out]
+def f(a, b):
+    a, b :: bytes
+    r0 :: i32
+    r1 :: bool
+L0:
+    r0 = CPyBytes_Endswith(a, b)
+    r1 = truncate r0: i32 to builtins.bool
+    return r1
+
 [case testBytesVsBytearray]
 def bytes_func(b: bytes) -> None: pass
 def bytearray_func(ba: bytearray) -> None: pass

mypyc/test-data/run-bytes.test

@@ -221,6 +221,27 @@ def test_startswith() -> None:
     assert test2.startswith(b'some')
     assert not test2.startswith(b'other')
 
+def test_endswith() -> None:
+    # Test default behavior
+    test = b'some string'
+    assert test.endswith(b'string')
+    assert test.endswith(b'some string')
+    assert not test.endswith(b'other')
+    assert not test.endswith(b'some string but longer')
+
+    # Test empty cases
+    assert test.endswith(b'')
+    assert b''.endswith(b'')
+    assert not b''.endswith(test)
+
+    # Test bytearray to verify slow paths
+    assert test.endswith(bytearray(b'string'))
+    assert not test.endswith(bytearray(b'other'))
+
+    test2 = bytearray(b'some string')
+    assert test2.endswith(b'string')
+    assert not test2.endswith(b'other')
+
 [case testBytesSlicing]
 def test_bytes_slicing() -> None:
     b = b'abcdefg'