BUG: memory leak on get_c_string (pandas-dev#16060)

* BUG: memory leak on get_c_string

* add except NULL

Author: chris-b1

doc/source/whatsnew/v0.20.0.txt

@@ -1523,7 +1523,7 @@ Performance Improvements
 ~~~~~~~~~~~~~~~~~~~~~~~~
 
 - Improved performance of ``pd.wide_to_long()`` (:issue:`14779`)
-- Improved performance of ``pd.factorize()`` by releasing the GIL with ``object`` dtype when inferred as strings (:issue:`14859`)
+- Improved performance of ``pd.factorize()`` by releasing the GIL with ``object`` dtype when inferred as strings (:issue:`14859`, :issue:`16057`)
 - Improved performance of timeseries plotting with an irregular DatetimeIndex
   (or with ``compat_x=True``) (:issue:`15073`).
 - Improved performance of ``groupby().cummin()`` and ``groupby().cummax()`` (:issue:`15048`, :issue:`15109`, :issue:`15561`, :issue:`15635`)

pandas/_libs/src/numpy_helper.h

@@ -87,16 +87,11 @@ PANDAS_INLINE PyObject* get_value_1d(PyArrayObject* ap, Py_ssize_t i) {
     return PyArray_Scalar(item, PyArray_DESCR(ap), (PyObject*)ap);
 }
 
+// returns ASCII or UTF8 (py3) view on python str
+// python object owns memory, should not be freed
 PANDAS_INLINE char* get_c_string(PyObject* obj) {
 #if PY_VERSION_HEX >= 0x03000000
-    PyObject* enc_str = PyUnicode_AsEncodedString(obj, "utf-8", "error");
-
-    char* ret;
-    ret = PyBytes_AS_STRING(enc_str);
-
-    // TODO(general): memory leak here
-
-    return ret;
+    return PyUnicode_AsUTF8(obj);
 #else
     return PyString_AsString(obj);
 #endif

pandas/_libs/src/util.pxd

@@ -17,7 +17,7 @@ cdef extern from "numpy_helper.h":
     inline cnp.int64_t get_nat()
     inline object get_value_1d(ndarray, Py_ssize_t)
     inline int floatify(object, double*) except -1
-    inline char *get_c_string(object)
+    inline char *get_c_string(object) except NULL
     inline object char_to_string(char*)
     inline void transfer_object_column(char *dst, char *src, size_t stride,
                                        size_t length)