Fix security and correctness issues from #25 Array API (#40)

* Fix security and correctness issues in array API

- Fix Py_BuildValue format specifiers: use 'L' for int64_t and 'K' for
  uint64_t instead of 'i' which silently truncates 64-bit values
- Fix signed integer overflow UB in best_within_dist when max_dist is
  large: use -1 sentinel instead of max_dist+1
- Replace new[]/delete[] with malloc/free + NULL check to prevent
  uncaught std::bad_alloc from crashing the process
- Fix NEON bug: max_dist > 0 should be max_dist >= 0 so that max_dist=0
  correctly delegates to the native path

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Add Python 3.14 to CI build matrix

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: mrecachinas

.github/workflows/pythonpackage.yml

@@ -14,7 +14,7 @@ jobs:
     runs-on: ubuntu-24.04
     strategy:
       matrix:
-        python_minor: [ "10" , "11" , "12" , "13" ]
+        python_minor: [ "10" , "11" , "12" , "13" , "14" ]
 
     steps:
       - uses: actions/checkout@v4.2.2
@@ -83,7 +83,7 @@ jobs:
     strategy:
       fail-fast: true
       matrix:
-        cibw_python: [ "cp310", "cp311", "cp312", "cp313" ]
+        cibw_python: [ "cp310", "cp311", "cp312", "cp313", "cp314" ]
         cibw_arch: [ "x86_64", "arm64" ]
 
     steps:
@@ -131,7 +131,7 @@ jobs:
       matrix:
         cibw_buildlinux: [ manylinux, musllinux ]
         cibw_arch: [ "x86_64", "aarch64" ]
-        cibw_python: [ "cp310", "cp311", "cp312", "cp313" ]
+        cibw_python: [ "cp310", "cp311", "cp312", "cp313", "cp314" ]
 
     steps:
       - uses: actions/checkout@v4.2.2
@@ -183,7 +183,7 @@ jobs:
       - name: 🛠 Build Hexhamming Python C extension
         run: cibuildwheel
         env:
-          CIBW_BUILD: "cp310-* cp311-* cp312-* cp313-*"
+          CIBW_BUILD: "cp310-* cp311-* cp312-* cp313-* cp314-*"
           CIBW_ARCHS_WINDOWS: "AMD64"
           CIBW_TEST_REQUIRES: pytest pytest-benchmark
           CIBW_TEST_COMMAND: "pytest -s {project}/test"

src/python_hexhamming.cc

@@ -1,3 +1,4 @@
+#include <cstdlib>
 #include <cstring>
 #include <string.h>
 #define PY_SSIZE_T_CLEAN
@@ -334,7 +335,7 @@ static PyObject * check_bytes_arrays_first_within_dist_wrapper(PyObject *self, P
 
     Py_END_ALLOW_THREADS
 
-    return Py_BuildValue("i", ret);
+    return Py_BuildValue("L", ret);
 }
 
 /**
@@ -376,7 +377,7 @@ static PyObject * check_bytes_arrays_best_within_dist_wrapper(PyObject *self, Py
         return NULL;
     }
 
-    int64_t best_dist = max_dist + 1;
+    int64_t best_dist = -1;
     int64_t best_index = -1;
 
     Py_BEGIN_ALLOW_THREADS
@@ -385,23 +386,19 @@ static PyObject * check_bytes_arrays_best_within_dist_wrapper(PyObject *self, Py
     uint64_t number_of_elements = big_array_size / small_array_size;
     uint8_t* pBig = big_array;
     for (uint64_t i = 0; i < number_of_elements; i++, pBig += small_array_size) {
-        if (ptr__hamming_distance_bytes(pBig, small_array, small_array_size, best_dist - 1) == 0)
+        int64_t threshold = (best_dist >= 0) ? best_dist - 1 : max_dist;
+        if (ptr__hamming_distance_bytes(pBig, small_array, small_array_size, threshold) == 0)
             continue;
         dist = ptr__hamming_distance_bytes(pBig, small_array, small_array_size, -1);
-        if (dist < best_dist) {
+        if (best_dist < 0 || (int64_t)dist < best_dist) {
             best_dist = dist;
             best_index = i;
         }
     }
-
-    // Anything found? If not, set dist to -1
-    if (best_index == -1) {
-        best_dist = -1;
-    }
 
     Py_END_ALLOW_THREADS
 
-    return Py_BuildValue("ii", best_dist, best_index);
+    return Py_BuildValue("LL", best_dist, best_index);
 }
 
 /**
@@ -444,7 +441,11 @@ static PyObject * check_bytes_arrays_all_within_dist_wrapper(PyObject *self, PyO
     }
 
     uint64_t number_of_elements = big_array_size / small_array_size;
-    uint64_t *out = new uint64_t[number_of_elements * 2];
+    uint64_t *out = (uint64_t *)malloc(number_of_elements * 2 * sizeof(uint64_t));
+    if (out == NULL) {
+        PyErr_NoMemory();
+        return NULL;
+    }
     uint64_t *o = out;
 
     Py_BEGIN_ALLOW_THREADS
@@ -465,25 +466,25 @@ static PyObject * check_bytes_arrays_all_within_dist_wrapper(PyObject *self, PyO
     PyObject *my_list = PyList_New(0);
     if (my_list == NULL)
     {
-        delete[] out;
+        free(out);
         PyErr_NoMemory();
         return NULL;
     }
 
     for (uint64_t *op = out; op < o; op += 2)
     {
-        PyObject *tup = Py_BuildValue("ii", op[0], op[1]);
+        PyObject *tup = Py_BuildValue("KK", op[0], op[1]);
         if (tup == NULL)
         {
-            delete[] out;
+            free(out);
             Py_DECREF(my_list);
             PyErr_NoMemory();
             return NULL;
         }
 
         if (PyList_Append(my_list, tup) == -1)
         {
-            delete[] out;
+            free(out);
             Py_DECREF(tup);
             Py_DECREF(my_list);
             PyErr_NoMemory();
@@ -492,7 +493,7 @@ static PyObject * check_bytes_arrays_all_within_dist_wrapper(PyObject *self, PyO
         Py_DECREF(tup);
     }
 
-    delete[] out;
+    free(out);
 
     return my_list;
 }

src/python_hexhamming.h

@@ -555,7 +555,7 @@ inline uint64_t hamming_distance_loop_string(const char* a, const char* b, const
     }
     static uint64_t hamming_distance_bytes__extra(const uint8_t* a, const uint8_t* b,
                                                   const uint64_t length, const int64_t max_dist) {
-        if (max_dist > 0)
+        if (max_dist >= 0)
             return hamming_distance_bytes__native(a, b, length, max_dist);   //This is faster on ARMs.
         uint64_t difference = 0;
         uint64_t i = 0;