ENH: Make internal per-op nditer flags type a typedef

There are 8 flags now, so to add another one requires
bumping up the size. This change makes that easy.
Also, because the internal structure of nditer is
not exposed to the ABI, we can change it at will.

3 files changed, 36 insertions, 34 deletions

diff --git a/numpy/core/src/multiarray/nditer_api.c b/numpy/core/src/multiarray/nditer_api.c
index a433df406..4e2005a31 100644
--- a/numpy/core/src/multiarray/nditer_api.c
+++ b/numpy/core/src/multiarray/nditer_api.c
@@ -771,7 +771,7 @@ NpyIter_RequiresBuffering(NpyIter *iter)
     /*int ndim = NIT_NDIM(iter);*/
     int iop, nop = NIT_NOP(iter);
 
-    char *op_itflags;
+    npyiter_opitflags *op_itflags;
 
     if (!(itflags&NPY_ITFLAG_BUFFER)) {
         return 0;
@@ -1217,7 +1217,7 @@ NpyIter_GetReadFlags(NpyIter *iter, char *outreadflags)
     /*int ndim = NIT_NDIM(iter);*/
     int iop, nop = NIT_NOP(iter);
 
-    char *op_itflags = NIT_OPITFLAGS(iter);
+    npyiter_opitflags *op_itflags = NIT_OPITFLAGS(iter);
 
     for (iop = 0; iop < nop; ++iop) {
         outreadflags[iop] = (op_itflags[iop]&NPY_OP_ITFLAG_READ) != 0;
@@ -1234,7 +1234,7 @@ NpyIter_GetWriteFlags(NpyIter *iter, char *outwriteflags)
     /*int ndim = NIT_NDIM(iter);*/
     int iop, nop = NIT_NOP(iter);
 
-    char *op_itflags = NIT_OPITFLAGS(iter);
+    npyiter_opitflags *op_itflags = NIT_OPITFLAGS(iter);
 
     for (iop = 0; iop < nop; ++iop) {
         outwriteflags[iop] = (op_itflags[iop]&NPY_OP_ITFLAG_WRITE) != 0;
@@ -1330,7 +1330,7 @@ NpyIter_GetInnerFixedStrideArray(NpyIter *iter, npy_intp *out_strides)
 
     if (itflags&NPY_ITFLAG_BUFFER) {
         NpyIter_BufferData *data = NIT_BUFFERDATA(iter);
-        char *op_itflags = NIT_OPITFLAGS(iter);
+        npyiter_opitflags *op_itflags = NIT_OPITFLAGS(iter);
         npy_intp stride, *strides = NBF_STRIDES(data),
                 *ad_strides = NAD_STRIDES(axisdata0);
         PyArray_Descr **dtypes = NIT_DTYPES(iter);
@@ -1751,14 +1751,14 @@ npyiter_allocate_buffers(NpyIter *iter, char **errmsg)
     int iop = 0, nop = NIT_NOP(iter);
 
     npy_intp i;
-    char *op_itflags = NIT_OPITFLAGS(iter);
+    npyiter_opitflags *op_itflags = NIT_OPITFLAGS(iter);
     NpyIter_BufferData *bufferdata = NIT_BUFFERDATA(iter);
     PyArray_Descr **op_dtype = NIT_DTYPES(iter);
     npy_intp buffersize = NBF_BUFFERSIZE(bufferdata);
     char *buffer, **buffers = NBF_BUFFERS(bufferdata);
 
     for (iop = 0; iop < nop; ++iop) {
-        char flags = op_itflags[iop];
+        npyiter_opitflags flags = op_itflags[iop];
 
         /*
          * If we have determined that a buffer may be needed,
@@ -1889,7 +1889,7 @@ npyiter_copy_from_buffers(NpyIter *iter)
     int maskop = NIT_MASKOP(iter);
     int first_maskna_op = NIT_FIRST_MASKNA_OP(iter);
 
-    char *op_itflags = NIT_OPITFLAGS(iter);
+    npyiter_opitflags *op_itflags = NIT_OPITFLAGS(iter);
     NpyIter_BufferData *bufferdata = NIT_BUFFERDATA(iter);
     NpyIter_AxisData *axisdata = NIT_AXISDATA(iter),
                     *reduce_outeraxisdata = NULL;
@@ -2116,7 +2116,7 @@ npyiter_copy_to_buffers(NpyIter *iter, char **prev_dataptrs)
     int iop, nop = NIT_NOP(iter);
     int first_maskna_op = NIT_FIRST_MASKNA_OP(iter);
 
-    char *op_itflags = NIT_OPITFLAGS(iter);
+    npyiter_opitflags *op_itflags = NIT_OPITFLAGS(iter);
     NpyIter_BufferData *bufferdata = NIT_BUFFERDATA(iter);
     NpyIter_AxisData *axisdata = NIT_AXISDATA(iter),
                     *reduce_outeraxisdata = NULL;
@@ -2636,7 +2636,7 @@ npyiter_checkreducesize(NpyIter *iter, npy_intp count,
     npy_intp reducespace = 1, factor;
     npy_bool nonzerocoord;
 
-    char *op_itflags = NIT_OPITFLAGS(iter);
+    npyiter_opitflags *op_itflags = NIT_OPITFLAGS(iter);
     char stride0op[NPY_MAXARGS];
 
     /* Default to no outer axis */
diff --git a/numpy/core/src/multiarray/nditer_constr.c b/numpy/core/src/multiarray/nditer_constr.c
index 941ce91c0..f1d0c5c38 100644
--- a/numpy/core/src/multiarray/nditer_constr.c
+++ b/numpy/core/src/multiarray/nditer_constr.c
@@ -27,14 +27,14 @@ static int
 npyiter_calculate_ndim(int nop, PyArrayObject **op_in,
                        int oa_ndim);
 static int
-npyiter_check_per_op_flags(npy_uint32 flags, char *op_itflags);
+npyiter_check_per_op_flags(npy_uint32 flags, npyiter_opitflags *op_itflags);
 static int
 npyiter_prepare_one_operand(PyArrayObject **op,
                         char **op_dataptr,
                         PyArray_Descr *op_request_dtype,
                         PyArray_Descr** op_dtype,
                         npy_uint32 flags,
-                        npy_uint32 op_flags, char *op_itflags);
+                        npy_uint32 op_flags, npyiter_opitflags *op_itflags);
 static int
 npyiter_prepare_operands(int nop, int first_maskna_op,
                     PyArrayObject **op_in,
@@ -43,16 +43,16 @@ npyiter_prepare_operands(int nop, int first_maskna_op,
                     PyArray_Descr **op_request_dtypes,
                     PyArray_Descr **op_dtype,
                     npy_uint32 flags,
-                    npy_uint32 *op_flags, char *op_itflags,
+                    npy_uint32 *op_flags, npyiter_opitflags *op_itflags,
                     npy_int8 *out_maskop,
                     npy_int8 *out_maskna_indices);
 static int
 npyiter_check_casting(int nop, PyArrayObject **op,
                     PyArray_Descr **op_dtype,
                     NPY_CASTING casting,
-                    char *op_itflags);
+                    npyiter_opitflags *op_itflags);
 static int
-npyiter_fill_axisdata(NpyIter *iter, npy_uint32 flags, char *op_itflags,
+npyiter_fill_axisdata(NpyIter *iter, npy_uint32 flags, npyiter_opitflags *op_itflags,
                     char **op_dataptr,
                     npy_uint32 *op_flags, int **op_axes,
                     npy_intp *itershape,
@@ -74,25 +74,25 @@ static void
 npyiter_find_best_axis_ordering(NpyIter *iter);
 static PyArray_Descr *
 npyiter_get_common_dtype(int first_maskna_op, PyArrayObject **op,
-                        char *op_itflags, PyArray_Descr **op_dtype,
+                        npyiter_opitflags *op_itflags, PyArray_Descr **op_dtype,
                         PyArray_Descr **op_request_dtypes,
                         int only_inputs, int output_scalars);
 static PyArrayObject *
 npyiter_new_temp_array(NpyIter *iter, PyTypeObject *subtype,
-                npy_uint32 flags, char *op_itflags,
+                npy_uint32 flags, npyiter_opitflags *op_itflags,
                 int op_ndim, npy_intp *shape,
                 PyArray_Descr *op_dtype, int *op_axes);
 static int
 npyiter_allocate_arrays(NpyIter *iter,
                         npy_uint32 flags,
                         PyArray_Descr **op_dtype, PyTypeObject *subtype,
-                        npy_uint32 *op_flags, char *op_itflags,
+                        npy_uint32 *op_flags, npyiter_opitflags *op_itflags,
                         int **op_axes, int output_scalars);
 static int
 npyiter_fill_maskna_axisdata(NpyIter *iter, int **op_axes);
 static void
 npyiter_get_priority_subtype(int first_maskna_op, PyArrayObject **op,
-                            char *op_itflags,
+                            npyiter_opitflags *op_itflags,
                             double *subtype_priority, PyTypeObject **subtype);
 static int
 npyiter_allocate_transfer_functions(NpyIter *iter);
@@ -120,7 +120,7 @@ NpyIter_AdvancedNew(int nop, PyArrayObject **op_in, npy_uint32 flags,
     /* Per-operand values */
     PyArrayObject **op;
     PyArray_Descr **op_dtype;
-    char *op_itflags;
+    npyiter_opitflags *op_itflags;
     char **op_dataptr;
 
     npy_int8 *perm;
@@ -916,7 +916,7 @@ npyiter_calculate_ndim(int nop, PyArrayObject **op_in,
  * Returns 1 on success, 0 on failure.
  */
 static int
-npyiter_check_per_op_flags(npy_uint32 op_flags, char *op_itflags)
+npyiter_check_per_op_flags(npy_uint32 op_flags, npyiter_opitflags *op_itflags)
 {
     if ((op_flags & NPY_ITER_GLOBAL_FLAGS) != 0) {
         PyErr_SetString(PyExc_ValueError,
@@ -1027,7 +1027,7 @@ npyiter_prepare_one_operand(PyArrayObject **op,
                         PyArray_Descr *op_request_dtype,
                         PyArray_Descr **op_dtype,
                         npy_uint32 flags,
-                        npy_uint32 op_flags, char *op_itflags)
+                        npy_uint32 op_flags, npyiter_opitflags *op_itflags)
 {
     /* NULL operands must be automatically allocated outputs */
     if (*op == NULL) {
@@ -1228,7 +1228,7 @@ npyiter_prepare_operands(int nop, int first_maskna_op, PyArrayObject **op_in,
                     PyArray_Descr **op_request_dtypes,
                     PyArray_Descr **op_dtype,
                     npy_uint32 flags,
-                    npy_uint32 *op_flags, char *op_itflags,
+                    npy_uint32 *op_flags, npyiter_opitflags *op_itflags,
                     npy_int8 *out_maskop,
                     npy_int8 *out_maskna_indices)
 {
@@ -1421,7 +1421,7 @@ static int
 npyiter_check_casting(int first_maskna_op, PyArrayObject **op,
                     PyArray_Descr **op_dtype,
                     NPY_CASTING casting,
-                    char *op_itflags)
+                    npyiter_opitflags *op_itflags)
 {
     int iop;
 
@@ -1559,7 +1559,7 @@ check_mask_for_writemasked_reduction(NpyIter *iter, int iop)
  * Returns 1 on success, 0 on failure.
  */
 static int
-npyiter_fill_axisdata(NpyIter *iter, npy_uint32 flags, char *op_itflags,
+npyiter_fill_axisdata(NpyIter *iter, npy_uint32 flags, npyiter_opitflags *op_itflags,
                     char **op_dataptr,
                     npy_uint32 *op_flags, int **op_axes,
                     npy_intp *itershape,
@@ -2522,7 +2522,7 @@ npyiter_find_best_axis_ordering(NpyIter *iter)
  */
 static PyArray_Descr *
 npyiter_get_common_dtype(int first_maskna_op, PyArrayObject **op,
-                        char *op_itflags, PyArray_Descr **op_dtype,
+                        npyiter_opitflags *op_itflags, PyArray_Descr **op_dtype,
                         PyArray_Descr **op_request_dtypes,
                         int only_inputs, int output_scalars)
 {
@@ -2588,7 +2588,7 @@ npyiter_get_common_dtype(int first_maskna_op, PyArrayObject **op,
  */
 static PyArrayObject *
 npyiter_new_temp_array(NpyIter *iter, PyTypeObject *subtype,
-                npy_uint32 flags, char *op_itflags,
+                npy_uint32 flags, npyiter_opitflags *op_itflags,
                 int op_ndim, npy_intp *shape,
                 PyArray_Descr *op_dtype, int *op_axes)
 {
@@ -2822,7 +2822,7 @@ static int
 npyiter_allocate_arrays(NpyIter *iter,
                         npy_uint32 flags,
                         PyArray_Descr **op_dtype, PyTypeObject *subtype,
-                        npy_uint32 *op_flags, char *op_itflags,
+                        npy_uint32 *op_flags, npyiter_opitflags *op_itflags,
                         int **op_axes, int output_scalars)
 {
     npy_uint32 itflags = NIT_ITFLAGS(iter);
@@ -3157,7 +3157,7 @@ npyiter_fill_maskna_axisdata(NpyIter *iter, int **op_axes)
     int first_maskna_op = NIT_FIRST_MASKNA_OP(iter);
     npy_int8 *perm;
 
-    char *op_itflags = NIT_OPITFLAGS(iter);
+    npyiter_opitflags *op_itflags = NIT_OPITFLAGS(iter);
     npy_int8 *maskna_indices = NIT_MASKNA_INDICES(iter);
     NpyIter_AxisData *axisdata;
     npy_intp sizeof_axisdata;
@@ -3291,7 +3291,7 @@ npyiter_fill_maskna_axisdata(NpyIter *iter, int **op_axes)
  */
 static void
 npyiter_get_priority_subtype(int first_maskna_op, PyArrayObject **op,
-                            char *op_itflags,
+                            npyiter_opitflags *op_itflags,
                             double *subtype_priority,
                             PyTypeObject **subtype)
 {
@@ -3317,7 +3317,7 @@ npyiter_allocate_transfer_functions(NpyIter *iter)
     int first_maskna_op = NIT_FIRST_MASKNA_OP(iter);
 
     npy_intp i;
-    char *op_itflags = NIT_OPITFLAGS(iter);
+    npyiter_opitflags *op_itflags = NIT_OPITFLAGS(iter);
     NpyIter_BufferData *bufferdata = NIT_BUFFERDATA(iter);
     NpyIter_AxisData *axisdata = NIT_AXISDATA(iter);
     PyArrayObject **op = NIT_OPERANDS(iter);
@@ -3334,7 +3334,7 @@ npyiter_allocate_transfer_functions(NpyIter *iter)
     int needs_api = 0;
 
     for (iop = 0; iop < nop; ++iop) {
-        char flags = op_itflags[iop];
+        npyiter_opitflags flags = op_itflags[iop];
         /*
          * Reduction operands may be buffered with a different stride,
          * so we must pass NPY_MAX_INTP to the transfer function factory.
diff --git a/numpy/core/src/multiarray/nditer_impl.h b/numpy/core/src/multiarray/nditer_impl.h
index ea02d5cc4..2b08436aa 100644
--- a/numpy/core/src/multiarray/nditer_impl.h
+++ b/numpy/core/src/multiarray/nditer_impl.h
@@ -147,6 +147,8 @@ struct NpyIter_InternalOnly {
 typedef struct NpyIter_AD NpyIter_AxisData;
 typedef struct NpyIter_BD NpyIter_BufferData;
 
+typedef npy_int8 npyiter_opitflags;
+
 /* Byte sizes of the iterator members */
 #define NIT_PERM_SIZEOF(itflags, ndim, nop) \
         NPY_INTP_ALIGNED(NPY_MAXDIMS)
@@ -161,7 +163,7 @@ typedef struct NpyIter_BD NpyIter_BufferData;
 #define NIT_OPERANDS_SIZEOF(itflags, ndim, nop) \
         ((NPY_SIZEOF_INTP)*(nop))
 #define NIT_OPITFLAGS_SIZEOF(itflags, ndim, nop) \
-        (NPY_INTP_ALIGNED(nop))
+        (NPY_INTP_ALIGNED(sizeof(npyiter_opitflags) * nop))
 #define NIT_BUFFERDATA_SIZEOF(itflags, ndim, nop) \
         ((itflags&NPY_ITFLAG_BUFFER) ? ((NPY_SIZEOF_INTP)*(6 + 9*nop)) : 0)
 
@@ -224,8 +226,8 @@ typedef struct NpyIter_BD NpyIter_BufferData;
         &(iter)->iter_flexdata + NIT_BASEOFFSETS_OFFSET(itflags, ndim, nop)))
 #define NIT_OPERANDS(iter) ((PyArrayObject **)( \
         &(iter)->iter_flexdata + NIT_OPERANDS_OFFSET(itflags, ndim, nop)))
-#define NIT_OPITFLAGS(iter) ( \
-        &(iter)->iter_flexdata + NIT_OPITFLAGS_OFFSET(itflags, ndim, nop))
+#define NIT_OPITFLAGS(iter) ((npyiter_opitflags *)( \
+        &(iter)->iter_flexdata + NIT_OPITFLAGS_OFFSET(itflags, ndim, nop)))
 #define NIT_BUFFERDATA(iter) ((NpyIter_BufferData *)( \
         &(iter)->iter_flexdata + NIT_BUFFERDATA_OFFSET(itflags, ndim, nop)))
 #define NIT_AXISDATA(iter) ((NpyIter_AxisData *)( \