py: Implement iterator support for object that has __getitem__.

Addresses Issue #203.

py/obj.h

@@ -39,7 +39,7 @@ typedef struct _mp_obj_base_t mp_obj_base_t;
 #define MP_OBJ_IS_SMALL_INT(o) ((((mp_small_int_t)(o)) & 1) != 0)
 #define MP_OBJ_IS_QSTR(o) ((((mp_small_int_t)(o)) & 3) == 2)
 #define MP_OBJ_IS_OBJ(o) ((((mp_small_int_t)(o)) & 3) == 0)
-#define MP_OBJ_IS_TYPE(o, t) (MP_OBJ_IS_OBJ(o) && (((mp_obj_base_t*)(o))->type == (t)))
+#define MP_OBJ_IS_TYPE(o, t) (MP_OBJ_IS_OBJ(o) && (((mp_obj_base_t*)(o))->type == (t))) // this does not work for checking a string, use below macro for that
 #define MP_OBJ_IS_STR(o) (MP_OBJ_IS_QSTR(o) || MP_OBJ_IS_TYPE(o, &str_type))
 
 #define MP_OBJ_SMALL_INT_VALUE(o) (((mp_small_int_t)(o)) >> 1)
@@ -231,6 +231,7 @@ mp_obj_t mp_obj_new_dict(int n_args);
 mp_obj_t mp_obj_new_set(int n_args, mp_obj_t *items);
 mp_obj_t mp_obj_new_slice(mp_obj_t start, mp_obj_t stop, mp_obj_t step);
 mp_obj_t mp_obj_new_bound_meth(mp_obj_t meth, mp_obj_t self);
+mp_obj_t mp_obj_new_getitem_iter(mp_obj_t *args);
 mp_obj_t mp_obj_new_module(qstr module_name);
 
 mp_obj_type_t *mp_obj_get_type(mp_obj_t o_in);

py/objgetitemiter.c

+#include <stdlib.h>
+#include <stdint.h>
+
+#include "nlr.h"
+#include "misc.h"
+#include "mpconfig.h"
+#include "qstr.h"
+#include "obj.h"
+#include "runtime.h"
+
+// this is a wrapper object that is turns something that has a __getitem__ method into an iterator
+
+typedef struct _mp_obj_getitem_iter_t {
+    mp_obj_base_t base;
+    mp_obj_t args[3];
+} mp_obj_getitem_iter_t;
+
+static mp_obj_t it_iternext(mp_obj_t self_in) {
+    mp_obj_getitem_iter_t *self = self_in;
+    nlr_buf_t nlr;
+    if (nlr_push(&nlr) == 0) {
+        // try to get next item
+        mp_obj_t value = rt_call_method_n_kw(1, 0, self->args);
+        self->args[2] = MP_OBJ_NEW_SMALL_INT(MP_OBJ_SMALL_INT_VALUE(self->args[2]) + 1);
+        nlr_pop();
+        return value;
+    } else {
+        // an exception was raised
+        if (MP_OBJ_IS_TYPE(nlr.ret_val, &exception_type) && mp_obj_exception_get_type(nlr.ret_val) == MP_QSTR_StopIteration) {
+            // return mp_const_stop_iteration instead of raising StopIteration
+            return mp_const_stop_iteration;
+        } else {
+            // re-raise exception
+            nlr_jump(nlr.ret_val);
+        }
+    }
+}
+
+static const mp_obj_type_t it_type = {
+    { &mp_const_type },
+    "iterator",
+    .iternext = it_iternext
+};
+
+// args are those returned from rt_load_method_maybe (ie either an attribute or a method)
+mp_obj_t mp_obj_new_getitem_iter(mp_obj_t *args) {
+    mp_obj_getitem_iter_t *o = m_new_obj(mp_obj_getitem_iter_t);
+    o->base.type = &it_type;
+    o->args[0] = args[0];
+    o->args[1] = args[1];
+    o->args[2] = MP_OBJ_NEW_SMALL_INT(0);
+    return o;
+}

py/objstr.c

@@ -77,7 +77,7 @@ mp_obj_t str_binary_op(int op, mp_obj_t lhs_in, mp_obj_t rhs_in) {
             if (MP_OBJ_IS_SMALL_INT(rhs_in)) {
                 uint index = mp_get_index(mp_obj_get_type(lhs_in), lhs_len, rhs_in);
                 if (MP_OBJ_IS_TYPE(lhs_in, &bytes_type)) {
-                    return MP_OBJ_NEW_SMALL_INT(lhs_data[index]);
+                    return MP_OBJ_NEW_SMALL_INT((mp_small_int_t)lhs_data[index]);
                 } else {
                     return mp_obj_new_str(lhs_data + index, 1, true);
                 }
@@ -549,7 +549,7 @@ mp_obj_t bytes_it_iternext(mp_obj_t self_in) {
     mp_obj_str_it_t *self = self_in;
     GET_STR_DATA_LEN(self->str, str, len);
     if (self->cur < len) {
-        mp_obj_t o_out = MP_OBJ_NEW_SMALL_INT(str[self->cur]);
+        mp_obj_t o_out = MP_OBJ_NEW_SMALL_INT((mp_small_int_t)str[self->cur]);
         self->cur += 1;
         return o_out;
     } else {

py/objtype.c

@@ -117,8 +117,8 @@ static mp_obj_t class_make_new(mp_obj_t self_in, uint n_args, uint n_kw, const m
 }
 
 // TODO somehow replace const char * with a qstr
-static const char *binary_op_method_name[] = {
-    [RT_BINARY_OP_SUBSCR] = "__getitem__",
+static const qstr binary_op_method_name[] = {
+    [RT_BINARY_OP_SUBSCR] = MP_QSTR___getitem__,
     /*
     RT_BINARY_OP_OR,
     RT_BINARY_OP_XOR,
@@ -126,8 +126,8 @@ static const char *binary_op_method_name[] = {
     RT_BINARY_OP_LSHIFT,
     RT_BINARY_OP_RSHIFT,
     */
-    [RT_BINARY_OP_ADD] = "__add__",
-    [RT_BINARY_OP_SUBTRACT] = "__sub__",
+    [RT_BINARY_OP_ADD] = MP_QSTR___add__,
+    [RT_BINARY_OP_SUBTRACT] = MP_QSTR___sub__,
     /*
     RT_BINARY_OP_MULTIPLY,
     RT_BINARY_OP_FLOOR_DIVIDE,
@@ -157,16 +157,16 @@ static const char *binary_op_method_name[] = {
     RT_COMPARE_OP_IS,
     RT_COMPARE_OP_IS_NOT,
     */
-    [RT_COMPARE_OP_EXCEPTION_MATCH] = "__not_implemented__",
+    [RT_COMPARE_OP_EXCEPTION_MATCH] = MP_QSTR_, // not implemented, used to make sure array has full size
 };
 
 static mp_obj_t class_binary_op(int op, mp_obj_t lhs_in, mp_obj_t rhs_in) {
     mp_obj_class_t *lhs = lhs_in;
-    const char *op_name = binary_op_method_name[op];
-    if (op_name == NULL) {
+    qstr op_name = binary_op_method_name[op];
+    if (op_name == 0) {
         return MP_OBJ_NULL;
     }
-    mp_obj_t member = mp_obj_class_lookup(lhs->base.type, QSTR_FROM_STR_STATIC(op_name));
+    mp_obj_t member = mp_obj_class_lookup(lhs->base.type, op_name);
     if (member != MP_OBJ_NULL) {
         return rt_call_function_2(member, lhs_in, rhs_in);
     } else {

py/py.mk

@@ -47,6 +47,7 @@ PY_O_BASENAME = \
 	objfloat.o \
 	objfun.o \
 	objgenerator.o \
+	objgetitemiter.o \
 	objint.o \
 	objint_longlong.o \
 	objlist.o \

py/qstrdefs.h

@@ -13,6 +13,10 @@ Q(__next__)
 Q(__qualname__)
 Q(__repl_print__)
 
+Q(__getitem__)
+Q(__add__)
+Q(__sub__)
+
 Q(micropython)
 Q(byte_code)
 Q(native)

py/runtime.c

@@ -87,7 +87,7 @@ void rt_init(void) {
     // init loaded modules table
     mp_map_init(&map_loaded_modules, 3);
 
-    // built-in exceptions (TODO, make these proper classes)
+    // built-in exceptions (TODO, make these proper classes, and const if possible)
     mp_map_add_qstr(&map_builtins, MP_QSTR_AttributeError, mp_obj_new_exception(MP_QSTR_AttributeError));
     mp_map_add_qstr(&map_builtins, MP_QSTR_IndexError, mp_obj_new_exception(MP_QSTR_IndexError));
     mp_map_add_qstr(&map_builtins, MP_QSTR_KeyError, mp_obj_new_exception(MP_QSTR_KeyError));
@@ -100,6 +100,7 @@ void rt_init(void) {
     mp_map_add_qstr(&map_builtins, MP_QSTR_OverflowError, mp_obj_new_exception(MP_QSTR_OverflowError));
     mp_map_add_qstr(&map_builtins, MP_QSTR_OSError, mp_obj_new_exception(MP_QSTR_OSError));
     mp_map_add_qstr(&map_builtins, MP_QSTR_AssertionError, mp_obj_new_exception(MP_QSTR_AssertionError));
+    mp_map_add_qstr(&map_builtins, MP_QSTR_StopIteration, mp_obj_new_exception(MP_QSTR_StopIteration));
 
     // built-in objects
     mp_map_add_qstr(&map_builtins, MP_QSTR_Ellipsis, mp_const_ellipsis);
@@ -805,7 +806,7 @@ mp_obj_t rt_load_attr(mp_obj_t base, qstr attr) {
     // use load_method
     mp_obj_t dest[2];
     rt_load_method(base, attr, dest);
-    if (dest[1] == NULL) {
+    if (dest[1] == MP_OBJ_NULL) {
         // load_method returned just a normal attribute
         return dest[0];
     } else {
@@ -814,9 +815,10 @@ mp_obj_t rt_load_attr(mp_obj_t base, qstr attr) {
     }
 }
 
-void rt_load_method(mp_obj_t base, qstr attr, mp_obj_t *dest) {
-    DEBUG_OP_printf("load method %p.%s\n", base, qstr_str(attr));
-
+// no attribute found, returns:     dest[0] == MP_OBJ_NULL, dest[1] == MP_OBJ_NULL
+// normal attribute found, returns: dest[0] == <attribute>, dest[1] == MP_OBJ_NULL
+// method attribute found, returns: dest[0] == <method>,    dest[1] == <self>
+static void rt_load_method_maybe(mp_obj_t base, qstr attr, mp_obj_t *dest) {
     // clear output to indicate no attribute/method found yet
     dest[0] = MP_OBJ_NULL;
     dest[1] = MP_OBJ_NULL;
@@ -830,7 +832,7 @@ void rt_load_method(mp_obj_t base, qstr attr, mp_obj_t *dest) {
     }
 
     // if nothing found yet, look for built-in and generic names
-    if (dest[0] == NULL) {
+    if (dest[0] == MP_OBJ_NULL) {
         if (attr == MP_QSTR___next__ && type->iternext != NULL) {
             dest[0] = (mp_obj_t)&mp_builtin_next_obj;
             dest[1] = base;
@@ -861,8 +863,14 @@ void rt_load_method(mp_obj_t base, qstr attr, mp_obj_t *dest) {
             }
         }
     }
+}
+
+void rt_load_method(mp_obj_t base, qstr attr, mp_obj_t *dest) {
+    DEBUG_OP_printf("load method %p.%s\n", base, qstr_str(attr));
+
+    rt_load_method_maybe(base, attr, dest);
 
-    if (dest[0] == NULL) {
+    if (dest[0] == MP_OBJ_NULL) {
         // no attribute/method called attr
         // following CPython, we give a more detailed error message for type objects
         if (MP_OBJ_IS_TYPE(base, &mp_const_type)) {
@@ -910,7 +918,16 @@ mp_obj_t rt_getiter(mp_obj_t o_in) {
     if (type->getiter != NULL) {
         return type->getiter(o_in);
     } else {
-        nlr_jump(mp_obj_new_exception_msg_varg(MP_QSTR_TypeError, "'%s' object is not iterable", type->name));
+        // check for __getitem__ method
+        mp_obj_t dest[2];
+        rt_load_method_maybe(o_in, qstr_from_str("__getitem__"), dest);
+        if (dest[0] != MP_OBJ_NULL) {
+            // __getitem__ exists, create an iterator
+            return mp_obj_new_getitem_iter(dest);
+        } else {
+            // object not iterable
+            nlr_jump(mp_obj_new_exception_msg_varg(MP_QSTR_TypeError, "'%s' object is not iterable", type->name));
+        }
     }
 }
 

tests/basics/getitem.py

+# create a class that has a __getitem__ method
+class A:
+    def __getitem__(self, index):
+        print('getitem', index)
+        if index > 10:
+            raise StopIteration
+
+# test __getitem__
+A()[0]
+A()[1]
+
+# iterate using a for loop
+for i in A():
+    pass
+
+# iterate manually
+it = iter(A())
+try:
+    while True:
+        next(it)
+except StopIteration:
+    pass