py: Fix check of small-int overflow when parsing ints.

Also unifies use of SMALL_INT_FITS macro across parser and runtime.

11 files changed, 37 insertions, 34 deletions

diff --git a/py/compile.c b/py/compile.c
index f925c8c1ff..31ecbf0b91 100644
--- a/py/compile.c
+++ b/py/compile.c
@@ -249,7 +249,7 @@ STATIC mp_parse_node_t fold_constants(compiler_t *comp, mp_parse_node_t pn, mp_m
                         // shouldn't happen
                         assert(0);
                     }
-                    if (MP_PARSE_FITS_SMALL_INT(arg0)) {
+                    if (MP_SMALL_INT_FITS(arg0)) {
                         //printf("%ld + %ld\n", arg0, arg1);
                         pn = mp_parse_node_new_leaf(MP_PARSE_NODE_SMALL_INT, arg0);
                     }
@@ -264,7 +264,7 @@ STATIC mp_parse_node_t fold_constants(compiler_t *comp, mp_parse_node_t pn, mp_m
                         // int * int
                         if (!mp_small_int_mul_overflow(arg0, arg1)) {
                             arg0 *= arg1;
-                            if (MP_PARSE_FITS_SMALL_INT(arg0)) {
+                            if (MP_SMALL_INT_FITS(arg0)) {
                                 pn = mp_parse_node_new_leaf(MP_PARSE_NODE_SMALL_INT, arg0);
                             }
                         }
@@ -337,7 +337,7 @@ STATIC mp_parse_node_t fold_constants(compiler_t *comp, mp_parse_node_t pn, mp_m
                         mp_load_method_maybe(elem->value, q_attr, dest);
                         if (MP_OBJ_IS_SMALL_INT(dest[0]) && dest[1] == NULL) {
                             machine_int_t val = MP_OBJ_SMALL_INT_VALUE(dest[0]);
-                            if (MP_PARSE_FITS_SMALL_INT(val)) {
+                            if (MP_SMALL_INT_FITS(val)) {
                                 pn = mp_parse_node_new_leaf(MP_PARSE_NODE_SMALL_INT, val);
                             }
                         }
diff --git a/py/obj.h b/py/obj.h
index 74bdc79774..185fe14db0 100644
--- a/py/obj.h
+++ b/py/obj.h
@@ -72,10 +72,6 @@ typedef struct _mp_obj_base_t mp_obj_base_t;
 
 // These macros check for small int, qstr or object, and access small int and qstr values
 
-// In SMALL_INT, next-to-highest bits is used as sign, so both must match for value in range
-#define MP_SMALL_INT_MIN ((mp_small_int_t)(((machine_int_t)WORD_MSBIT_HIGH) >> 1))
-#define MP_SMALL_INT_MAX ((mp_small_int_t)(~(MP_SMALL_INT_MIN)))
-#define MP_OBJ_FITS_SMALL_INT(n) ((((n) ^ ((n) << 1)) & WORD_MSBIT_HIGH) == 0)
 // these macros have now become inline functions; see below
 //#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)
diff --git a/py/objint.c b/py/objint.c
index 7d6258b7d4..f631d698f3 100644
--- a/py/objint.c
+++ b/py/objint.c
@@ -35,6 +35,7 @@
 #include "qstr.h"
 #include "obj.h"
 #include "parsenum.h"
+#include "smallint.h"
 #include "mpz.h"
 #include "objint.h"
 #include "runtime0.h"
@@ -251,7 +252,7 @@ mp_obj_t mp_obj_new_int_from_uint(machine_uint_t value) {
 }
 
 mp_obj_t mp_obj_new_int(machine_int_t value) {
-    if (MP_OBJ_FITS_SMALL_INT(value)) {
+    if (MP_SMALL_INT_FITS(value)) {
         return MP_OBJ_NEW_SMALL_INT(value);
     }
     nlr_raise(mp_obj_new_exception_msg(&mp_type_OverflowError, "small int overflow"));
diff --git a/py/objint_longlong.c b/py/objint_longlong.c
index 381246dfe4..82db9e6608 100644
--- a/py/objint_longlong.c
+++ b/py/objint_longlong.c
@@ -34,6 +34,7 @@
 #include "misc.h"
 #include "qstr.h"
 #include "obj.h"
+#include "smallint.h"
 #include "mpz.h"
 #include "objint.h"
 #include "runtime0.h"
@@ -140,7 +141,7 @@ mp_obj_t mp_obj_int_binary_op(int op, mp_obj_t lhs_in, mp_obj_t rhs_in) {
 }
 
 mp_obj_t mp_obj_new_int(machine_int_t value) {
-    if (MP_OBJ_FITS_SMALL_INT(value)) {
+    if (MP_SMALL_INT_FITS(value)) {
         return MP_OBJ_NEW_SMALL_INT(value);
     }
     return mp_obj_new_int_from_ll(value);
diff --git a/py/objint_mpz.c b/py/objint_mpz.c
index 733dc096a6..9cdbb71688 100644
--- a/py/objint_mpz.c
+++ b/py/objint_mpz.c
@@ -35,6 +35,7 @@
 #include "qstr.h"
 #include "parsenumbase.h"
 #include "obj.h"
+#include "smallint.h"
 #include "mpz.h"
 #include "objint.h"
 #include "runtime0.h"
@@ -239,7 +240,7 @@ mp_obj_t mp_obj_int_binary_op(int op, mp_obj_t lhs_in, mp_obj_t rhs_in) {
 }
 
 mp_obj_t mp_obj_new_int(machine_int_t value) {
-    if (MP_OBJ_FITS_SMALL_INT(value)) {
+    if (MP_SMALL_INT_FITS(value)) {
         return MP_OBJ_NEW_SMALL_INT(value);
     }
     return mp_obj_new_int_from_ll(value);
diff --git a/py/parse.c b/py/parse.c
index 93bced843b..af09c335f2 100644
--- a/py/parse.c
+++ b/py/parse.c
@@ -36,6 +36,7 @@
 #include "lexer.h"
 #include "parsenumbase.h"
 #include "parse.h"
+#include "smallint.h"
 
 #define RULE_ACT_KIND_MASK      (0xf0)
 #define RULE_ACT_ARG_MASK       (0x0f)
@@ -311,13 +312,13 @@ STATIC void push_result_token(parser_t *parser, const mp_lexer_t *lex) {
         int i = mp_parse_num_base(str, len, &base);
         bool overflow = false;
         for (; i < len; i++) {
-            machine_int_t old_val = int_val;
+            int dig;
             if (unichar_isdigit(str[i]) && str[i] - '0' < base) {
-                int_val = base * int_val + str[i] - '0';
+                dig = str[i] - '0';
             } else if (base == 16 && 'a' <= str[i] && str[i] <= 'f') {
-                int_val = base * int_val + str[i] - 'a' + 10;
+                dig = str[i] - 'a' + 10;
             } else if (base == 16 && 'A' <= str[i] && str[i] <= 'F') {
-                int_val = base * int_val + str[i] - 'A' + 10;
+                dig = str[i] - 'A' + 10;
             } else if (str[i] == '.' || str[i] == 'e' || str[i] == 'E' || str[i] == 'j' || str[i] == 'J') {
                 dec = true;
                 break;
@@ -325,17 +326,18 @@ STATIC void push_result_token(parser_t *parser, const mp_lexer_t *lex) {
                 small_int = false;
                 break;
             }
-            if (int_val < old_val) {
-                // If new value became less than previous, it's overflow
+            // add next digi and check for overflow
+            if (mp_small_int_mul_overflow(int_val, base)) {
                 overflow = true;
-            } else if ((old_val ^ int_val) & WORD_MSBIT_HIGH) {
-                // If signed number changed sign - it's overflow
+            }
+            int_val = int_val * base + dig;
+            if (!MP_SMALL_INT_FITS(int_val)) {
                 overflow = true;
             }
         }
         if (dec) {
             pn = mp_parse_node_new_leaf(MP_PARSE_NODE_DECIMAL, qstr_from_strn(str, len));
-        } else if (small_int && !overflow && MP_PARSE_FITS_SMALL_INT(int_val)) {
+        } else if (small_int && !overflow && MP_SMALL_INT_FITS(int_val)) {
             pn = mp_parse_node_new_leaf(MP_PARSE_NODE_SMALL_INT, int_val);
         } else {
             pn = mp_parse_node_new_leaf(MP_PARSE_NODE_INTEGER, qstr_from_strn(str, len));
diff --git a/py/parse.h b/py/parse.h
index 43c74e73b8..6950139e7b 100644
--- a/py/parse.h
+++ b/py/parse.h
@@ -37,13 +37,6 @@ struct _mp_lexer_t;
 //  - xx...x10010: a string of bytes; bits 5 and above are the qstr holding the value
 //  - xx...x10110: a token; bits 5 and above are mp_token_kind_t
 
-// TODO: these can now be unified with MP_OBJ_FITS_SMALL_INT(x)
-// makes sure the top 2 bits of x are all cleared (positive number) or all set (negavite number)
-// these macros can probably go somewhere else because they are used more than just in the parser
-#define MP_UINT_HIGH_2_BITS (~((~((machine_uint_t)0)) >> 2))
-// parser's small ints are different from VM small int
-#define MP_PARSE_FITS_SMALL_INT(x) (((((machine_uint_t)(x)) & MP_UINT_HIGH_2_BITS) == 0) || ((((machine_uint_t)(x)) & MP_UINT_HIGH_2_BITS) == MP_UINT_HIGH_2_BITS))
-
 #define MP_PARSE_NODE_NULL      (0)
 #define MP_PARSE_NODE_SMALL_INT (0x1)
 #define MP_PARSE_NODE_ID        (0x02)
diff --git a/py/parsenum.c b/py/parsenum.c
index 842a9e9599..9729ffe64a 100644
--- a/py/parsenum.c
+++ b/py/parsenum.c
@@ -34,6 +34,7 @@
 #include "obj.h"
 #include "parsenumbase.h"
 #include "parsenum.h"
+#include "smallint.h"
 
 #if MICROPY_ENABLE_FLOAT
 #include <math.h>
@@ -70,16 +71,16 @@ mp_obj_t mp_parse_num_integer(const char *restrict str, uint len, int base) {
     machine_int_t int_val = 0;
     const char *restrict str_val_start = str;
     for (; str < top; str++) {
-        machine_int_t old_val = int_val;
+        // get next digit as a value
         int dig = *str;
         if (unichar_isdigit(dig) && dig - '0' < base) {
             // 0-9 digit
-            int_val = base * int_val + dig - '0';
+            dig = dig - '0';
         } else if (base == 16) {
             dig |= 0x20;
             if ('a' <= dig && dig <= 'f') {
                 // a-f hex digit
-                int_val = base * int_val + dig - 'a' + 10;
+                dig = dig - 'a' + 10;
             } else {
                 // unknown character
                 break;
@@ -88,11 +89,13 @@ mp_obj_t mp_parse_num_integer(const char *restrict str, uint len, int base) {
             // unknown character
             break;
         }
-        if (int_val < old_val) {
-            // If new value became less than previous, it's overflow
+
+        // add next digi and check for overflow
+        if (mp_small_int_mul_overflow(int_val, base)) {
             goto overflow;
-        } else if ((old_val ^ int_val) & WORD_MSBIT_HIGH) {
-            // If signed number changed sign - it's overflow
+        }
+        int_val = int_val * base + dig;
+        if (!MP_SMALL_INT_FITS(int_val)) {
             goto overflow;
         }
     }
diff --git a/py/runtime.c b/py/runtime.c
index f487807256..a9d57460ac 100644
--- a/py/runtime.c
+++ b/py/runtime.c
@@ -413,7 +413,7 @@ mp_obj_t mp_binary_op(int op, mp_obj_t lhs, mp_obj_t rhs) {
                     goto unsupported_op;
             }
             // TODO: We just should make mp_obj_new_int() inline and use that
-            if (MP_OBJ_FITS_SMALL_INT(lhs_val)) {
+            if (MP_SMALL_INT_FITS(lhs_val)) {
                 return MP_OBJ_NEW_SMALL_INT(lhs_val);
             } else {
                 return mp_obj_new_int(lhs_val);
diff --git a/py/smallint.c b/py/smallint.c
index 186c9c754f..5543f126c3 100644
--- a/py/smallint.c
+++ b/py/smallint.c
@@ -28,6 +28,7 @@
 #include "mpconfig.h"
 #include "qstr.h"
 #include "obj.h"
+#include "smallint.h"
 
 bool mp_small_int_mul_overflow(machine_int_t x, machine_int_t y) {
     // Check for multiply overflow; see CERT INT32-C
diff --git a/py/smallint.h b/py/smallint.h
index 9c1de42ad0..36b55f9197 100644
--- a/py/smallint.h
+++ b/py/smallint.h
@@ -26,6 +26,11 @@
 
 // Functions for small integer arithmetic
 
+// In SMALL_INT, next-to-highest bits is used as sign, so both must match for value in range
+#define MP_SMALL_INT_MIN ((mp_small_int_t)(((machine_int_t)WORD_MSBIT_HIGH) >> 1))
+#define MP_SMALL_INT_MAX ((mp_small_int_t)(~(MP_SMALL_INT_MIN)))
+#define MP_SMALL_INT_FITS(n) ((((n) ^ ((n) << 1)) & WORD_MSBIT_HIGH) == 0)
+
 bool mp_small_int_mul_overflow(machine_int_t x, machine_int_t y);
 machine_int_t mp_small_int_modulo(machine_int_t dividend, machine_int_t divisor);
 machine_int_t mp_small_int_floor_divide(machine_int_t num, machine_int_t denom);