py/parsenum: Extend mp_parse_num_integer() to parse long long.

If big integer support is 'long long' then mp_parse_num_integer() can
parse to it directly instead of failing over from small int. This means
strtoll() is no longer pulled in, and fixes some bugs parsing long long
integers (i.e. can now parse negative values correctly, can now parse
values which aren't NULL terminated).

The (default) smallint parsing compiled code should stay the same here,
macros and a typedef are used to abstract some parts of it out.

When bigint is long long we parse to 'unsigned long long' first (to avoid
the code size hit of pulling in signed 64-bit math routines) and the
convert to signed at the end.

One tricky case this routine correctly overflows on is
int("9223372036854775808") which is one more than LLONG_MAX in decimal. No
unit test case added for this as it's too hard to detect 64-bit long
integer mode.

This work was funded through GitHub Sponsors.

Signed-off-by: Angus Gratton <angus@redyak.com.au>

.github/workflows/ports_unix.yml

@@ -134,6 +134,20 @@ jobs:
       if: failure()
       run: tests/run-tests.py --print-failures
 
+  longlong:
+    runs-on: ubuntu-22.04 # use 22.04 to get python2, and libffi-dev:i386
+    steps:
+    - uses: actions/checkout@v4
+    - name: Install packages
+      run: source tools/ci.sh && ci_unix_32bit_setup
+    - name: Build
+      run: source tools/ci.sh && ci_unix_longlong_build
+    - name: Run main test suite
+      run: source tools/ci.sh && ci_unix_longlong_run_tests
+    - name: Print failures
+      if: failure()
+      run: tests/run-tests.py --print-failures
+
   float:
     runs-on: ubuntu-latest
     steps:

ports/unix/variants/longlong/mpconfigvariant.h

@@ -0,0 +1,37 @@
+/*
+ * This file is part of the MicroPython project, http://micropython.org/
+ *
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2016 Damien P. George
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+// This config exists to test that the MICROPY_LONGINT_IMPL_LONGLONG variant
+// (i.e. minimal form of "big integer" that's backed by 64-bit int only) builds
+// and passes tests.
+
+#define MICROPY_LONGINT_IMPL           (MICROPY_LONGINT_IMPL_LONGLONG)
+
+// Set base feature level.
+#define MICROPY_CONFIG_ROM_LEVEL (MICROPY_CONFIG_ROM_LEVEL_EXTRA_FEATURES)
+
+// Enable extra Unix features.
+#include "../mpconfigvariant_common.h"

ports/unix/variants/longlong/mpconfigvariant.mk

@@ -0,0 +1,8 @@
+# build interpreter with "bigints" implemented as "longlong"
+
+# otherwise, small int is essentially 64-bit
+MICROPY_FORCE_32BIT := 1
+
+MICROPY_PY_FFI := 0
+
+MPY_TOOL_FLAGS = -mlongint-impl longlong

py/misc.h

@@ -33,10 +33,15 @@
 #include <stdbool.h>
 #include <stdint.h>
 #include <stddef.h>
+#include <limits.h>
 
 typedef unsigned char byte;
 typedef unsigned int uint;
 
+#ifndef __has_builtin
+#define __has_builtin(x) (0)
+#endif
+
 /** generic ops *************************************************/
 
 #ifndef MIN
@@ -374,26 +379,23 @@ static inline bool mp_check(bool value) {
 static inline uint32_t mp_popcount(uint32_t x) {
     return __popcnt(x);
 }
-#else
+#else // _MSC_VER
 #define mp_clz(x) __builtin_clz(x)
 #define mp_clzl(x) __builtin_clzl(x)
 #define mp_clzll(x) __builtin_clzll(x)
 #define mp_ctz(x) __builtin_ctz(x)
 #define mp_check(x) (x)
-#if defined __has_builtin
 #if __has_builtin(__builtin_popcount)
 #define mp_popcount(x) __builtin_popcount(x)
-#endif
-#endif
-#if !defined(mp_popcount)
+#else
 static inline uint32_t mp_popcount(uint32_t x) {
     x = x - ((x >> 1) & 0x55555555);
     x = (x & 0x33333333) + ((x >> 2) & 0x33333333);
     x = (x + (x >> 4)) & 0x0F0F0F0F;
     return (x * 0x01010101) >> 24;
 }
-#endif
-#endif
+#endif // __has_builtin(__builtin_popcount)
+#endif // _MSC_VER
 
 #define MP_FIT_UNSIGNED(bits, value) (((value) & (~0U << (bits))) == 0)
 #define MP_FIT_SIGNED(bits, value) \
@@ -426,4 +428,93 @@ static inline uint32_t mp_clz_mpi(mp_int_t x) {
     #endif
 }
 
+// Overflow-checked operations for long long
+
+// Integer overflow builtins were added to GCC 5, but __has_builtin only in GCC 10
+//
+// Note that the builtins has a defined result when overflow occurs, whereas the custom
+// functions below don't update the result if an overflow would occur (to avoid UB).
+#define MP_GCC_HAS_BUILTIN_OVERFLOW (__GNUC__ >= 5)
+
+#if __has_builtin(__builtin_umulll_overflow) || MP_GCC_HAS_BUILTIN_OVERFLOW
+#define mp_mul_ull_overflow __builtin_umulll_overflow
+#else
+inline static bool mp_mul_ull_overflow(unsigned long long int x, unsigned long long int y, unsigned long long int *res) {
+    if (y > 0 && x > (ULLONG_MAX / y)) {
+        return true; // overflow
+    }
+    *res = x * y;
+    return false;
+}
+#endif
+
+#if __has_builtin(__builtin_smulll_overflow) || MP_GCC_HAS_BUILTIN_OVERFLOW
+#define mp_mul_ll_overflow __builtin_smulll_overflow
+#else
+inline static bool mp_mul_ll_overflow(long long int x, long long int y, long long int *res) {
+    bool overflow;
+
+    // Check for multiply overflow; see CERT INT32-C
+    if (x > 0) { // x is positive
+        if (y > 0) { // x and y are positive
+            overflow = (x > (LLONG_MAX / y));
+        } else { // x positive, y nonpositive
+            overflow = (y < (LLONG_MIN / x));
+        } // x positive, y nonpositive
+    } else { // x is nonpositive
+        if (y > 0) { // x is nonpositive, y is positive
+            overflow = (x < (LLONG_MIN / y));
+        } else { // x and y are nonpositive
+            overflow = (x != 0 && y < (LLONG_MAX / x));
+        } // End if x and y are nonpositive
+    } // End if x is nonpositive
+
+    if (!overflow) {
+        *res = x * y;
+    }
+
+    return overflow;
+}
+#endif
+
+#if __has_builtin(__builtin_saddll_overflow) || MP_GCC_HAS_BUILTIN_OVERFLOW
+#define mp_add_ll_overflow __builtin_saddll_overflow
+#else
+inline static bool mp_add_ll_overflow(long long int lhs, long long int rhs, long long int *res) {
+    bool overflow;
+
+    if (rhs > 0) {
+        overflow = (lhs > LLONG_MAX - rhs);
+    } else {
+        overflow = (lhs < LLONG_MIN - rhs);
+    }
+
+    if (!overflow) {
+        *res = lhs + rhs;
+    }
+
+    return overflow;
+}
+#endif
+
+#if __has_builtin(__builtin_ssubll_overflow) || MP_GCC_HAS_BUILTIN_OVERFLOW
+#define mp_sub_ll_overflow __builtin_ssubll_overflow
+#else
+inline static bool mp_sub_ll_overflow(long long int lhs, long long int rhs, long long int *res) {
+    bool overflow;
+
+    if (rhs > 0) {
+        overflow = (lhs < LLONG_MIN + rhs);
+    } else {
+        overflow = (lhs > LLONG_MAX + rhs);
+    }
+
+    if (!overflow) {
+        *res = lhs - rhs;
+    }
+
+    return overflow;
+}
+#endif
+
 #endif // MICROPY_INCLUDED_PY_MISC_H

py/objint_longlong.c

@@ -31,6 +31,7 @@
 #include "py/smallint.h"
 #include "py/objint.h"
 #include "py/runtime.h"
+#include "py/misc.h"
 
 #if MICROPY_PY_BUILTINS_FLOAT
 #include <math.h>
@@ -43,6 +44,10 @@
 const mp_obj_int_t mp_sys_maxsize_obj = {{&mp_type_int}, MP_SSIZE_MAX};
 #endif
 
+static void raise_long_long_overflow(void) {
+    mp_raise_msg(&mp_type_OverflowError, MP_ERROR_TEXT("result overflows long long storage"));
+}
+
 mp_obj_t mp_obj_int_from_bytes_impl(bool big_endian, size_t len, const byte *buf) {
     int delta = 1;
     if (!big_endian) {
@@ -120,7 +125,6 @@ mp_obj_t mp_obj_int_unary_op(mp_unary_op_t op, mp_obj_t o_in) {
         // small int if the value fits without truncation
         case MP_UNARY_OP_HASH:
             return MP_OBJ_NEW_SMALL_INT((mp_int_t)o->val);
-
         case MP_UNARY_OP_POSITIVE:
             return o_in;
         case MP_UNARY_OP_NEGATIVE:
@@ -147,6 +151,8 @@ mp_obj_t mp_obj_int_unary_op(mp_unary_op_t op, mp_obj_t o_in) {
 mp_obj_t mp_obj_int_binary_op(mp_binary_op_t op, mp_obj_t lhs_in, mp_obj_t rhs_in) {
     long long lhs_val;
     long long rhs_val;
+    bool overflow = false;
+    long long result;
 
     if (mp_obj_is_small_int(lhs_in)) {
         lhs_val = MP_OBJ_SMALL_INT_VALUE(lhs_in);
@@ -167,13 +173,16 @@ mp_obj_t mp_obj_int_binary_op(mp_binary_op_t op, mp_obj_t lhs_in, mp_obj_t rhs_i
     switch (op) {
         case MP_BINARY_OP_ADD:
         case MP_BINARY_OP_INPLACE_ADD:
-            return mp_obj_new_int_from_ll(lhs_val + rhs_val);
+            overflow = mp_add_ll_overflow(lhs_val, rhs_val, &result);
+            break;
         case MP_BINARY_OP_SUBTRACT:
         case MP_BINARY_OP_INPLACE_SUBTRACT:
-            return mp_obj_new_int_from_ll(lhs_val - rhs_val);
+            overflow = mp_sub_ll_overflow(lhs_val, rhs_val, &result);
+            break;
         case MP_BINARY_OP_MULTIPLY:
         case MP_BINARY_OP_INPLACE_MULTIPLY:
-            return mp_obj_new_int_from_ll(lhs_val * rhs_val);
+            overflow = mp_mul_ll_overflow(lhs_val, rhs_val, &result);
+            break;
         case MP_BINARY_OP_FLOOR_DIVIDE:
         case MP_BINARY_OP_INPLACE_FLOOR_DIVIDE:
             if (rhs_val == 0) {
@@ -199,9 +208,21 @@ mp_obj_t mp_obj_int_binary_op(mp_binary_op_t op, mp_obj_t lhs_in, mp_obj_t rhs_i
 
         case MP_BINARY_OP_LSHIFT:
         case MP_BINARY_OP_INPLACE_LSHIFT:
-            return mp_obj_new_int_from_ll(lhs_val << (int)rhs_val);
+            if ((int)rhs_val < 0) {
+                // negative shift not allowed
+                mp_raise_ValueError(MP_ERROR_TEXT("negative shift count"));
+            }
+            result = lhs_val << (int)rhs_val;
+            // Left-shifting of negative values is implementation defined in C, but assume compiler
+            // will give us typical 2s complement behaviour unless the value overflows
+            overflow = rhs_val > 0 && ((lhs_val >= 0 && result < lhs_val) || (lhs_val < 0 && result > lhs_val));
+            break;
         case MP_BINARY_OP_RSHIFT:
         case MP_BINARY_OP_INPLACE_RSHIFT:
+            if ((int)rhs_val < 0) {
+                // negative shift not allowed
+                mp_raise_ValueError(MP_ERROR_TEXT("negative shift count"));
+            }
             return mp_obj_new_int_from_ll(lhs_val >> (int)rhs_val);
 
         case MP_BINARY_OP_POWER:
@@ -213,18 +234,18 @@ mp_obj_t mp_obj_int_binary_op(mp_binary_op_t op, mp_obj_t lhs_in, mp_obj_t rhs_i
                 mp_raise_ValueError(MP_ERROR_TEXT("negative power with no float support"));
                 #endif
             }
-            long long ans = 1;
-            while (rhs_val > 0) {
+            result = 1;
+            while (rhs_val > 0 && !overflow) {
                 if (rhs_val & 1) {
-                    ans *= lhs_val;
+                    overflow = mp_mul_ll_overflow(result, lhs_val, &result);
                 }
-                if (rhs_val == 1) {
+                if (rhs_val == 1 || overflow) {
                     break;
                 }
                 rhs_val /= 2;
-                lhs_val *= lhs_val;
+                overflow = mp_mul_ll_overflow(lhs_val, lhs_val, &lhs_val);
             }
-            return mp_obj_new_int_from_ll(ans);
+            break;
         }
 
         case MP_BINARY_OP_LESS:
@@ -242,6 +263,12 @@ mp_obj_t mp_obj_int_binary_op(mp_binary_op_t op, mp_obj_t lhs_in, mp_obj_t rhs_i
             return MP_OBJ_NULL; // op not supported
     }
 
+    if (overflow) {
+        raise_long_long_overflow();
+    }
+
+    return mp_obj_new_int_from_ll(result);
+
 zero_division:
     mp_raise_msg(&mp_type_ZeroDivisionError, MP_ERROR_TEXT("divide by zero"));
 }
@@ -265,22 +292,12 @@ mp_obj_t mp_obj_new_int_from_ll(long long val) {
 }
 
 mp_obj_t mp_obj_new_int_from_ull(unsigned long long val) {
-    // TODO raise an exception if the unsigned long long won't fit
     if (val >> (sizeof(unsigned long long) * 8 - 1) != 0) {
-        mp_raise_msg(&mp_type_OverflowError, MP_ERROR_TEXT("ulonglong too large"));
+        raise_long_long_overflow();
     }
     return mp_obj_new_int_from_ll(val);
 }
 
-mp_obj_t mp_obj_new_int_from_str_len(const char **str, size_t len, bool neg, unsigned int base) {
-    // TODO this does not honor the given length of the string, but it all cases it should anyway be null terminated
-    // TODO check overflow
-    char *endptr;
-    mp_obj_t result = mp_obj_new_int_from_ll(strtoll(*str, &endptr, base));
-    *str = endptr;
-    return result;
-}
-
 mp_int_t mp_obj_int_get_truncated(mp_const_obj_t self_in) {
     if (mp_obj_is_small_int(self_in)) {
         return MP_OBJ_SMALL_INT_VALUE(self_in);

py/parsenum.c

@@ -46,6 +46,27 @@ static MP_NORETURN void raise_exc(mp_obj_t exc, mp_lexer_t *lex) {
     nlr_raise(exc);
 }
 
+#if MICROPY_LONGINT_IMPL != MICROPY_LONGINT_IMPL_LONGLONG
+// For the common small integer parsing case, we parse directly to mp_int_t and
+// check that the value doesn't overflow a smallint (in which case we fail over
+// to bigint parsing if supported)
+typedef mp_int_t parsed_int_t;
+
+#define PARSED_INT_MUL_OVERFLOW mp_small_int_mul_overflow
+#define PARSED_INT_FITS MP_SMALL_INT_FITS
+#else
+// In the special case where bigint support is long long, we save code size by
+// parsing directly to long long and then return either a bigint or smallint
+// from the same result.
+//
+// To avoid pulling in (slow) signed 64-bit math routines we do the initial
+// parsing to an unsigned long long and only convert to signed at the end.
+typedef unsigned long long parsed_int_t;
+
+#define PARSED_INT_MUL_OVERFLOW mp_mul_ull_overflow
+#define PARSED_INT_FITS(I) ((I) <= (unsigned long long)LLONG_MAX)
+#endif
+
 mp_obj_t mp_parse_num_integer(const char *restrict str_, size_t len, int base, mp_lexer_t *lex) {
     const byte *restrict str = (const byte *)str_;
     const byte *restrict top = str + len;
@@ -76,7 +97,7 @@ mp_obj_t mp_parse_num_integer(const char *restrict str_, size_t len, int base, m
     str += mp_parse_num_base((const char *)str, top - str, &base);
 
     // string should be an integer number
-    mp_int_t int_val = 0;
+    parsed_int_t parsed_val = 0;
     const byte *restrict str_val_start = str;
     for (; str < top; str++) {
         // get next digit as a value
@@ -98,25 +119,29 @@ mp_obj_t mp_parse_num_integer(const char *restrict str_, size_t len, int base, m
             break;
         }
 
-        // add next digi and check for overflow
-        if (mp_small_int_mul_overflow(int_val, base)) {
+        // add next digit and check for overflow
+        if (PARSED_INT_MUL_OVERFLOW(parsed_val, base, &parsed_val)) {
             goto overflow;
         }
-        int_val = int_val * base + dig;
-        if (!MP_SMALL_INT_FITS(int_val)) {
+        parsed_val += dig;
+        if (!PARSED_INT_FITS(parsed_val)) {
             goto overflow;
         }
     }
 
-    // negate value if needed
-    if (neg) {
-        int_val = -int_val;
-    }
-
-    // create the small int
-    ret_val = MP_OBJ_NEW_SMALL_INT(int_val);
-
+    #if MICROPY_LONGINT_IMPL != MICROPY_LONGINT_IMPL_LONGLONG
+    // The PARSED_INT_FITS check above ensures parsed_val fits in small int representation
+    ret_val = MP_OBJ_NEW_SMALL_INT(neg ? (-parsed_val) : parsed_val);
 have_ret_val:
+    #else
+    // The PARSED_INT_FITS check above ensures parsed_val won't overflow signed long long
+    long long signed_val = parsed_val;
+    if (neg) {
+        signed_val = -signed_val;
+    }
+    ret_val = mp_obj_new_int_from_ll(signed_val); // Could be large or small int
+    #endif
+
     // check we parsed something
     if (str == str_val_start) {
         goto value_error;
@@ -135,6 +160,7 @@ have_ret_val:
     return ret_val;
 
 overflow:
+    #if MICROPY_LONGINT_IMPL != MICROPY_LONGINT_IMPL_LONGLONG
     // reparse using long int
     {
         const char *s2 = (const char *)str_val_start;
@@ -142,6 +168,9 @@ overflow:
         str = (const byte *)s2;
         goto have_ret_val;
     }
+    #else
+    mp_raise_msg(&mp_type_OverflowError, MP_ERROR_TEXT("result overflows long long storage"));
+    #endif
 
 value_error:
     {

py/runtime.c

@@ -505,13 +505,14 @@ mp_obj_t MICROPY_WRAP_MP_BINARY_OP(mp_binary_op)(mp_binary_op_t op, mp_obj_t lhs
                     }
                     #endif
 
-                    if (mp_small_int_mul_overflow(lhs_val, rhs_val)) {
+                    mp_int_t int_res;
+                    if (mp_small_int_mul_overflow(lhs_val, rhs_val, &int_res)) {
                         // use higher precision
                         lhs = mp_obj_new_int_from_ll(lhs_val);
                         goto generic_binary_op;
                     } else {
                         // use standard precision
-                        return MP_OBJ_NEW_SMALL_INT(lhs_val * rhs_val);
+                        return MP_OBJ_NEW_SMALL_INT(int_res);
                     }
                 }
                 case MP_BINARY_OP_FLOOR_DIVIDE:
@@ -552,19 +553,19 @@ mp_obj_t MICROPY_WRAP_MP_BINARY_OP(mp_binary_op)(mp_binary_op_t op, mp_obj_t lhs
                         mp_int_t ans = 1;
                         while (rhs_val > 0) {
                             if (rhs_val & 1) {
-                                if (mp_small_int_mul_overflow(ans, lhs_val)) {
+                                if (mp_small_int_mul_overflow(ans, lhs_val, &ans)) {
                                     goto power_overflow;
                                 }
-                                ans *= lhs_val;
                             }
                             if (rhs_val == 1) {
                                 break;
                             }
                             rhs_val /= 2;
-                            if (mp_small_int_mul_overflow(lhs_val, lhs_val)) {
+                            mp_int_t int_res;
+                            if (mp_small_int_mul_overflow(lhs_val, lhs_val, &int_res)) {
                                 goto power_overflow;
                             }
-                            lhs_val *= lhs_val;
+                            lhs_val = int_res;
                         }
                         lhs_val = ans;
                     }

py/smallint.c

@@ -26,7 +26,7 @@
 
 #include "py/smallint.h"
 
-bool mp_small_int_mul_overflow(mp_int_t x, mp_int_t y) {
+bool mp_small_int_mul_overflow(mp_int_t x, mp_int_t y, mp_int_t *res) {
     // Check for multiply overflow; see CERT INT32-C
     if (x > 0) { // x is positive
         if (y > 0) { // x and y are positive
@@ -49,6 +49,9 @@ bool mp_small_int_mul_overflow(mp_int_t x, mp_int_t y) {
             }
         } // End if x and y are nonpositive
     } // End if x is nonpositive
+
+    // Result doesn't overflow
+    *res = x * y;
     return false;
 }
 

py/smallint.h

@@ -68,7 +68,10 @@
 // The number of bits in a MP_SMALL_INT including the sign bit.
 #define MP_SMALL_INT_BITS (MP_IMAX_BITS(MP_SMALL_INT_MAX) + 1)
 
-bool mp_small_int_mul_overflow(mp_int_t x, mp_int_t y);
+// Multiply two small ints.
+// If returns false, the correct result is stored in 'res'
+// If returns true, the multiplication would have overflowed. 'res' is unchanged.
+bool mp_small_int_mul_overflow(mp_int_t x, mp_int_t y, mp_int_t *res);
 mp_int_t mp_small_int_modulo(mp_int_t dividend, mp_int_t divisor);
 mp_int_t mp_small_int_floor_divide(mp_int_t num, mp_int_t denom);
 

tests/basics/int_64_basics.py

@@ -0,0 +1,137 @@
+# test support for 64-bit long integers
+# (some ports don't support arbitrary precision but do support these)
+
+# this test is adapted from int_big1.py with numbers kept within 64-bit signed range
+
+# to test arbitrary precision integers
+
+x = 1000000000000000000
+xn = -1000000000000000000
+y = 2000000000000000000
+
+# printing
+print(x)
+print(y)
+print('%#X' % (x - x)) # print prefix
+print('{:#,}'.format(x)) # print with commas
+
+# construction
+print(int(x))
+
+# addition
+print(x + 1)
+print(x + y)
+print(x + xn == 0)
+print(bool(x + xn))
+
+# subtraction
+print(x - 1)
+print(x - y)
+print(y - x)
+print(x - x == 0)
+print(bool(x - x))
+
+# multiplication
+print(x * 2)
+print(1090511627776 * 1048500)
+
+# integer division
+print(x // 2)
+print(y // x)
+
+# bit inversion
+print(~x)
+print(~(-x))
+
+# left shift
+print("left shift positive")
+x = 0x40000000
+for i in range(32):
+    x = x << 1
+    print(x)
+
+# right shift
+print("right shift positive")
+x = 0x2000000000000000  # TODO: why can't second-tip bit be set?
+for i in range(64):
+    x = x >> 1
+    print(x)
+
+# left shift of a negative number
+print("left shift negative")
+for i in range(8):
+    print(-10000000000000000 << i)
+    print(-10000000000000001 << i)
+    print(-10000000000000002 << i)
+    print(-10000000000000003 << i)
+    print(-10000000000000004 << i)
+
+
+# right shift of a negative number
+print("right shift negative")
+for i in range(8):
+    print(-1000000000000000000 >> i)
+    print(-1000000000000000001 >> i)
+    print(-1000000000000000002 >> i)
+    print(-1000000000000000003 >> i)
+    print(-1000000000000000004 >> i)
+
+# conversion from string
+print(int("1234567890123456789"))
+print(int("-1234567890123456789"))
+print(int("1234567890abcdef", 16))
+print(int("1234567890ABCDEF", 16))
+print(int("-1234567890ABCDEF", 16))
+print(int("ijklmnopqrsz", 36))
+
+# numbers close to 64-bit limits
+print(int("-9111222333444555666"))
+print(int("9111222333444555666"))
+
+# numbers with preceding 0s
+print(int("-00000000000000000000009111222333444555666"))
+print(int("0000000000000000000000009111222333444555666"))
+
+# invalid characters in string
+try:
+    print(int("1234567890abcdef"))
+except ValueError:
+    print('ValueError');
+try:
+    print(int("123456789\x01"))
+except ValueError:
+    print('ValueError');
+
+# test parsing ints just on threshold of small to big
+# for 32 bit archs
+x = 1073741823 # small
+x = -1073741823 # small
+x = 1073741824 # big
+x = -1073741824 # big
+# for 64 bit archs
+x = 4611686018427387903 # small
+x = -4611686018427387903 # small
+x = 4611686018427387904 # big
+x = -4611686018427387904 # big
+
+# sys.maxsize is a constant bigint, so test it's compatible with dynamic ones
+import sys
+if hasattr(sys, "maxsize"):
+    print(sys.maxsize - 1 + 1 == sys.maxsize)
+else:
+    print(True)  # No maxsize property in this config
+
+# test extraction of big int value via mp_obj_get_int_maybe
+x = 1 << 62
+print('a' * (x + 4 - x))
+
+# negative shifts are invalid
+try:
+    print((1 << 48) >> -4)
+except ValueError as e:
+    print(e)
+
+try:
+    print((1 << 48) << -6)
+except ValueError as e:
+    print(e)

tests/extmod/json_loads_int_64.py

@@ -0,0 +1,16 @@
+# Parse 64-bit integers from JSON payloads.
+#
+# This also exercises parsing integers from strings
+# where the value may not be null terminated (last line)
+try:
+    import json
+except ImportError:
+    print("SKIP")
+    raise SystemExit
+
+
+print(json.loads("9111222333444555666"))
+print(json.loads("-9111222333444555666"))
+print(json.loads("9111222333444555666"))
+print(json.loads("-9111222333444555666"))
+print(json.loads("[\"9111222333444555666777\",9111222333444555666]"))

tests/extmod/uctypes_addressof.py

@@ -12,5 +12,8 @@ for i in range(8):
     print(uctypes.addressof(uctypes.bytearray_at(1 << i, 8)))
 
 # Test address that is bigger than the greatest small-int but still within the address range.
+try:
     large_addr = maxsize + 1
     print(uctypes.addressof(uctypes.bytearray_at(large_addr, 8)) == large_addr)
+except OverflowError:
+    print(True)  # systems with 64-bit bigints will overflow on the above operation

tests/extmod/uctypes_array_load_store.py

@@ -6,6 +6,13 @@ except ImportError:
     print("SKIP")
     raise SystemExit
 
+# 'int' needs to be able to represent UINT64 for this test
+try:
+    int("FF" * 8, 16)
+except OverflowError:
+    print("SKIP")
+    raise SystemExit
+
 N = 5
 
 for endian in ("NATIVE", "LITTLE_ENDIAN", "BIG_ENDIAN"):

tests/feature_check/int_64.py

@@ -0,0 +1,2 @@
+# Check whether 64-bit long integers are supported
+print(1 << 62)

tests/feature_check/int_64.py.exp

@@ -0,0 +1 @@
+4611686018427387904

tests/perf_bench/bm_pidigits.py

@@ -5,6 +5,12 @@
 # This benchmark stresses big integer arithmetic.
 # Adapted from code on: http://benchmarksgame.alioth.debian.org/
 
+try:
+    int("0x10000000000000000", 16)
+except:
+    print("SKIP")  # No support for >64-bit integers
+    raise SystemExit
+
 
 def compose(a, b):
     aq, ar, as_, at = a

tests/run-tests.py

@@ -628,6 +628,7 @@ def run_tests(pyb, tests, args, result_dir, num_threads=1):
     skip_tests = set()
     skip_native = False
     skip_int_big = False
+    skip_int_64 = False
     skip_bytearray = False
     skip_set_type = False
     skip_slice = False
@@ -658,6 +659,11 @@ def run_tests(pyb, tests, args, result_dir, num_threads=1):
         if output != b"1000000000000000000000000000000000000000000000\n":
             skip_int_big = True
 
+        # Check if 'long long' precision integers are supported, even if arbitrary precision is not
+        output = run_feature_check(pyb, args, "int_64.py")
+        if output != b"4611686018427387904\n":
+            skip_int_64 = True
+
         # Check if bytearray is supported, and skip such tests if it's not
         output = run_feature_check(pyb, args, "bytearray.py")
         if output != b"bytearray\n":
@@ -885,7 +891,12 @@ def run_tests(pyb, tests, args, result_dir, num_threads=1):
         test_name = os.path.splitext(os.path.basename(test_file))[0]
         is_native = test_name.startswith("native_") or test_name.startswith("viper_")
         is_endian = test_name.endswith("_endian")
-        is_int_big = test_name.startswith("int_big") or test_name.endswith("_intbig")
+        is_int_big = (
+            test_name.startswith("int_big")
+            or test_name.endswith("_intbig")
+            or test_name.startswith("ffi_int")  # these tests contain large integer literals
+        )
+        is_int_64 = test_name.startswith("int_64") or test_name.endswith("_int64")
         is_bytearray = test_name.startswith("bytearray") or test_name.endswith("_bytearray")
         is_set_type = test_name.startswith(("set_", "frozenset")) or test_name.endswith("_set")
         is_slice = test_name.find("slice") != -1 or test_name in misc_slice_tests
@@ -899,6 +910,7 @@ def run_tests(pyb, tests, args, result_dir, num_threads=1):
         skip_it |= skip_native and is_native
         skip_it |= skip_endian and is_endian
         skip_it |= skip_int_big and is_int_big
+        skip_it |= skip_int_64 and is_int_64
         skip_it |= skip_bytearray and is_bytearray
         skip_it |= skip_set_type and is_set_type
         skip_it |= skip_slice and is_slice

tools/ci.sh

@@ -695,6 +695,14 @@ function ci_unix_nanbox_run_tests {
     ci_unix_run_tests_full_no_native_helper nanbox PYTHON=python2.7
 }
 
+function ci_unix_longlong_build {
+    ci_unix_build_helper VARIANT=longlong
+}
+
+function ci_unix_longlong_run_tests {
+    ci_unix_run_tests_full_helper longlong
+}
+
 function ci_unix_float_build {
     ci_unix_build_helper VARIANT=standard CFLAGS_EXTRA="-DMICROPY_FLOAT_IMPL=MICROPY_FLOAT_IMPL_FLOAT"
     ci_unix_build_ffi_lib_helper gcc