Port, board and/or hardware

any port configured for MICROPY_LONGINT_IMPL_LONGLONG (tested on Unix, Windows and ARM Cortex M4)

MicroPython version

MicroPython v1.25.0-preview.389.gf187c77da.dirty on 2025-03-14; linux [GCC 13.3.0]

Reproduction

The LONGLONG implementation of mp_obj_new_int_from_str_len is broken. It does not take the sign into account, and assumes the string buffer is nul-terminated, which is wrong as mp_parse_num_integer is mostly used on vstr buffers.

To reproduce, make a standard Unix build with

#define MICROPY_LONGINT_IMPL                    MICROPY_LONGINT_IMPL_LONGLONG

1. Try to evaluate

json.loads("-9111222333444555666")

It should return a negative number, but returns a positive number.

2. Try to evaluate

["9111222333444555666777", 9111222333444555666]

It fails with SyntaxError: invalid syntax for integer with base 10: '9111222333444555666'
Removing the first element of array will make the number parse properly.
This issue can be reproduced using json.loads() as well.

Test cases that exhibit the issues have been added to tests/extmod/json_loads.py and submitted as PR #16931

Expected behaviour

1. Should return -9111222333444555666
2. Should return ["9111222333444555666777", 9111222333444555666]

Observed behaviour

1. Returns 9111222333444555666
2. Fails with SyntaxError: invalid syntax for integer with base 10: '9111222333444555666'

Additional Information

Replacing the call to strtoll by code similar to mp_parse_num_integer but using a long long to accumulate the result does fix the issue. It might however be more efficient in terms of code size to use directly a long long in mp_parse_num_integer when compiled with MICROPY_LONGINT_IMPL_LONGLONG

Code of Conduct

Yes, I agree