[arib06]

blst_256_t::to(const unsigned char*, size_t, bool) imports a variable-length byte string into a field element, the C++ mirror of the C blst_scalar_from_be_bytes() entry point. It sizes the leading partial digit with rem = (n - 1) % 32 + 1, but unlike the C version it never special-cases n == 0. When a caller imports an empty value the (n - 1) subtraction underflows to SIZE_MAX, rem lands on 32, and the following n -= rem drives n to SIZE_MAX - 31, so the digit loop keeps reading 32-byte chunks well past the end of the buffer. I ran into it while exercising zero-length imports and AddressSanitizer reported a stack-buffer-overflow read on the first loop iteration. The C siblings in exports.c already guard this with n ? (n - 1) % 32 + 1 : 0, so I applied the same guard here, which makes an empty input decode to the zero element and leaves every n >= 1 path unchanged.

[dot-asm]

Thanks!