It turned out that signed integer division and sign extension is harder to get right than I naïvely thought. However, more test cases helped to figure out the corner cases either not covered, or causing UB. If you are already using this library, please update!
A C++20 implementation of safe (wrap around) integers following MISRA C++ rules.
An #ifdef
ed C++17 implementation is available in branch C++17.
It provides the following types in namspace psssint
and corresponding UDL operators in namespace psssint::literals
:
// unsigned
enum class ui8; auto a = 1_ui8;
enum class ui16; auto b = 2_ui16;
enum class ui32; auto c = 3_ui32;
enum class ui64; auto d = 4_ui64;
// signed
enum class si8; auto e = 5_si8;
enum class si16; auto f = 6_si16;
enum class si32; auto g = 7_si32;
enum class si64; auto h = 8_si64;
You can play around with it on compiler explorer starting here
Some extra features:
- an output operator is defined and will output
ui8
andsi8
as integer values. - The UDL operators check for range.
- not recommended features:
to_int(val)
promotes the safe integer val to built-in integer keeping its signednessto_underlying(val)
allows access to the underlying-type's value, use with care in operations because of promotion. This is useful for existing function overloads.from_int(val)
converts one of the allowed integer type's value to corresponding safe integer type. This is useful for integration with not-yet adapted code. Prevents conversion fromchar
andbool
, Note that depending on the definition of the types in<cstdint>
not all built-in integer types will convert, for example, ifstd::int32_t
is defined asint
andstd::int64_t
is defined aslong long
, you cannot convert a42L
, because its conversion rank differs from both.from_int_to<safeinttype>(val)
converts integer to safe integer type with range check (throws if mismatch).
The following MISRA C++ recommendations for integer usage are implemented:
- No mixing of signed and unsigned types or with any standard-defined arithmetic types
- Integral promotion (only when mixing types in operations) will keep signedness
- Operations using the same operand type result in said type (no implicit promotion)
- Operations using mixed types of the same signedness promote to the bigger type first
- All operations wrap safely (2's complement), even for signed operands and small unsigned operands
- Bitwise operations are only defined for unsigned types
- Negation is only defined for signed types
- Comparison is only available through operands of the same type, using built-in comparison (
<=>
in 20)
What else do you want?