How to properly use C++ union on embedded software

Question

I'm an embedded software (firmware) developer (C++) and sometimes I need to convert data between types, for example

• an array of 4 bytes, need to be written to the hardware in a single 32bit register
• a float variable needs to be transmitted in a message as bytes

In order to do the conversion properly, I am using a union, to store the data, so I can read/write it in both ways, for example

union {
    uint8_t asByte[lengthInBytes]{};         // interprete the data as 16 bytes
    uint32_t asUint32[lengthInWords];        // interprete the data as 4 32bit words
} state;

Now I am running static code checking on my code (using SonarCloud) and this is flagging an issue on this construct :

Add a discriminant to this wrapped undiscriminated union or replace it with an "std::variant"

I don't want to use the std:: library containers because I need to avoid using dynamic memory allocation, and often it's also an overkill...

So how can I resolve this 'codeSmell', or should I just ignore/accept it in my use case ?

Answer 1

Thanks for all the useful remarks,

Seems like std::bitcast is the right way to do this.
It requires C++20 but I think I'm ok with that.