Make casting negative numbers more predictable. about fixedpointsarduino HOT 2 OPEN

To cut a long story short, it's because internally the conversion is using unsigned shift operations (i.e. 'logical' shifting is being forced rather than 'arithmetic' shifting).

I can't remember whether this was intentional or not, but it quite possibly was because technically signed right shift is 'implementation defined' prior to C++20, whilst unsigned shifting is well defined.

The fix would be fairly simple, it would just be a matter of removing a few casts and consequently some unneeded local type aliases.

However, even with that the result won't be quite perfect because of the inherant loss of precision involved.

from fixedpointsarduino.

A reminder for myself:
https://gist.github.com/Pharap/04e6fff6fe754641048faabe3ecb4af5

from fixedpointsarduino.