David Laight: david@l8s.co.uk wrote:

        And that doesn't DTRT either, since the compiler then has to generate
        code to allow for the structure itself being misaligned - so on systems
        which cannot do misaligned transfers (eg ARM) the compiler has to read
        each byte of a 32bit value separately and then shift and 'or' them together.
        
        I wonder what the guys at ARM were on when they decided on that structure
        padding rule! (I'm presuming that gcc is following some ARM ABI).

It was Acorn that originally invented that rule, so a longggg time ago. 

ARM on NetBSD has never used that structure size rounding; and the ARM
EABI doesn't either.

R.