11 May
2016
11 May
'16
6:37 a.m.
doesn't the compile error tells us that this is not the case here?
If you are triggering that warning then the resulting code was broken (that is, not atomic) before.
it is an error not a warning. i'm a bit confused: was it completely broken (as this statement suggest) or was the generated code correct when alignment was ensured by user code (as stl's statement suggested)?
It's not a bincompat break if it never actually worked.
well, isn't it an ABI change if you need to recompile everything with a specific compiler version? users will run into exactly the same issue, if they use structs as my tagged_index, which is not under their control.