g++ and aliasing bools

[Joe Buck]

me:
> > Good catch.  OK, Daniel, Mark has demonstrated that he was correct in
> > asking you for a proof: your proposal was wrong.

Daniel:
> Not quite, Mark has caught a case that is already handled properly.
> It is impossible for the type involved as a member to have been given an 
> alias set other than 0, by the same test.  So we would never improperly 
> generate code because of it.

My analysis was based on the pseudocode, where this this was not obvious.

But if this is the case, there is no bug, so it seems that Mark's pre-approval
would apply to your patch, and it can go in (for 3.2).