Re: C++ PATCH: PR 16405

[Mark Mitchell <mark at codesourcery dot com>]

Eric Botcazou wrote:

> > Whatever alignment it has, it has.  If the cast makes the alignment "2",
> > when it's really "4", then the change I made is an optimization because
> > it allows us to do the read more efficiently.  If the cast makes the
> > alignment "4", when it's really "2", then it's a lie, and the program
> > was invalid.
>
>
> Well, in Ada we have "aligning" nodes that explicitly step up the alignment of 
> an expression that is known to be over-aligned, but for which this can't be 
> deduced from the expression itself.  Removing these nodes may lead to the 
> creation of unwanted temporaries later.  See TYPE_ALIGN_OK.

So, you are saying that you might have an expression, of an object type, 
with an alignment of (say) 2, but then apply some casts that give an 
alignment of 4?  Why not just set the alignment correctly on the 
original expression?  And, are these types really compatible, in the 
sense of the langhook?  Your point seems to be that you can't 
interchange the expressions; therefore, it seems, their types should not 
be considered compatible.

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304