Re: More on VIEW_CONVERT_EXPR

["Giovanni Bajo" <giovannibajo at libero dot it>]

Richard Kenner wrote:

>     I'm not sure if there's a way around that.  I was just trying to
>     think of a way to keep this on the front-end's side of the fence.
>
> We did it as an Ada-specific node for the first few years of GNAT's
> history, but it was causing so many problems that we added
> VIEW_CONVERT_EXPR to
> the middle-end a few years ago, specifically, back in November of
> 2001.

Well, this might very well be fixed now with tree-ssa. Do you foresee any
problems switching this back to an Ada-private node?

I think I understand the semantic of VIEW_CONVERT_EXPR, but I don't understand
why it is needed. For instance, in another post, you said:

>You can have a packed record that contains another record at a position that's
not at
> its natural alignment and then do a V_C_E of that field.

Isn't this equivalent in C to a nested structure with some special alignment
attribute on it? And C doesn't use V_C_E. Also C can do unions, which to me
looks like a perfect example of something that *would* need V_C_E but in
practice does not.
-- 
Giovanni Bajo