Re: GCC 4.2.0 Status Report (2007-03-22)

[kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)]

> > But how is it "supposed to mark the variable as addressable"?  If this
> > just means setting TREE_ADDRESSABLE, what's the point of having the hook?
> 
> It also issues language specific warnings

Then one suggestion is that we rename the langhook to "warn_addressable" and
set TREE_ADDRESSABLE explicitly in the middle-end.  But another is to move
the entire processing into the front end and get rid of the langhook
entirely!  It's the front end that's emitting the ADDR_EXPR in the first
place, so it certainly knows it's taking the address of something!

> It's not supposed to be doing other things for languages that don't
> want to emit warnings (or do other front-endy things with the answer)

And where is this documented?

> You haven't explained what you think needs undoing in *any current
> language that uses the hook*.
> We shouldn't be adding hooks just because some theoretical language
> that doesn't exist might want to do things we don't want it to do
> *anyway*.
> No wonder we have so many langhooks.

I was suggesting *removing* a langhook.

But the fact that "no current frontend does something" is not a reason why
"something" isn't valid!  To the greatest extent possible, we are trying to
form a documented interace between the front end and the middle-end.  To do
that, we need to know precisely what a front-end can and can't do and not
rely on what some particular set of them do now.

Specifically, I suggest that we remove the present mark_addressable langhook.
Have the middle-end set TREE_ADDRESSABLE.  If necessary (and I don't think it
is: see above), call a new langhook that *notifies* the front end about the
addressability for the purpose of a warning, but the front end isn't allowed
to change anything else or to rely on or assume that the address will still
be taken when final code is general.

I don't think the alternative of making a new langhook to unmark addressable
is the best because then you'd actually have to make the marking "private"
and have yet another lang hook to see if something is marked!