Re: More MEM tracking stuff

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

    Huh?  What's the logic here?  This seems very wrong.

See the old code that I deleted in store_field.

    Whatever the logic, this is a *drastic* change in the meaning of alias
    set 0.  Which, among other things, is the "I didn't get around to
    initializing this field" value.

No, that's -1.

The problem is that if a record has alias set 0, we don't want to use
another alias set to reference a field in the record since the temp slot
for that record can be reused for *any* type, including that of the field.
This was a problem reported in some PR and what I did was the suggested
long-term fix.

    Moreover, have you forgotten that new_alias_set forces zero
    values when -fno-strict-aliasing?

No, why does that matter?  In that case, you would use alias set zero
for the field anyway.

    > + hook_get_alias_set_0 (t)

    lang_hook_...

I was going back and forth on that.  My logic for a different prefix was
that I didn't want it to be confused with a default hook, but perhaps you
are right.