Re: Added error message (for too-large array)

[Jason Merrill <jason_merrill at redhat dot com>]

>>>>> "Richard" == Richard Kenner <kenner@vlsi1.ultra.nyu.edu> writes:

> What was the purpose of adding an error message in layout_type when an
> array's size cannot be represented in size_t?

If the size cannot be represented in size_t, it isn't represented
internally in gcc in any useful way; it overflows to some number bearing
little relation to what the user intended.

Now I notice that the C frontend already caught this, in grokdeclarator.  I
suppose the C++ frontend could handle it similarly, but I didn't see how a
type too large to represent could be useful for any language.

> We should give an error if the size of an *object* can't be
> representable, but why a type?

I suppose that moving the test for overflowing TYPE_SIZE to layout_decl
would be plausible.

> Note that you don't check for RECORD_TYPE, just ARRAY_TYPE.

I suppose that would make sense for completeness, though the problem is
much less likely there.

> It's a fairly common idiom in Ada to make an array type whose bounds
> are all integers to represent "all memory".  If it's an array of bytes
> that's OK, but not if it's an array of anything wider.  So this error
> causes problems there.

It would think the Ada frontend should set TYPE_SIZE itself for such an
ARRAY_TYPE, rather than let the middle-end come up with a meaningless
value.

Jason