This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Improve PR30911 and PR31023, Ada and VRP
On Tue, 1 Apr 2008, Richard Kenner wrote:
> > How would you communicate this to the middle-end then?
>
> My rough proposal is the following: we add a flag to a VAR_DECL that
> means the value in it is invalid. I think we also need a similar flag
> on the other two ways of making a value: INDIRECT_REF and FUNCTION_CALL.
> VRP propagates that flag the same way it propagates values with the fairly
> obvious propagation rules. If that flag is clear for an expression, VRP
> uses the ranges from the subtype, otherwise from the type.
>
> Then the Ada front-end is responsible for the initialization cases I spoke
> about and for setting this flag in the few statically-determinable cases
> where it needs to be set.
This looks equivalent to dropping sub-types (thus by default VRP would
only assume a value lies in the range of the "base-types") and at the
"few statically-determinable cases where it needs to be set" the Ada FE
would place ASSERT_EXPRs that assert the produced value is in the
range of the Ada sub-type. VRP propagates that ranges with the fairly
obvious propagation rules then.
Richard.