On Wed, Aug 7, 2019, 2:35 PM Marek Polacek <pola...@redhat.com> wrote:

> On Wed, Aug 07, 2019 at 12:23:25PM -0400, Jason Merrill wrote:
> > On Wed, Aug 7, 2019, 9:11 AM Marek Polacek <pola...@redhat.com> wrote:
> >
> > > On Tue, Aug 06, 2019 at 09:25:45PM -0400, Jason Merrill wrote:
> > > > Let's downgrade the errors in earlier standard modes to pedwarn. Ok
> with
> > > > that change.
> > >
> > > Works for me, here's what I'll apply once it passes testing.
> > >
> > > I removed the diagnostic in potential_constant_expression_1/ASM_EXPR so
> > > that
> > > we don't generate duplicate pedwarns for the same thing.  Hope that's
> OK.
> > >
> > > 2019-08-07  Marek Polacek  <pola...@redhat.com>
> > >
> > >         PR c++/91346 - Implement P1668R1, allow unevaluated asm in
> > > constexpr.
> > >         * constexpr.c (cxx_eval_constant_expression): Handle ASM_EXPR.
> > >         (potential_constant_expression_1) <case ASM_EXPR>: Allow.
> > >         * cp-tree.h (finish_asm_stmt): Adjust.
> > >         * parser.c (cp_parser_asm_definition): Grab the locaion of
> "asm"
> > > and
> > >         use it.  Change an error to a pedwarn.  Allow asm in C++2a,
> warn
> > >         otherwise.
> > >         * pt.c (tsubst_expr): Pass a location down to finish_asm_stmt.
> > >         * semantics.c (finish_asm_stmt): New location_t parameter.
> Use it.
> > >
> > >         * g++.dg/cpp2a/inline-asm1.C: New test.
> > >         * g++.dg/cpp2a/inline-asm2.C: New test.
> > >         * g++.dg/cpp1y/constexpr-neg1.C: Adjust dg-error.
> > >
> > > diff --git gcc/cp/constexpr.c gcc/cp/constexpr.c
> > > index 36a66337433..e86b0789b84 100644
> > > --- gcc/cp/constexpr.c
> > > +++ gcc/cp/constexpr.c
> > > @@ -5289,6 +5289,18 @@ cxx_eval_constant_expression (const
> constexpr_ctx
> > > *ctx, tree t,
> > >        r = void_node;
> > >        break;
> > >
> > > +    case ASM_EXPR:
> > > +      if (!ctx->quiet)
> > > +       {
> > > +         error_at (cp_expr_loc_or_input_loc (t),
> > > +                   "inline assembly is not a constant expression");
> > > +         inform (cp_expr_loc_or_input_loc (t),
> > > +                 "only unevaluated inline assembly is allowed in a "
> > > +                 "%<constexpr%> function in C++2a");
> > > +       }
> > > +      *non_constant_p = true;
> > > +      return t;
> > > +
> > >      default:
> > >        if (STATEMENT_CODE_P (TREE_CODE (t)))
> > >         {
> > > @@ -6469,13 +6481,18 @@ potential_constant_expression_1 (tree t, bool
> > > want_rval, bool strict, bool now,
> > >        /* GCC internal stuff.  */
> > >      case VA_ARG_EXPR:
> > >      case TRANSACTION_EXPR:
> > > -    case ASM_EXPR:
> > >      case AT_ENCODE_EXPR:
> > >      fail:
> > >        if (flags & tf_error)
> > >         error_at (loc, "expression %qE is not a constant expression",
> t);
> > >        return false;
> > >
> > > +    case ASM_EXPR:
> > > +      /* In C++2a, unevaluated inline assembly is permitted in
> constexpr
> > > +        functions.  If it's used in earlier standard modes, we
> pedwarn in
> > > +        cp_parser_asm_definition.  */
> > > +      return true;
> > >
> >
> > Actually, do we need this change? If it's (possibly) unevaluated, we
> > shouldn't get here.
>
> We can get here when using asm() in ({ }) like this (ugh):
>
> constexpr int
> foo (bool b)
> {
>   if (b)
>    {
>      constexpr int i = ({ asm(""); 42; });
>      return i;
>     }
>   else
>     return 42;
> }
> static_assert(foo(false) == 42, "");
>
> With the current state of potential_constant_expression_1, we generate
>
> inline-asm3.C: In function ‘constexpr int foo(bool)’:
> inline-asm3.C:10:27: error: inline assembly is not a constant expression
>    10 |      constexpr int i = ({ asm(""); 42; });
>       |                           ^~~
> inline-asm3.C:10:27: note: only unevaluated inline assembly is allowed in
> a ‘constexpr’ function in C++2a
>
> which I thought was better than what we emit with the hunk revered:
>
> inline-asm3.C: In function ‘constexpr int foo(bool)’:
> inline-asm3.C:10:27: error: expression ‘<statement>’ is not a constant
> expression
>    10 |      constexpr int i = ({ asm(""); 42; });
>       |                           ^~~
>
> But I'm happy to revert that hunk if you want.
>

Sure, perhaps we could share the error message between the two functions.
In general it's better to reject something in potential_... if we can.

>

Reply via email to