Re: [PATCH] c++, v2: Don't defer local statics initialized with constant expressions [PR108702]

[Jason Merrill <jason@redhat.com>]

On 3/3/23 10:18, Jakub Jelinek wrote:
> On Thu, Mar 02, 2023 at 11:48:04AM -0500, Jason Merrill wrote:
>>> The stmtexpr19.C testcase used to be rejected as it has a static
>>> variable in statement expression in constexpr context, but as that
>>> static variable is initialized by constant expression, when P2647R1
>>> was implemented we agreed to make it valid.
>>>
>>> Now, as reported, the testcase compiles fine, but doesn't actually link
>>> because the static variable isn't defined anywhere, and with -flto ICEs
>>> because of this problem.  This is because we never
>>> varpool_node::finalize_decl those vars, the constant expression in which
>>> the DECL_EXPR is present for the static VAR_DECL is folded (constant
>>> evaluated) into just the address of the VAR_DECL.
>>
>> Would it make sense to define it when we see the DECL_EXPR in constant
>> evaluation?
> 
> So like this?

OK, thanks.

> Passes GXX_TESTSUITE_STDS=98,11,14,17,20,2b make check-g++ so far.
> 
> 2023-03-03  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR c++/108702
> 	* constexpr.cc: Include toplev.h.
> 	(cxx_eval_constant_expression) <case DECL_EXPR>: When seeing a local
> 	static initialized by constant expression outside of a constexpr
> 	function which has been deferred by make_rtl_for_nonlocal_decl,
> 	call rest_of_decl_compilation on it.
> 
> 	* g++.dg/ext/stmtexpr19.C: Use dg-do link rather than dg-do compile.
> 
> --- gcc/cp/constexpr.cc.jj	2023-03-03 00:34:44.113679918 +0100
> +++ gcc/cp/constexpr.cc	2023-03-03 13:26:57.602871900 +0100
> @@ -40,6 +40,7 @@ along with GCC; see the file COPYING3.
>   #include "attribs.h"
>   #include "fold-const.h"
>   #include "intl.h"
> +#include "toplev.h"
>   
>   static bool verify_constant (tree, bool, bool *, bool *);
>   #define VERIFY_CONSTANT(X)						\
> @@ -7127,6 +7128,24 @@ cxx_eval_constant_expression (const cons
>   	    break;
>   	  }
>   
> +	/* make_rtl_for_nonlocal_decl could have deferred emission of
> +	   a local static var, but if it appears in a statement expression
> +	   which is constant expression evaluated to e.g. just the address
> +	   of the variable, its DECL_EXPR will never be seen during
> +	   gimple lowering's record_vars_into as the statement expression
> +	   will not be in the IL at all.  */
> +	if (VAR_P (r)
> +	    && TREE_STATIC (r)
> +	    && !DECL_REALLY_EXTERN (r)
> +	    && DECL_FUNCTION_SCOPE_P (r)
> +	    && !var_in_maybe_constexpr_fn (r)
> +	    && decl_constant_var_p (r))
> +	  {
> +	    varpool_node *node = varpool_node::get (r);
> +	    if (node == NULL || !node->definition)
> +	      rest_of_decl_compilation (r, 0, at_eof);
> +	  }
> +
>   	if (AGGREGATE_TYPE_P (TREE_TYPE (r))
>   	    || VECTOR_TYPE_P (TREE_TYPE (r)))
>   	  {
> --- gcc/testsuite/g++.dg/ext/stmtexpr19.C.jj	2023-02-09 15:52:29.623359240 +0100
> +++ gcc/testsuite/g++.dg/ext/stmtexpr19.C	2023-03-03 12:24:20.217186735 +0100
> @@ -1,6 +1,6 @@
>   // PR c++/81073
>   // { dg-options "" }
> -// { dg-do compile { target c++11 } }
> +// { dg-do link { target c++11 } }
>   
>   struct test { const int *addr; };
>   
> 
> 
> 	Jakub
>