On Fri, 5 Mar 2021, Tobias Burnus wrote:
> Nested functions are permitted for C but not C++ as extension.
> They are also permitted for Fortran, which generates DECL_CONTEXT
> == NAMESPACE_DECL for module variables.
>
> That causes the gcc_assert (decl_function_context (decl) == info->context)
> to fail in tree-nested.c's lookup_field_for_decl.
>
> The call happens in convert_local_reference_stmt for:
> 2524 case GIMPLE_ASSIGN:
> 2525 if (gimple_clobber_p (stmt))
> 2528 if (DECL_P (lhs)
> 2529 && !use_pointer_in_frame (lhs)
> 2530 && lookup_field_for_decl (info, lhs, NO_INSERT))
>
> The latter runs into the assert mentioned above. And the
> '= {CLOBBER}' occurs in gfortran due to the intent(out).
>
> As additional ingredient, a nested (internal) procedure involved,
> obviously as otherwise tree-nested.c wouldn't be involved.
>
> The patch fixes the assert and in terms of the assert it makes
> sense, but I am not sure whether there are assumptions
> elsewhere which are wrong for NAMESPACE_DECL.
I think the bug is elsewhere. We're not expecting non-local
(non-auto) variables to be queried with lookup_field_for_decl.
I think a reasonable fix would be sth like (also see
convert_local_reference_op on how it processes decls).
diff --git a/gcc/tree-nested.c b/gcc/tree-nested.c
index addd6eef9ab..5619ebc85d4 100644
--- a/gcc/tree-nested.c
+++ b/gcc/tree-nested.c
@@ -2526,6 +2526,7 @@ convert_local_reference_stmt (gimple_stmt_iterator
*gsi, bool *handled_ops_p,
{
tree lhs = gimple_assign_lhs (stmt);
if (DECL_P (lhs)
+ && auto_var_p (lhs)
&& !use_pointer_in_frame (lhs)
&& lookup_field_for_decl (info, lhs, NO_INSERT))
{
> OK for mainline? GCC 10?
>
> Tobias
>
> -----------------
> Mentor Graphics (Deutschland) GmbH, Arnulfstrasse 201, 80634 München
> Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung, Frank
> Thürauf
>
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imendörffer; HRB 36809 (AG Nuernberg)
