> It is fine to merge decls across static and external flags, but I am not
> sure this is a safe solution to the problem. In C it is perfectly normal to
> have one decl more specified or with different attributes. Like:
>
> extern int a __attribute__ ((warning("bar"));
>
> int a=7 __attribute__ ((warning("foo")));
>
> which must prevent merging otherwise the warnings will go out wrong way. In
> GCC 6 timeframe we decided to live with duplicated declarations and added
> support for syntactic aliases. Tree merging should be optional WRT
> correctness, so I think bug is in the canonical type calculation:
>
> /* For array types hash the domain bounds and the string flag. */
> if (TREE_CODE (type) == ARRAY_TYPE && TYPE_DOMAIN (type))
> {
> hstate.add_int (TYPE_STRING_FLAG (type));
> /* OMP lowering can introduce error_mark_node in place of
> random local decls in types. */
> if (TYPE_MIN_VALUE (TYPE_DOMAIN (type)) != error_mark_node)
> inchash::add_expr (TYPE_MIN_VALUE (TYPE_DOMAIN (type)), hstate);
> if (TYPE_MAX_VALUE (TYPE_DOMAIN (type)) != error_mark_node)
> inchash::add_expr (TYPE_MAX_VALUE (TYPE_DOMAIN (type)), hstate);
> }
>
> which boils down to:
>
> if (tclass == tcc_declaration)
> {
> /* DECL's have a unique ID */
> hstate.add_wide_int (DECL_UID (t));
> }
>
> (in asd_expr)
>
> It is bit ugly, but I think for static/external/public decls we need to use
> assembler name here (as we can't rely on symtab to be around all the time)
> which will result in unstability WRT symbol renaming and also give false
> positives for static symbols. But even for static symbols it is not
> guaranteed they are not duplicated. It may be possible to use combination
> of assembler name and translation unit that is available from symtab node,
> but not all decls will have it defined.
So something akin to what I initially proposed?
https://gcc.gnu.org/ml/gcc-patches/2016-07/msg00182.html
And, as mentioned in my previous message, I'm not sure that we would be able
to merge all the global types with the unification scheme, it is too strict.
--
Eric Botcazou
