[Bug target/77587] [5/6/7 Regression] C compiler produces incorrect stack alignment with __attribute__((weak))

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77587

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
From what I can see, this happens because expand_call does:
2736      if (fndecl)
2737        {
2738          struct cgraph_rtl_info *i = cgraph_node::rtl_info (fndecl);
2739          /* Without automatic stack alignment, we can't increase preferred
2740             stack boundary.  With automatic stack alignment, it is
2741             unnecessary since unless we can guarantee that all callers
will
2742             align the outgoing stack properly, callee has to align its
2743             stack anyway.  */
2744          if (i
2745              && i->preferred_incoming_stack_boundary
2746              && i->preferred_incoming_stack_boundary <
preferred_stack_boundary)
2747            preferred_stack_boundary =
i->preferred_incoming_stack_boundary;

and

1951    cgraph_rtl_info *
1952    cgraph_node::rtl_info (tree decl)
1953    {
1954      gcc_assert (TREE_CODE (decl) == FUNCTION_DECL);
1955      cgraph_node *node = get (decl);
1956      if (!node)
1957        return NULL;
1958      node = node->ultimate_alias_target ();
1959      if (node->decl != current_function_decl
1960          && !TREE_ASM_WRITTEN (node->decl))
1961        return NULL;
1962      /* Allocate if it doesnt exist.  */
1963      if (node->ultimate_alias_target ()->rtl == NULL)
1964        node->ultimate_alias_target ()->rtl =
ggc_cleared_alloc<cgraph_rtl_info> ();
1965      return node->ultimate_alias_target ()->rtl;

First of all, it is confusing, because it calls ultimate_alias_target up to 3
times, wouldn't one call be enough?

And the second issue is that ultimate_alias_target and its helpers return foo's
node even when the alias is a weak alias (global) to a local routine, which
could (and is in this case) be overridden with something different at link
time.

Honza, shall ultimate_alias_target ignore such aliases, something else?