https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125327
--- Comment #1 from Christopher Bazley <Chris.Bazley at arm dot com> ---
This existing statement appears to be completely wrong and misleading:
tree prev_target_node = build_target_option_node (&global_options,
&global_options_set);
The so-called prev_target_node contains the ambient global option values rather
than the values associated with the function (if any). Because prev_target_node
is not really the previous target node of the function, it is much more likely
to match the rebuilt target node, in which case the real previous target node
is not replaced and the bug remains latent (as in the case of i386).
