[C++ Patch] PR 66644

[Paolo Carlini <paolo dot carlini at oracle dot com>]

Hi,

only when Jakub bumped some bugs in preparation for the release I noted 
that this one remained assigned to me for way too much time...

Roughly speaking, the problem is caused by the fact that when we have a 
GNU anonymous struct inside a union the fields are flattened out and 
appear to be just individual fields of the union. Thus we end up 
rejecting test1 below because multiple fields are initialized and we 
don't simply handle them later on as NSDMIs. A GNU anonymous struct is 
required for the issue to show up, thus the details of the way we want 
to handle such code are debatable, but we know that, eg, both EDG and 
clang accept test1 too (in relaxed GNU mode, at least), besides test2 
and test3.

Given the underlying cause of the rejection I could easily imagine other 
less straightforward ways to match the cases at issues in 
check_field_decl (eg, along the lines DECL_CONTEXT (field) == t ??), but 
the below certainly passes testing on x86_64-linux.

Thanks,
Paolo.

////////////////////////////

Attachment: CL_66644
Description: Text document

Attachment: patch_66644
Description: Text document