Re: [LTO] Question about comdat linkage of implicit ctor and dtor.

[Rafael Espindola <espindola at google dot com>]

2008/11/20 Doug Kwan (éæå) <dougkwan@google.com>:
> Hi Mark,
>
>    I ran into to a bug where ILs in different files contain identical
> DECLs of an implicit virtual destructor of an instantiated template
> class.  That happened in the BINFOs.  Rafael just removed the BINFOs
> in the IL but the underlying problem is the there.  The problem was
> caused by lack of vague linkage indication in the DECLs so LTO
> complained about multiple definition.  Actually the destructor was not
> defined in either of the object files.  It was not emitted.  So gcc
> did to called import_export_decl to determine the linkage.  As far as
> I know it is an error to re-define an implicitly declared
> constructor/destructor.  So gcc should be able to determine the
> linkage early, when such a constructor/destructor is declared
> implictly.  I tried to look at the logic of import_export_decl and it
> is not simple because of additional interactions of the -frepo and
> -fweak.  Do you know if it is safe to have import_export_decl to
> compute the right linkage setting early for implicit constructors and
> destructors? Thanks.

Unfortunately I am not familiar with the C++ FE so I don't know when
it safe to decide the linkage. It must be possible to do so before
writing to disk, since we (at least implicitly) do so in a normal
compilation when printing assembly :-)

> -Doug
>


Cheers,
-- 
Rafael Avila de Espindola

Google | Gordon House | Barrow Street | Dublin 4 | Ireland
Registered in Dublin, Ireland | Registration Number: 368047