newest collect2 patch

[Andrey Slepuhin]

On 28-Jan-98 Jeffrey A Law wrote:
>   In message <XFMail.980127201411.pooh@msu.ru>you write:
>   > this header. The problem is because in the second
>   > case typeinfo nodes should be generated using ".lcomm"
>   > assembler directive instead of ".comm", i.e. they
>   > should not be global.
>   > If this problem will be fixed, we really do not need
>   > any checks for typeinfo nodes in collect2.
> Actually I think these need to be global & exported.
> (assuming that we're moving to an export list for libgcc).

Indeed, typeinfo problem doesn't refer to libgcc. For example, imagine
that I have generic class my_exception; this class is never instantiated
itself. But inherited classes are agressively used both in program
modules and in shared objects. In every module where I use
a class which inherits my_exception, typeinfo node for my_exception
is instantiated. As a result I get duplicated symbols. If I
disable exporting symbols for typeinfo nodes in collect2
(to avoid warnings about duplicated symbols), I get undefined
typeinfo nodes for libstdc++ classes which are declared
using "#pragma interface" directive.
Of course, there are no problems if I don't use shared libraries,
but I do this...

Andrey.