[Darwin] Patch, take 2: enable linkonce support for Darwin

[Matt Austern]

On Feb 2, 2004, at 10:47 AM, Jason Merrill wrote:

> On Fri, 30 Jan 2004 17:06:17 -0800, Matt Austern <austern@apple.com> 
> wrote:
>
>> Here's the case where we use COMMON even on SUPPORTS_ONE_ONLY
>> targets:
>>   if (TREE_CODE (decl) == VAR_DECL
>>        && (DECL_INITIAL (decl) == 0 || DECL_INITIAL (decl) ==
>> error_mark_node))
>>      DECL_COMMON (decl) = 1;
>>    else if (SUPPORTS_ONE_ONLY)
>>      ...
>>
>> I agree, it looks funny.  Unless someone can explain to me why this 
>> is a
>> good idea, I'd be happy to exchange the order of those clauses so 
>> that we
>> never use COMMON if we've got ONE_ONLY available.
>
> I was doing it that way out of a philosophy of using a more advanced
> feature only when necessary.  COMMON objects are merged by any Unix 
> linker,
> and they take up no space in the object file.  But if using COMMON is
> problematic, I'm certainly not attached to it.

OK!  As you may have noticed, version 3 of my patch avoids COMMON
on all platforms where ONE_ONLY is available.  This is the only part of
my patch that changes the compiler's behavior for any target other than
Darwin.

>
>>>> Finally, one new user-visible feature: -fhidden-weak.  This flag 
>>>> makes
>>>> the compiler automatically give hidden visibility to symbols that 
>>>> have
>>>> vague linkage.
>
> Sounds like this would block sharing of template instantiations and 
> RTTI
> information between an executable and a shared library, which would 
> break a
> lot of code.

Not necessarily as much code as you might think!  Most Unix linkers have
the property that implicit instantiations are shared between different 
shared
libraries, but that's not the way it works on MS-Windows.  Anyone who's
writing code that's portable between Unix and Windows isn't relying on 
that
sort of sharing.

But yes, you're  basically right: this is a fairly big change in 
semantics.
That's why it's not the default, even on Darwin.

			--Matt