weak for Darwin

[Geoff Keating]

On 01/11/2004, at 1:37 PM, Aaron W. LaFramboise wrote:

> Geoff Keating wrote:
>
>> On 01/11/2004, at 11:21 AM, Aaron W. LaFramboise wrote:
>
>>>> 2. Applied to a symbol that isn't defined in the current translation
>>>> unit it makes a weak undefined symbol (i.e. one whose address must 
>>>> be
>>>> tested before the symbol is used, because at runtime it may or may
>>>> not have a definition).
>>
>> How can something be 'not external' and yet not 'defined in the 
>> current
>> translation unit' either?  Is weak on PECOFF just a way of saying 
>> "this
>> symbol is NULL"?
>
> Quick map of PECOFF weak symantics:
>
> If symbol is weak 'w' in a particular object
>
>   If symbol is defined 'D' in any translation unit
>
>     References to that symbol are resolved normally to the
>     defined symbol (weak has no effect in this case).
>
>   If symbol is not defined in any other translation unit
>
>     References within the object that has the weak symbol
>     are resolved to the default value for the weak symbol
>     (which may or may not be 0, depending on whether it
>     has been specified a value or alias locally.)
>
> In particular, 'U' undefined symbols in obj1.o will never be resolved 
> to
> 'w' symbols in obj2.o.
>
> This still preserves the normal usage and semantics we expect of weak
> symbols.  For example, the example from
> <http://gcc.gnu.org/onlinedocs/libstdc++/debug.html#mem> works as 
> expected:
>
>       extern void* __dso_handle __attribute__ ((__weak__));
>       __cxa_atexit((void (*) (void *)) __libc_freeres, NULL,
>                    &__dso_handle ? __dso_handle : NULL);

Aah, I see.  You have both #2 and #1 but with the extra semantic that a 
weak symbol never satisfies any reference.

>> I really do want the semantics of 'weak' as on ELF.  In particular, I
>> want the property that:
>>
>>>> 1. Applied to a symbol that's defined in the current translation
>>>> unit, it makes it a weak definition. (i.e. one that can be 
>>>> overridden
>>>> by a strong definition elsewhere).
>
> This is the part that confuses me.  I thought it was the usual case 
> that
> symbols in libraries such as libsupc++ are always overridable 
> (certainly
> they are on ELF and COFF) by user code.  Is this different with 
> Darwin's
> linker or loader?

Yes, it's different.  It's a very long explanation as to why this 
particular approach seems best.

It's not true that they're *always* overridable; it depends on the link 
ordering.

>> It would be possible to have PECOFF disable the weak-ness using a 
>> macro.
>
> OK.  What do you think about adding a macro named something like
> _GLIBCXX_WEAK_DEFINITION that is __attribute__((weak)) for Darwin,
> nothing for PE targets, and whatever the concensus is for other 
> targets?

Sure.  Will you make a patch, or should I?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 2410 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20041102/8266695a/attachment.p7s>