[patch] Don't use appending linkage for embeded bitcode

[Rafael Espíndola via cfe-commits]

You can probably use collectUsedGlobalVariables.

You can probably also delete the FIXME about llvm.used and appending
linkage. I think the way to fix this is to make "can be dropped" an
independent property from the linkage and then we don't need llvm.used
at all.

Cheers,
Rafael


On 13 May 2016 at 17:01, Steven Wu <stevenwu at apple.com> wrote:
> Attach a patch using private linkage type and adding to llvm.used. I have to recreate llvm.used when embedding bitcode. I don't really like it but I don't have better solutions.
> Few other options:
> 1. Not allowing re-embedded bitcode will simplify the code a bit but not a lot.
> 2. Create a new "llvm.*.used" variable in llvm.metadata section.
> 3. Teach optimizer do not optimize away llvm.* variables.
>
>
>
>
> Steven
>
>> On May 13, 2016, at 10:10 AM, Rafael Espíndola <rafael.espindola at gmail.com> wrote:
>>
>> On 13 May 2016 at 13:02, Steven Wu <stevenwu at apple.com> wrote:
>>> Hi Rafael
>>>
>>> Thanks for notice this! That would definitely cause duplicated symbol error and I should definitely change that.
>>> Here is some background:
>>> ld64 in Xcode 7+ knows how to handle the embedded bitcode correctly but not the ones in earlier Xcode. The old ld64 will simply concatenate the bitcode files which is not the right thing to do. So there is a symbol generated at the place to prevent user to link the bitcode object file with old ld64 because older ld64 will fail and report duplicated symbols.
>>> I have a radar tracking to change the linkage type when upstream but I dropped the ball on that one. The correct thing to do is to make it internal and add to llvm.used. I will come up with a patch.
>>
>> Thank you so much!
>>
>> Cheers,
>> Rafael
>
>