In D145862#4187915, @jhuber6 wrote:

Do you need someone to land this for you?

No issues with default PIE and SSP on Arch for over 4 years now. It's nice to see this being upstreamed; thanks for doing it. This patch worked fine.

In D102453#2761170, @MaskRay wrote:

For global variables, I have a feature request https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100483

Is it possible that future GCC versions would optimize global variables like Clang does when -fno-semantic-interposition is used?

This proposed change might avoid triggering the reloading issue with the LLVM gold plugin. (At least the simple ar r ab.a a.o b.o from my previous comment doesn't trigger it.)

Yes, someone please commit this for me.

In D50416#1192161, @mgorny wrote:

Also, are you maybe aware of how soon is this going to hit binutils users? It kinda looks like a ticking time bomb, so maybe it would be worth fixing it for LLVM 7 release as well.

In D50416#1192106, @mgorny wrote:

I don't really know BFD plugin API, so I don't know if this could have any implications (I suppose upstream's answer may be helpful there). However, I don't think it likely to cause any real trouble.

Does this make sense or could it cause issues with other modules? I haven't thought of possible implications, but it seems to work OK when tested on top of LLVM 6.0.1.

In D30997#1095549, @tstellar wrote:

I think this option should be dropped. When I added it, I was unaware of D31524, which was being developed at the same time.

Apologies for the random a-year-later comment, but how does this relate to D31524 (which already seems to version libLLVM.so's symbols)?

You have a point that any further tweaks to llvm_config should be created and reviewed separately. What I described above is not too great either; it shouldn't be required to specify USE_SHARED.

+1 from me; it fixes llvm_config(foo USE_SHARED ...) for out-of-tree builds.

One more observation: It might be preferable to avoid defining set(LLVM_DYLIB_COMPONENTS all) in non-DYLIB builds. It prevents llvm_config(foo USE_SHARED ...) from opportunistically linking to the DYLIB while still having the static libraries as a fallback.

Scratch that, I see what's going on! The body of the llvm_config macro needs to check if(ARG_USE_SHARED) instead of if(USE_SHARED).

This change seems to make llvm_config() usable for out-of-tree builds, neat! I tried a standalone build of LLDB and it works fine after having rebuilt llvm with this change.

The latest patch still applies (relatively cleanly) to Clang 3.7.x which is nice. The infinite recursion segfaults appear to be gone, thanks!

In D12834#314136, @elizafox wrote:

In D12834#311890, @foutrelis wrote:

We have received a few reports of clang crashes after applying the abi_tag support patch to our llvm/clang package in Arch Linux.

Why would you put a patch clearly marked as "needs review" into a distribution?!?!?!?!

We have received a few reports of clang crashes after applying the abi_tag support patch to our llvm/clang package in Arch Linux.

Thanks for the review.

Just a note that this change is specific to CMake since autotools already installs clang-query (along with some other tools).