Shipping custom libc++ on MacOS

Hello,

We are working on an application where we want to have full support
for C++17 but still ship on older versions of macOS (10.9/10.10 in
this case). Our plan was to build our own libc++ from llvm and ship it
with the application and then pass "-D_LIBCPP_DISABLE_AVAILABILITY"

This seemed to work fine in internal testing - but then we ran into a
similar issue as the one described here:
https://reviews.llvm.org/D74489#2497044

Now I can patch my copy of libc++ with the upstream fix for this - but
the comment below made me think otherwise:
https://reviews.llvm.org/D74489#2505156

We have run into problem where if you ship your own libc++ the system
frameworks loads the system libc++ and some strange errors happen - we
thought we could work around these issues - but maybe this is just us
being naive.

So I guess my question is - is it possible to ship a custom libc++ in
a .app archive for older versions of macOS in order to use C++17? Or
is my only hope to raise the minimum version of our app and always
link to the system libc++?

Thanks,
Tobias

Hi,

I can't speak for macOS specifically, but in general you can have problems on *NIX platforms if you have two libraries that implement the same symbols loaded. Everything in libc++ is in the namespace defined by _LIBCPP_ABI_NAMESPACE and then aliased into std. Can you not define this to something custom? That should give you different symbol names and so if you depend on another library that links a different version of libc++ (statically or dynamically) you won't conflict.

David

Generally, I would say this is a tricky thing to do. As you mention, the problem is that if you link against any other library that does link against the system-provided libc++.dylib, you'll have ODR violations because some symbols will be found in both libraries. As a result, I believe you could also end up in a situation where you end up using globals from one library when the globals from the other library has been initialized. So basically, nothing good happens.

If you can ensure that your application doesn't pull in the system libc++.dylib (even transitively), then this would in theory be safe to do. Due to the brittleness of the solution, my immediate reaction would be to not recommend that. The reason why some parts of C++17 aren't available on older platforms is that we put the vtables required for exception classes in the dylib. That is a code size optimization that we think is worth the trouble. You can generally use the parts of C++17 APIs that can't throw exceptions even on older platforms because they don't require those vtable symbols - perhaps that is something you can consider?

Louis

Also, Chrome is known to do something similar to what you’re inquiring about. They build libc++ statically with hidden visibility, and then they link it into their application. I assume they also make sure to not export any symbol from libc++ that would be created by using the libc++ headers (e.g. by instantiating some function template). I don’t know their setup very well, but I assume they make sure to not link against any other system library that does require libc++.dylib from the system, or else things get trickier as I explained above. But as long as you satisfy that requirements, I believe things have worked fairly well for them so far. You just have to be able and willing to walk that fine line.

Louis

On MacPorts we are just getting into this issue, in our case trying to build a new libc++.dylib that will be accepted by the existing system frameworks on 10.7 to 10.13 to support c++17 software on older systems without resorting to libstdc++ from gcc.

Is this macro of value in helping avoid ODR violations? I’m just investigating that now:

_LIBCPP_HIDE_FROM_ABI_PER_TU

Ken

We use this approach in Fuchsia as well where we always statically link libc++ into our host tools and we haven’t encountered any issues, but these are command line development tools which generally don’t use any frameworks or host libraries. The only minor problem we ran into is that Clang’s Darwin driver doesn’t support -static-libstdc++ flag, so we instead have to pass -nostdlib++ ${clang_prefix}/../lib/libc++.a through linker flags which is a bit unfortunate but not a big deal.

On MacPorts we are just getting into this issue, in our case trying to build a new libc++.dylib that will be accepted by the existing system frameworks on 10.7 to 10.13 to support c++17 software on older systems without resorting to libstdc++ from gcc.

Is this macro of value in helping avoid ODR violations? I’m just investigating that now:

_LIBCPP_HIDE_FROM_ABI_PER_TU

Pasting from the documentation:

_LIBCPP_HIDE_FROM_ABI_PER_TU
This macro controls whether symbols hidden from the ABI with _LIBCPP_HIDE_FROM_ABI
are local to each translation unit in addition to being local to each final
linked image. This macro is defined to either 0 or 1. When it is defined to
1, translation units compiled with different versions of libc++ can be linked
together, since all non ABI-facing functions are local to each translation unit.
This allows static archives built with different versions of libc++ to be linked
together. This also means that functions marked with _LIBCPP_HIDE_FROM_ABI
are not guaranteed to have the same address across translation unit boundaries.

When the macro is defined to 0, there is no guarantee that translation units
compiled with different versions of libc++ can interoperate. However, this
leads to code size improvements, since non ABI-facing functions can be
deduplicated across translation unit boundaries.

This macro can be defined by users to control the behavior they want from
libc++. The default value of this macro (0 or 1) is controlled by whether
_LIBCPP_HIDE_FROM_ABI_PER_TU_BY_DEFAULT is defined, which is intended to
be used by vendors only (see below).

Basically, this macro allows mixing different TUs or static archives built with different versions of libc++. That is relevant if you have multiple versions of libc++ within the same executable, which doesn’t appear to be your case here.

Louis