[CMake] Android NDK, clang, and ODR violations

[Christopher Broadbent]

On 28/07/2017 02:28, Dan Albert wrote:
> (in case threading doesn't work since I'm not a list member and can't 
> reply to the original message: 
> https://cmake.org/pipermail/cmake/2017-July/065886.html)
> 
> I'm pretty sure -Wl,--detect-odr-violations is just giving you false 
> positives. Note that the warning message says "/possible/ ODR 
> violation". We tried using this flag in the Android platform recently 
> and found lots of false positives too.
> 
> Those symbols are weak in libc++abi, so this is not an ODR violation 
> (they're not weak in libstdc++.so, and that's a bug, but not a harmful 
> one; I'm fixing it right now either way).
> 
> FWIW, CMake shouldn't be linking libstdc++.so /and/ libc++_static.a in 
> the same link either way, but this is actually something that both 
> upstream CMake, the NDK's CMake toolchain file, and ndk-build all get 
> wrong. It's something that we should resolve eventually, but it doesn't 
> actually cause any problems so it hasn't been prioritized. Could argue 
> that it has caused this problem, but this would still be firing on false 
> positives if you'd added your own new/delete replacements, and this 
> warning is still pretty much guaranteed to fire for libc++ because 
> libandroid_support will do exactly the same thing and try to replace a 
> bunch of things in libc.

Thanks, I'll move ahead with the switch, and look in to filtering out 
the warnings on these symbols, rather than disabling the ODR checks 
entirely.