[MacPorts] #54766: Does port:clang*'s libstdcxx have to be default?

René J.V. Bertin rjvbertin at gmail.com
Sat Sep 9 08:15:27 UTC 2017


On Friday September 08 2017 21:06:35 Jeremy Huddleston Sequoia wrote:

> The reason for this is that many OSS projects based on autoconf, cmake, etc have an assumption baked into the build system that the SDK matches the minimum level of support.  

Well, I think that *is* the easier/safer assumption to make. Esp. on OS versions that are somehow on some kind of bifurcation. I don't really know what to call it, but the 10.10 SDK on OS X 10.9 would be an example. From what I understand it remained stuck at 10.10.1 or so and never received newer APIs that were added in 10.10.2 or later. That makes it complicated enough to do API checks that for instance the Qt 5.9LTS release dropped official support for 10.9.

>> Well, Qt does `xcrun -sdk $QMAKE_MAC_SDK -find $QMAKE_CXX`
>
>That scares me, but hopefully they know what they're doing with taht.

Their contraption builds reliably and building with anything other than the "official" toolchain is not their concern. That's all I can say...
>
>> and as far as I've seen that means you always get the clang from Xcode no matter what QMAKE_CXX is set to.
>
>Nope.
>
>$ xcrun -sdk macosx -find clang-mp-devel
>/opt/local/bin/clang-mp-devel

Hmm, I may need to check again, it's been a while since I figured out I needed to remove the compilers and linker from that lookup step. And truth be told, I rarely use a clang-mp version if there's no significant benefit to it. Every newer clang version is slower than the previous and as you know Apple's clang has always been faster whenever I compared it.
(I hope the llvm 4.0 -> 5.0 jump is explained by enough internal changes to reverse the slowness trend but am not holding my breath.)

R.


More information about the macports-dev mailing list