ken.cunningham.webuse at gmail.com
Tue Jan 17 20:08:44 UTC 2017
Re: 10.5 Intel - for me, that platform is just a vehicle to 10.5 PPC.
I use it for cross compiling for 10.5 PPC. All these intel machines
can be upgraded to at least 10.7 (10.6 if you want Rosetta). Snow
Leopard DVDs on eBay are $15 if you don't have one (I have six or so
from various Apple Dev and other purchases over the past 20 years). So
there might not be a lot of point to expending a great deal of effort
porting to 10.5 Intel, unless it leads to 10.5 PPC. Am I missing
10.5 PPC is where I'm having some fun trying to get things working.
It's a good challenge. gcc6 works great there for compiling cxx11
against libgcc, but falls down compiling newer things against the
Apple SDKs. I have clang-3.6 / llvm-3.6 working, but the code bus
errors and crashes and seems unreliable.
I have llvm-3.8 built cleanly ( I believe ) on PPC using gcc6 to build
it - gcc6 puts out reliable ppc code, so hopefully that might work
better than some previous versions of llvm - but so far, I can't build
clang-3.8 with anything on ppc. This might be something that could be
overcome. Even then, I don't know if this llvm-3.8 ppc code will be
reliable - probably not very. Is there any way to steer clang-3.6
output into llvm-3.8 for compiling?
the llvm group has stated they are dropping ppc support (there is a
thread about that recently). So I don't know if clang/llvm will ever
exist for 10.5 ppc in a fashion that outputs reliablle reliable code.
Jeremy is in the inner circles on this. Many bugs are indicated a
So it looks like gcc6 (or maybe 7 and beyond) and apple-gcc-4.2 are
probably going to be the end of the road for PPC. Still -- a good run.
Just for fun, I started a build of octave using gcc6 on 10.5 PPC just
now. Will see how far it gets...
Anyway, point of all this is that I'm extremely impressed with the way
the macports devs and Jeremy have enabled libc++ / modern llvm-clang
on 10.6 and 10.7, and would hate to see us move very far away from
that or overly confuse the issue. The current situation is just
fantastic for 10.6/10.7, and uses largely the same portfile structure
as 10.9+, minimizing hassles for port authors.
More information about the macports-dev