compilers that support thread-local storage?

[Michael Dickens]

Or ... maybe not! Looks like the cxx11 1.1 PG conflicts with the compilers PG & I end up with just g95 for the Fortran compiler selection ... oops!

I'll just replicate the compiler blocklist per your description & go with that. Will work on older OSX & won't harm on newer OSX. - MLD

On Wed, Feb 13, 2019, at 1:41 PM, Michael Dickens wrote:
> OK wow that's quite the analysis! I'm going to go with just 
> incorporating the c++11 1.1 PG ... it'll work for now, and once "we" get 
> around to dealing with the changes to base from PR #88, it'll be a 
> "universal" change to all Portfiles that use c++11 PGs ... so we'd be 
> good to go! Thanks! - MLD
> 
> On Wed, Feb 13, 2019, at 1:23 PM, Ken Cunningham wrote:
> > __thread came first, then thread_local a bit later. 
> > 
> > the difference is that thread_local allows more complicated initializers 
> > and destructors ("non-trivial"). It is c++11, as you said.
> > 
> > 
> > quite old gcc versions support __thread  I think the earliest one was 
> > gcc 4.1 
> > <https://gcc.gnu.org/onlinedocs/gcc-4.1.2/gcc/Thread_002dLocal.html>
> > 
> > gcc 4.8+  supports thread_local 
> > <https://www.gnu.org/software/gcc/gcc-4.8/changes.html>
> > 
> > 
> > For Open Source clang, we have this reference 
> > <https://clang.llvm.org/cxx_status.html> that shows clang-3.3+ supported 
> > it with a proper runtime. That runtime exists on 10.7+. As you know, I 
> > have recently enabled "emulated thread_local" on clang-5.0+ (to match up 
> > with the c++11 PG) for both stdlib=libc++ and stdlib=macports-libstdc++ 
> > (our c++11 runtimes). I wasn't planning on porting that back any 
> > further.
> > 
> > 
> > For Xcode clang, you already have that -- 900+. Apple introduced a 
> > better-performance thread_local system that (as I understand it) 
> > involved integration into dyld .
> > 
> > 
> > 
> > So gcc-4.8+ and clang-5.0+ is reliable on all systems using a c++11 
> > runtime, although there are some systems that can use thread_local with 
> > clang-3.4 to clang-4.0 it appears (untested by me).
> > 
> > My idea was to make the cxx11 PG more or less == thread_local capability. 
> > 
> > This has _all_ been changed in base recently by Marcus' base PR #88, so 
> > it's a moving target.
> > 
> > Ken
> >