Update release notes to mention dropping OS X 10.7 support #9022

1. 
   fanquake commented at 10:07 AM on October 26, 2016: member

   Should close #8577.

2. fanquake added the label Docs and Output on Oct 26, 2016
3. fanquake added this to the milestone 0.13.1 on Oct 26, 2016
4. in doc/release-notes.md:None in d89e603e7a outdated

     35 | @@ -36,6 +36,13 @@ No attempt is made to prevent installing or running the software on Windows XP,
     36 |  you can still do so at your own risk, but do not expect it to work: do not
     37 |  report issues about Windows XP to the issue tracker.
     38 |  
     39 | +From 0.13.1 onwards OS X 10.7 is no longer supported. Issues with the libc++
   

   ---

   ---

   

   laanwj commented at 10:10 AM on October 26, 2016:

   Should this be "from 0.13.0 onward"? My impression (from #8577 (comment) ) was that 0.13.0 already was more or less broken on 10.7? @theuni

   ---

   

   theuni commented at 4:53 PM on October 26, 2016:

   @laanwj Yes. I just did a fresh download/run of 0.13.0 on a 10.7 vm to verify. Broken.

   ---

   

   theuni commented at 4:59 PM on October 26, 2016:

   So maybe instead this should read something like: "0.13.0 was intended to work on 10.7+, but severe issues with the libc++ version on 10.7.x keep it from running reliably. 0.13.1 now requires 10.8+, and will communicate that to 10.7 users, rather than crashing unexpectedly".

5. Update release notes for dropping osx 10.7 support 1d1246314f
6. fanquake force-pushed on Oct 27, 2016
7. laanwj merged this on Oct 27, 2016
8. laanwj closed this on Oct 27, 2016

   ---

9. laanwj referenced this in commit 03422e564b on Oct 27, 2016
10. fanquake deleted the branch on Oct 27, 2016
11. 
    jonasschnelli commented at 6:22 AM on October 27, 2016: contributor

    Post merge ACK. Though, 10.8+ is only required for our binary distribution, 10.7 works fine if you self-compile. But we don't need to mention that imo.

12. 
    laanwj commented at 6:48 AM on October 27, 2016: member

    That's quite a subtle detail that may be worth mentioning, yes

13. 
    luke-jr commented at 6:51 AM on October 27, 2016: member

    Though, 10.8+ is only required for our binary distribution, 10.7 works fine if you self-compile.

    Does it? If 10.7's C++11 libraries are broken, how can it?

14. MarcoFalke referenced this in commit ba26d415c0 on Oct 27, 2016
15. 
    droark commented at 1:02 AM on October 28, 2016: contributor

    Though, 10.8+ is only required for our binary distribution, 10.7 works fine if you self-compile.

    Does it? If 10.7's C++11 libraries are broken, how can it?

    We ran into something with Armory that sounds similar. IIRC, there was a C++11 call that was present in libc before 2.17 came out, at which point it was moved to librt. If you compiled in Ubuntu 12.04, the call would picked up in libc. (There were other issues, though, that made compilation on 12.04 impossible unless you used clang.) If you compiled it on 12.10 and beyond, the system expected the call to be in librt, which would cause it to fail due to dynamic linking falling apart. Our solution was to force a static link with a special call such that the relevant code would be included in the binary. This way, 12.04 users could run our binaries without issues. A bit ugly but it worked.

16. 
    jonasschnelli commented at 12:02 PM on October 28, 2016: contributor

    Does it? If 10.7's C++11 libraries are broken, how can it?

    IMO its a ABI difference. We compile against 10.9's c++ libraries which probably must have ABI differences to 10.7 ones. But better ask @theuni to be certain.

17. 
    laanwj commented at 9:51 AM on October 29, 2016: member

    Hacky but makes sense. We do a similar thing for glibc compat configure flag. In the case of ancient OSX it's not worth it though, OSX users seem to have an extremely high update acceptance, no need to support a 5 year old OS (wish we could say the same of windows)

18. 
    theuni commented at 8:16 PM on October 31, 2016: member

    See a detailed explanation here: #8577 (comment)

    Yes, it's an abi difference in the 10.7 libc++ implementation. Unfortunately I wasn't able to narrow down all of the culprits, but I believe there are a few.

19. luke-jr referenced this in commit 1f69022b24 on Apr 21, 2017
20. MarcoFalke locked this on Sep 8, 2021

Contributors

fanquakelaanwjtheunijonasschnelliluke-jrdroark

Milestone
0.13.1