Move to C++1y was more difficult than expected

I’ve managed to get all the build machinery and test runners for PyStan moved to use virtual machines / images which have a version of gcc which has support for C++1y/C++14. This took a surprising amount of work. C++11 is widely supported. I think it might be another year or two before support for C++14 becomes (near) universal. At least in the Python world, most continuous integration stuff is stuck on Ubuntu 14.04 which has gcc 4.8.4. Things have only just started to move to Ubuntu 16.04, which has gcc 5.3.1.

I trust there are no plans to move to C++17 this year?

1 Like

I also had a rough experience.

In using Stan for building tools (in R) for third party users it is important that the software installation is easy, and the overload is not too much compared to alternatives.

At the moment there are many steps to do for installing/running Stan, especially for servers that are conservative about their framework.

I don’t think Rtools is going to move from 4.9.3 for the upcoming R 3.6, in which case even if there were plans to move toward using C++17 features, they would be ruined. But I don’t think there really were such plans.

1 Like