Re: chosing a C++ version/compiler (was: icu wants c++11)

[Jonathan Perkin <jperkin%joyent.com@localhost>]

* On 2017-04-24 at 13:50 BST, Joerg Sonnenberger wrote:

> On Mon, Apr 24, 2017 at 02:07:17PM +0200, Edgar Fuß wrote:
> > Is C++n a superset of C++m for n>m?
> 
> Mostly. There are a couple of things that explicitly create problems for
> m == 2003 and there is the problem of at least one semantic change that
> can potentially result in miscompiled code (destructors are tagged as
> not throwing by default in 2011). For m>=2011 it should be transparent.

In that case what about something like this:

  https://github.com/joyent/pkgsrc/commit/e33dc10357d54315a5152ccd9895dde5e1d0f256

tl;dr: introduce CXX_REQD as a similar mechanism to GCC_REQD, packages
append to it what they need, then we distil it to the most recent
version and use that for the build.

todo: set GCC_REQD based on it, i.e. if CXX_REQD>= c++11 then
automatically set GCC_REQD+=4.7 or whatever

This is running through a bulk build at present.

> There is also the ABI breaking mess in libstdc++ to consider of course,
> i.e. depending on how libstdc++ from GCC 5.1+ is installed, code
> compiled with -std=c++11 may or may not link correctly with -std=c++03.

I'm coming to the conclusion that we basically punt on this, and if
users want to ensure that things work well they need to choose
up-front a suitable compiler to use for all their builds that will
satisfy all of their *_REQD.  This is basically what we (Joyent) do
currently, though at some point we'll need to go to newer than GCC 4.9
for the latest C++ stuff.

-- 
Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com