Re: Once more, C++11 and older 'base' compilers

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

* On 2017-08-14 at 02:01 BST, coypu%sdf.org@localhost wrote:

> On Sat, Aug 12, 2017 at 04:37:41PM +0100, Jonathan Perkin wrote:
> > * On 2017-08-09 at 18:57 BST, coypu%sdf.org@localhost wrote:
> > 
> > > I'm proposing this option:
> > > per OS/arch combos _TARGET_HAS_PKGSRC_GCC.
> > 
> > I don't really understand this.  Surely every target has a pkgsrc gcc?
> 
> It won't build on many. I will probably use this whitelist:
> 
> linux-*
> sunos-x86 (maybe * not sure about sparc)
> netbsd-x86
> freebsd-x86, dragonflybsd
> 
> For things like Tru64 and such, GCC 4.8 won't work at all as they were
> removed as GCC targets before it.

So it's basically like an early BROKEN or NOT_FOR_PLATFORM check?  I
don't really like adding more ways of doing what we can already do
with existing infrastructure.  It would also be ripe for bitrot, as
we'd now have multiple places where identical information is stored.

We've discussed in the past overhauling the GCC selection process and
how we interact with GCC_REQD.  I'd rather see us continue in that
direction so that we have a consistent and logical selection process
based on mk/platform and mk.conf inputs.  My fear is that this current
proposal would only make things more unclear (what do we do when the
minimum GCC_REQD changes per-platform, for example?)

The first thing we need to do is agree on the direction we should
take, and I'm not sure we've even got there yet.  Once we have a clear
idea about that, the design should follow relatively quickly, and we
can test in bulk builds to ensure we got things right.

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