COMPILER_TYPE and installworld

[Warner Losh]

On Thu, Nov 5, 2015 at 9:02 AM, Adrian Chadd <adrian at freebsd.org> wrote:

> On 4 November 2015 at 23:46, NGie Cooper <yaneurabeya at gmail.com> wrote:
> >
> >> On Nov 4, 2015, at 23:08, O'Connor, Daniel <darius at dons.net.au> wrote:
> >>
> >>
> >>> On 5 Nov 2015, at 16:31, Adrian Chadd <adrian at freebsd.org> wrote:
> >>> So! In the short term, is there any way that during
> >>> installworld/installkernel we can also set CC, etc appropriately?
> >>> Otherwise we're either using the wrong compiler logic/version
> >>> (installkernel) or we're just plain not going to work (installworld.)
> >>
> >> Why does installworld/kernel care what the compiler type is?
> >>
> >> Or is it just something incidental which gets pulled in that complains
> because it can't work it out? If so perhaps the easiest way would be to add
> a flag which turns off compiler detection and set it for install*
> >>
> >> If install* _do_ need to know the compiler type then, bleh..
> >
> > It shouldn’t care. bsd.compiler.mk just has the intelligence of a stump
> and Makefile.inc1 doesn’t tell bsd.compiler.mk to do otherwise in the
> distribution/install phases.
> >
> > If it were up to me, CC would be false when make install* is run from
> Makefile.inc1.
>
> I don't mind what people do, I'd just really appreciate some help
> fixing this up. I think we're pretty close to being able to transition
> mips over to using 100% external toolchain only, and this is the
> starting point.
>
> So, what can we do? Pretty please? :)
>

Sadly, CC=true (or CC=false) doesn't solve the problem.
The defaults on the system depend, stupidly[*], on the compiler
you are using and it's capabilities. This means you won't install
all the bits that need C++11 to build, for example.

I'm joining this discussion late. What circumstances do we get CC
wrong in install*? For ages, that's been working properly. What
changed? I suspect 'nothing' and this is nothing more than a
pilot error, but it's possible it's a real bug. Most likely it's the
specific flavor of external toolchain that Bapt added isn't quite
setting variables the way it should.

Warner

[*] I've railed against this in the past to no good effect, mind you.