Jenkins build is still unstable: FreeBSD_HEAD #564

[Julio Merino]

On Mon, Aug 29, 2016 at 2:45 PM, Bryan Drewery <bdrewery at freebsd.org> wrote:

> On 8/29/2016 11:38 AM, Julio Merino wrote:
> > On Mon, Aug 29, 2016 at 2:28 PM, Ngie Cooper (yaneurabeya)
> > <yaneurabeya at gmail.com <mailto:yaneurabeya at gmail.com>> wrote:
> >
> >
> >     > On Aug 29, 2016, at 11:20, Bryan Drewery <bdrewery at FreeBSD.org>
> wrote:
> >
> >
> >     > Trust me, you're wasting your time, this is not the problem.  If
> it were
> >     > then _nothing_ would compile from the /usr/bin/cc compiler.  It's
> only a
> >     > problem with running the ATF tests which means it is likely
> storing the
> >     > flags it is built with somewhere and reusing them.
> >
> >     Then it’s potentially this variable (from lib/atf/libatf-c/Makefile
> ):
> >
> >     CFLAGS+=        -DATF_BUILD_CFLAGS='"${_CFLAGS}”’
> >
> >     I’ll look through the Jenkins logs to confirm the CFLAGS that get
> >     passed in to lib/atf.
> >
> >
> > Yes. If we ought to not use any of those flags at run-time (and we
> > should not, as Bryan pointed out), then yes, the problem is in the
> > setting of ATF_BUILD_CFLAGS et. al.
>
> Are there specific kinds of flags this is wanting to pickup and use
> later? Such as -m or -f or -W?  If we can limit to what is wanted and
> safe it might be better.  We can exclude
> --sysroot/-isystem/-B/-L/-target as well but it isn't as trivial as it
> seems since -isystem has a space after it.
>

This should mostly only pick -I, -L, and -Wl,-R flags to ensure the build
action can find any needed dependencies, and it would be nice if it picked
user-provided flags also.

For atf in the base system, I *think* we could go with an empty set of
flags (unless we have to do something special for private libraries). cc
and c++ should just do the right thing without any flags to compile the
simple things that the failing tests build.

-- 
Julio Merino -- http://julio.meroh.net/