[yocto] what is the proper way to build with fedora rawhide/gcc-5.0?

[Robert P. J. Day]

On Tue, 31 Mar 2015, Khem Raj wrote:

> > On Mar 31, 2015, at 12:54 PM, Robert P. J. Day <rpjday at crashcourse.ca> wrote:
> >
> >  i asked about this a few weeks ago, finally getting back to it
> > ... for better or worse, i'm running fedora rawhide, updated to
> > the point where i have gcc-5.0.0:
> >
> > $ gcc --version
> > gcc (GCC) 5.0.0 20150208 (Red Hat 5.0.0-0.10)
> > Copyright (C) 2015 Free Software Foundation, Inc.
> > This is free software; see the source for copying conditions.  There is NO
> > warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
> > $
> >
> > which causes a number of build issues trying to build for something as
> > simple as qemux86.
> >
> >  first, there is a linemarkers issue with gcc-5.0 that i got around
> > by adding to local.conf:
> >
> >  CPPFLAGS_append= " -P”
>
> This should not be required. Can you post the failing package with error details
> it should be fixed.
> >
> > and there were two native build issues due to gcc-5.0 being far
> > pickier with warnings that i sidestepped with the cheap hack:
> >
> >  ASSUME_PROVIDED += "elfutils-native"
> >  ASSUME_PROVIDED += "binutils-native”
>
> There is 2.25 update available on my contrib tree.
> http://git.openembedded.org/openembedded-core-contrib/log/?h=kraj/master
> should help with binutils.

  i *think* i mentioned earlier that the first failing package build
is ncurses-native:

| In file included from
/home/rpjday/oe/builds/qemux86/tmp/work/x86_64-linux/ncurses-native/5.9-r15.1/ncurses-5.9/ncurses/curses.priv.h:283:0,
|                  from ../ncurses/lib_gen.c:19:
| _9187.c:835:15: error: expected ')' before 'int'
| ../include/curses.h:1594:56: note: in definition of macro
'mouse_trafo'
|  #define mouse_trafo(y,x,to_screen)
wmouse_trafo(stdscr,y,x,to_screen)
|                                                         ^
| gcc  -DHAVE_CONFIG_H -I../ncurses
-I/home/rpjday/oe/builds/qemux86/tmp/work/x86_64-linux/ncurses-native/5.9-r15.1/ncurses-5.9/ncurses
-isystem/home/rpjday/oe/builds/qemux86/tmp/sysroots/x86_64-linux/usr/include
-D_GNU_SOURCE -DNDEBUG -I. -I../include
-I/home/rpjday/oe/builds/qemux86/tmp/work/x86_64-linux/ncurses-native/5.9-r15.1/ncurses-5.9/ncurses/../include
-I/home/rpjday/oe/builds/qemux86/tmp/sysroots/x86_64-linux/usr/include
-isystem/home/rpjday/oe/builds/qemux86/tmp/sysroots/x86_64-linux/usr/include
-D_GNU_SOURCE -O2 -pipe  --param max-inline-insns-single=1200 -fPIC -c
/home/rpjday/oe/builds/qemux86/tmp/work/x86_64-linux/ncurses-native/5.9-r15.1/ncurses-5.9/ncurses/base/lib_isendwin.c
-o ../obj_s/lib_isendwin.o
| Makefile:1682: recipe for target '../obj_s/lib_gen.o' failed

  this appears to be related to the issues listed in this post
describing a mass rebuild of fedora packages with gcc-5.0.0:

https://lists.fedoraproject.org/pipermail/devel/2015-February/207549.html

particularly this:

"Main offender this time is probably the gnu11 change that entails
different inline semantics, enables some warnings by default, bumps
the __STDC_VERSION__, and so on.  Hopefully it won't take too long
till these packages catch on. Many packages were not prepared for the
new major version of GCC.  There's also been quite a lot of churn
because of the preprocessor now emitting linemarkers in the output
when the -P option is not turned on.  The C++ compiler now rejects
some code that it used to accept.  Furthermore, GCC 5 has a batch of
new warnings, which, combined with -Werror, caused some additional
failures."

  sure enough, down that page, ncurses and a bunch of other packages
are identified as having the issue:

"these packages failed to build because of the changes in the
preprocessor; gcc started to generate line directives to better detect
whether a macro tokens come from a system header - see
http://gcc.gnu.org/PR60723 The fix is to use the -P option if the code
isn't prepared to deal with such directives."

  i see that your oe-contrib repo has a newer version of binutils, but
not of ncurses, although your ncurses SRC_URI is different. anyway,
this is just what i've tripped over lately on my rawhide system.

rday

-- 

========================================================================
Robert P. J. Day                                 Ottawa, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================