On Thu, Apr 14, 2016 at 03:25:37PM +0200, Stefan Richter wrote:
On Apr 14 Jonathan Woithe wrote:
On Thu, Apr 14, 2016 at 09:30:21AM +0200, Stefan
Richter wrote:
gcc (Gentoo 4.9.3 p1.2, pie-0.6.3) 4.9.3"
[...]
Build log:
http://pastebin.com/fRHScVVH
Not sure what changed since the last successful build. gcc was possibly
the same.
In Nikita's report the errors are reported in src/libutil/cmd_serialize.h.
This file hasn't been changed for ages and certainly not any time this year.
There may be something rather obscure going on, but it's hard to see that
any changes have been made to FFADO to cause Nikita's errors. I wonder
there is something strange going on with the compiler or the compilation
environment.
In Stefan's report the error is different:
Unlike Nikita, I haven't added -std=c++11 yet.
Yes, so it seems - my bad. Somehow I managed to convince myself that I'd
seen it in the pastebin log you posted, but I've just rechecked it and it
clearly isn't in there. Evidently I need more sleep.
/usr/include/glibmm-2.4/glibmm/ustring.h:267:12: error: expected â?????;â?????? at
end of member declaration
~ustring() noexcept;
^
This file is included via libxml++'s exception.h header file. I know there
have been some issues with libxml++ and newer gcc versions so I wonder
whether any of those are behind this. I remember that libxml++ did throw me
some curveballs when I moved my development system from gcc 4.5.x to 4.8.x
but I can't remember the details - I'll look them up tonight. It wasn't
anything that affected FFADO though.
On the mainline Gentoo package repo (the only one which I use), there are
now only the packages
dev-cpp/libxmlpp-2.38.1
dev-cpp/libxmlpp-2.40.1
available. I have 2.40.1 installed right now; downgrading to 2.38.1 is
currently impossible since its build fails as described in
https://bugs.gentoo.org/show_bug.cgi?id=567526.
A downgrade of libxmlpp is probably not needed since it seems you have the
same version that resulted in a successful build in the Debian build log
that Adrian linked to.
Note that my build log also contains errors in
/usr/lib/gcc/x86_64-pc-linux-gnu/4.9.3/include/g++-v4/exception for
example.
Yes, further down the log. They are of the same form as the ustring.h
errors though, so whatever's tripping ustring.h up is probably responsible
for the g++-v4/exception woes as well.
The Debian build log with successful build, which
Adrian linked to, lists
the following Debian package versions:
gcc-5_5.2.1-26
libxml++2.6-dev_2.40.1-1
libglibmm-2.4-dev_2.46.2-1
etc.
I guess I attempt an upgrade to gcc-5 next.
It's strange, since I think you mentioned that the same thing happened for
you with a gcc 4.8.x on that system. I guess 4.9.x could break something
which worked in 4.8.x but such behaviour would be more likely with gcc 5.
Still, stranger things have happened, and there's still the issue that the
Debian build worked fine.
On Apr 14 Jano Svitok wrote:
can you add -std=c++11 to CXX_FLAGS or how is it
called to enable C++11
extensions? Most of the errors come from these - noexcept, override,...
I tried multiple things, among them
$ export CXXFLAGS=-std=c++11
$ export CUSTOM_ENV=True
$ scons -c
$ scons
but so far I did not manage to inspire scons to add -std=c++11 to the gcc
command line.
So where did it come from? It must be something the ebuild is adding.
Regards
jonathan