I'm busy fighting with the python3-base-3.2 build, which seems to be unpredictable atm, due to blender (against the protests of distro
packagers) switching to that version. The 11.4 x86_64 build fails in devel:languages:python:Factory but all the others succeed, I've had
varying build results by linking home:plater:blender python3-base to home:plater:blender python3 and devel:languages:python:Factory
python3, when linked to devel:languages:python:Factory python3 it mimics the devel build but when linked to home:plater:blender python3
I've had complete build failures and all x86_64 failures with i586 successes with a fresh untouched branched python3 after only adding a
fedora conditional for BuildRequires first. I understand that the build service servers aren't 100% in fact the connection was down again a
few moments ago and up again now, I don't know if this has any effect that can cause such build behaviour? It also builds differently on my
local x86_64 11.4 system using rpmbuild.
As it stands now, I've added a few build requirements which seemed to be missing and have experienced an i586 failure due to -pg and
-fomit-frame-pointer being incompatible and redefining $RPM_OPT_FLAGS omitting -fomit-frame-pointer (without diving into gcc) can this
cause problems?
It looks like I've a successful build atm but only after changing a few build options. Here is the diff with comments:
%build
# use rpm_opt_flags
+# Redefine opt flags due to error: -pg and -fomit-frame-pointer are incompatible
+export RPM_OPT_FLAGS="-fmessage-length=0 -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector -funwind-tables -fasynchronous-unwind-tables -g"
export OPT="$RPM_OPT_FLAGS"
#export OPT="-O0 -g"
@@ -201,10 +210,12 @@
--docdir=%{_docdir}/python \
--enable-ipv6 \
--with-fpectl \
- --enable-shared
+ --enable-shared \
+ --with-cxx-main=%{_bindir}/c++ configure started asking for a c++ compiler
LD_LIBRARY_PATH=.:$LD_LIBRARY_PATH \
- make %{?_smp_mflags} profile-opt
+ make %{?_smp_mflags} This is the change that had the most effect on the build
succeeding.
+# profile-opt
%check
# test_file fails in autobuild env - "stdin.seek(-1)" wrongly succeeds. probably an issue with autobuild's stdin
@@ -244,10 +255,10 @@
# cleanup parts that don't belong
for so in _md5 _sha1 _sha256 _sha512; do
- rm $RPM_BUILD_ROOT%{sitedir}/lib-dynload/$so.*
+ rm -f $RPM_BUILD_ROOT%{sitedir}/lib-dynload/$so.* _md5 and _sha1 don't exist, I haven't checked the others but when I see
things missing that existed before I get worried.
done
for dir in curses dbm sqlite3 tkinter; do
- rm -r $RPM_BUILD_ROOT%{sitedir}/$dir
+ rm -rf $RPM_BUILD_ROOT%{sitedir}/$dir
done
# install site-specific tweaks
@@ -277,7 +288,7 @@
# move idle config into /etc
install -d -m 755 ${RPM_BUILD_ROOT}/etc/idle3
-(
+(
cd ${RPM_BUILD_ROOT}/%{sitedir}/idlelib/
for file in *.def ; do
mv $file ${RPM_BUILD_ROOT}/etc/idle3/
I use kwrite for spec files and it saves in UTF8 also removes trailing white spaces.
I would appreciate expert help with this after which python3-3.2 could hopefully be ready for tumbleweed, blender is about to be 2.57 RC1
and will be released in about a month, the current blender in 11.4 is beta plus svn and would have been newer if the switch to python 3.2
hadn't happened. I think blender is the main consumer of python3 along with a couple of gnome apps.
Thanks
Dave P
--
To unsubscribe, e-mail: opensuse-packaging+unsubscribe(a)opensuse.org
For additional commands, e-mail: opensuse-packaging+help(a)opensuse.org