PolyBoRi fails to build on OpenSolaris with gcc-4.6.0

[drkirkby]

When compiling with gcc, PolyBoRi generates position independent code because of the compiler flag -fPIC. It does this on Solaris, and I expect Linux and OS X too.

However, on Solaris PolyBoRi also generates the compiler flag "-KPIC" which is the flag used by Sun (now Oracle) compilers to generate position independent code. GCC has always ignored this flag. However, the latest gcc does not, and considers this an error.

g++ -o Cudd/obj/cuddObj.pic.o -c -O3 -Wno-long-long -Wreturn-type -g -fPIC -ftemplate-depth-100 -O3 -Wno-long-long -Wreturn-type -g -fPIC -KPIC -DNDEBUG -DHAVE_GD -DHAVE_TR1_UNORDERED_MAP -DPACKED -DHAVE_M4RI -DHAVE_GD -DHAVE_IEEE_754 -DBSD -I/export/home/drkirkby/sage-new-gcc/sage-4.7.alpha2/local/include -I/export/home/drkirkby/sage-new-gcc/sage-4.7.alpha2/local/include/python2.6 -Ipolybori/include -ICudd/obj -ICudd/util -ICudd/cudd -ICudd/mtr -ICudd/st -ICudd/epd Cudd/obj/cuddObj.cc
g++: error: unrecognized option '-KPIC'
scons: *** [Cudd/obj/cuddObj.pic.o] Error 1
scons: building terminated because of errors.
Error building PolyBoRi.

I found this on OpenSolaris, but I'm 99% sure the same would happen on Solaris, since I know the option "-KPIC" has always been there on both systems.

I'm cc'ing the PolyBoRi developers.

Summary

• SPKG:  http://boxen.math.washington.edu/home/jdemeyer/spkg/polybori-0.7.0.p2.spkg

[drkirkby]

I've sent a note to polybori-discuss@lists.sourceforge.net outlining the problem a few minutes ago.

[AlexanderDreyer]

Thanks for pointing this out. Since the gcc-installation of t2 seems to be broken. Is there any machine around, where I can get access to?

[AlexanderDreyer]

Meanwhile. please try out this hack:

diff -r 7bb357180e7b SConstruct
--- a/SConstruct        Sun Mar 13 22:34:17 2011 +0100
+++ b/SConstruct        Wed Mar 30 10:05:01 2011 +0200
@@ -276,6 +276,9 @@


 env = Environment(ENV = getenv, options = opts, tools = tools, toolpath = '.')
+if (env['PLATFORM'] == "sunos") and ('gcc' in env['TOOLS']):
+    if '-KPIC' in  env['SHCCFLAGS']:
+      env['SHCCFLAGS'].remove('-KPIC')

 if 'dump' in COMMAND_LINE_TARGETS:
   print env.Dump()

If the problem persists, please add replace scons by scons dump and post the output.

My best,

Alexander

[drkirkby]

Replying to AlexanderDreyer:

Thanks for pointing this out. Since the gcc-installation of t2 seems to be broken. Is there any machine around, where I can get access to?

gcc on t2.math works fine. William did break it, but I fixed ages ago. t2.math is regularly used to build Sage with the buildbot. It was last built successufully on 20^th March - only 10 days ago.

 http://build.sagemath.org/sage/buildslaves/t2-1

I just put a note in /etc/motd to add this to ones $HOME/.profile

. /usr/local/bin/t2-setup

However, there is currently no gcc-4.6.0 on there, but it should be possible to see if the patch removes the -KPIC option.

Do you want an account on my Sun Ultra 27? That runs OpenSolaris, not Solaris and is x86 not SPARC.

Although there are some differences, most things are the same. The advantage of that machine is that it is much faster than t2.math. You can build Sage and run all the long doctests in little over an hour if the machine is idle.

[AlexanderDreyer]

I just put a note in /etc/motd to add this to ones $HOME/.profile

. /usr/local/bin/t2-setup

Thanks, I didn't know that. Is there a 4.7alpha2 build somewhere?

However, there is currently no gcc-4.6.0 on there, but it should be possible to see if the patch removes the -KPIC option.

Do you want an account on my Sun Ultra 27? That runs OpenSolaris, not Solaris and is x86 not SPARC.

This would be nice, I'll contact you off-Trac

My best

Alexander

[AlexanderDreyer]

It seems, I finally found out where this -KPIC comes in and managed to fix it. See the updated spkg here:  http://sage.math.washington.edu/home/dreyer/spkg/polybori-0.7.0.p2.spkg

My best,

Alexander

[drkirkby]

Thank you. That has solved it.

[jdemeyer]

SPKG.txt needs to updated:

• backport  http://trac.sagemath.org/sage_trac/raw-attachment/ticket/10797/polybori-SPKG.txt.diff
• mention this ticket

[AlexanderDreyer]

Does this mean that it fails again?

[AlexanderDreyer]

Or should does one needs to rebundle the spkg?

[jdemeyer]

Only the file SPKG.txt in the spkg should be changed. This ticket should be documented in the spkg, that's it.

[AlexanderDreyer]

Ok, I put it on the same location here;

 http://sage.math.washington.edu/home/dreyer/spkg/polybori-0.7.0.p2.spkg

(the previous one is at  http://sage.math.washington.edu/home/dreyer/spkg/polybori-0.7.0.p2.spkg.0)

[jdemeyer]

I removed SPKG.txt~, new spkg:  http://boxen.math.washington.edu/home/jdemeyer/spkg/polybori-0.7.0.p2.spkg

[drkirkby]

That looks fine

real	3m44.578s
user	3m21.196s
sys	0m16.574s
Successfully installed polybori-0.7.0.p2

Positive review.