pari compilation failure on Ubuntu Karmic

[Dan Drake]

I installed an Ubuntu Karmic virtual machine to see how Sage compiles
there...and right now, it doesn't.

I've tried on a 64-bit and 32-bit VM, with between 768 and about 1250 MB
of RAM allocated to the VM, and it consistently fails while compiling
base3.c from pari (src/basemath/base3.c).

When it gets to base3, the compiler gradually uses up more and more
memory, until it either gets killed by the kernel OOM stuff, or dies
with "gcc: Internal error: Killed (program cc1)".

This is with 4.1.2.alpha1, but the same behavior occurred with 4.1.1. In
Karmic, the compiler is:

$ gcc -v
Using built-in specs.
Target: i486-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Ubuntu
4.4.1-3ubuntu3' --with-bugurl=file:///usr/share/doc/gcc-4.4/README.Bugs
--enable-languages=c,c++,fortran,objc,obj-c++ --prefix=/usr
--enable-shared --enable-multiarch --enable-linker-build-id
--with-system-zlib --libexecdir=/usr/lib --without-included-gettext
--enable-threads=posix --with-gxx-include-dir=/usr/include/c++/4.4
--program-suffix=-4.4 --enable-nls --enable-clocale=gnu
--enable-libstdcxx-debug --enable-mpfr --enable-objc-gc
--enable-targets=all --disable-werror --with-arch-32=i486
--with-tune=generic --enable-checking=release --build=i486-linux-gnu
--host=i486-linux-gnu --target=i486-linux-gnu
Thread model: posix
gcc version 4.4.1 (Ubuntu 4.4.1-3ubuntu3)

On a current Ubuntu Hardy system (like sage.math, but a 32-bit VM with
768 MB RAM) the compiler goes through base3.c with no trouble.

Can anyone else confirm (or perhaps even fix) this?

Dan

--
--- Dan Drake
----- http://mathsci.kaist.ac.kr/~drake
-------

[Minh Nguyen]

Hi Dan,

On Thu, Sep 10, 2009 at 5:48 PM, Dan Drake <dr...@kaist.edu> wrote:

<SNIP>

> Can anyone else confirm (or perhaps even fix) this?

Jaap Spies reported the same error when compiling Sage 4.1.2.alpha0 on
32-bit Fedora 11. See

http://groups.google.com/group/sage-devel/browse_thread/thread/2ea99a973a54581c/41944cbfee79bcf8

--
Regards
Minh Van Nguyen

[Dan Drake]

I tried the compilation on an Arch Linux VM with gcc 4.4.1 and it worked
fine. So this is something related to Ubuntu's and Fedora's gcc
packages.

[William Stein]

This is a GCC bug which has been discussed a lot on the Pari listed,
and reported in at least three places to GCC and Ubuntu. The only
stated workaround at present is to build pari with -O0 instead of -O3:

http://www.nabble.com/Ubuntu-9.10-and-gcc-4.4-failure-td24853751.html

William

William

[Jason Grout]

Following the links leads to this bug report:

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41101#c23

which indicates that this may have been fixed yesterday in the GCC trunk.

Jason

[Tan Swee Heng]

A quick fix I used was to change OPTFLAGS=-O3 in PARI's get_cc script to OPTFLAGS=-O1. I wrote down the specific steps here: <http://thesweeheng.wordpress.com/2009/08/28/building-sage-4-1-1-on-fedora-11/>. Hope that helps.

Swee Heng

[J Elaych]

The problem is now fixed with the latest gcc/g++ in Ubuntu Karmic.

>  signature.asc
> < 1KViewDownload