GC seems to kill process

[Andi Vajda]

On Wed, 22 Mar 2006, Bryce McKinlay wrote:

> Andi,
>
> Can you provide a test case, or some directions to reproduce the problem?

I haven't been able to isolate this out of PyLucene and Python, and only on 
Linux so far. Here are instructions for how to build all this:

   - Build your own gcc 4.1.0 using this (I used the release source tarball):
     . add #define LARGE_CONFIG to gcc-4.1.0/boehm-gc/include/private/gc_priv.h
     . configure --enable-threads=posix --prefix=/usr/local/gcc-4.1.0
                 --enable-languages=c,c++,java
       (I'm sure you all know how to build gcc yourselves, I just wanted to
        list the configure flags I used).
     . make bootstrap; make install

   - Build Python 2.4.2 with that or another gcc, it doesn't seem to make a
     difference. Building python is really simple:
      ./configure; make; make install
     If you already have a version of Python 2.4.x installed, you can skip this
     step.

   - Download swig 1.3.24 from http://www.swig.org and build it. It must be
     swig 1.3.24, later versions are not working with PyLucene 1.9.1
     ./configure; make; make install

   - Download the PyLucene source tarball from
     http://downloads.osafoundation.org/PyLucene/src/PyLucene-src-1.9.1.tar.gz

   - Edit PyLucene's Makefile Linux section to set some variables vital to its
     build, this may be tricky if your gcc installation is all spread out, but
     really simple if you built your own gcc and installed it all in one tree.
     No need to compile DB support in. Ant and a JDK are *not* needed when
     building from a PyLucene source tarball.

   - make; make install or make DEBUG=1; make DEBUG=1 install if you want a
     DEBUG version with symbols for gdb.

   - start python and reproduce the bug:
     > python
     >>> import PyLucene
     >>> PyLucene.System.gc()
     --> process killed

Andi..