On Thu, 1 Jul 2010, Bill Janssen wrote:
Andi Vajda <va...@apache.org> wrote:
I'm trying to build Lucene 2.9.2 on Fedora with gcc 4.3 and Python 2.5.1
and OpenJDK 6, but hit this issue with JArray:
[...]
building 'lucene._lucene' extension
creating build/temp.linux-i686-2.5
creating build/temp.linux-i686-2.5/build
creating build/temp.linux-i686-2.5/build/_lucene
gcc -pthread -fno-strict-aliasing -DNDEBUG -O2 -g -pipe -Wall
-Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector
--param=ssp-buffer-size=4 -m32 -march=i386 -mtune=generic
-fasynchronous-unwind-tables -D_GNU_SOURCE -fPIC -fPIC -DPYTHON=1
-D_jcc_shared=1 -I/usr/lib/jvm/java-6-openjdk/include
-I/usr/lib/jvm/java-6-openjdk/include/linux -Ibuild/_lucene
-I/usr/lib/python2.5/site-packages/JCC-2.4.1-py2.5-linux-i686.egg/jcc/sources
-I/usr/include/python2.5 -c build/_lucene/JArray.cpp -o
build/temp.linux-i686-2.5/build/_lucene/JArray.o -fno-strict-aliasing
-Wno-write-strings
This error indicates that the C++ compiler thinks that jboolean and
jbyte are the same types.
Actually, the problem is the same old one that we've discussed before.
There is no "/usr/lib/jvm/java-6-openjdk" directory. When I built JCC,
I had the environment variable JCC_JDK set to point to
/usr/lib/jvm/java-1.6.0-openjdk, which is where openjdk decided to
install itself. But, at least in 2.9.2, JCC doesn't remember that and
re-use that knowledge. Instead, in the above, it generates a completely
bogus and non-existent path to pull the include file headers from. And
gcc, confused, falls back to the gcj headers.
If you're talking about jcc's forgetting this between 'build' and 'install'
when run separately, that's been fixed in jcc 2.6, the version currently
awaiting votes for release.
In 2.6, the config.py file is not written unless it needs to be.
See code in helpers/build.py for more details.
Starting with the upcoming pylucene 2.9.3 / 3.0.2 releases, all pylucene
versions use the same version of jcc, currently 2.6.
Andi..
