Re: Patch: RFA: set LD_LIBRARY_PATH

[Andrew Haley <aph at redhat dot com>]

Thomas Fitzsimmons writes:
 > On Mon, 2005-09-05 at 14:05 +0100, Andrew Haley wrote:
 > > Thomas Fitzsimmons writes:
 > >  > 
 > >  > Sun's JRE includes $JAVA_HOME/lib/<arch> in its LD_LIBRARY_PATH so that
 > >  > users don't need to specify LD_LIBRARY_PATH manually.  This patch makes
 > >  > libgcj do the same when the --with-java-home configure option is
 > >  > specified.
 > >  > 
 > >  > OK for mainline?
 > > 
 > > In what circumstances is this necessary?
 > 
 > The specific use case I'm trying to support with this patch is running
 > Java applications that link against libjawt.so.  Sun installs this
 > library in $JAVA_HOME/lib/<arch>.  See:
 > 
 > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21741
 > 
 > >   Should not the appropriate
 > > libraries be installed somewhere where LD_LIBRARY_PATH isn't needed?
 > 
 > We tried that before, see:
 > 
 > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20790
 > 
 > When our JAWT implementation was installed in GCC's library prefix it
 > was overriding Sun's libjawt.so when compiling JAWT-using applications
 > against proprietary VMs.

OK, so we renamed the lib to libgcjawt.so.  Why can't libgcjawt.so be
installed in the common place for shared libraries, /usr/lib?  Its
name prevents it from colliding with other libraries.

 > > What if a user specifically sets LD_LIBRARY_PATH in order to _avoid_
 > > $JAVA_HOME/lib/<arch>?
 > 
 > They can't avoid it.  But I don't know why they would want to.  The idea
 > here is that LD_LIBRARY_PATH only includes $JAVA_HOME/lib/<arch> when --
 > with-java-home is specified.

I understand that, but most of our users don't build their own libgcj.

Making libgcj mess with LD_LIBRARY_PATH at runtime is a pretty evil
hack.  If users are going to link their own programs with -ljawt, then
they're going to be setting LD_LIBRARY_PATH anyway.

I guess the problem is I still don't understand the scenario you have
in mind.

Andrew.