libltdl proposal for libgcj

[Andrew Haley]

Anthony Green writes:
 > On Tue, 2004-07-13 at 11:55, Bryce McKinlay wrote:
 > > I wonder if, with the new ABI, we ought to think about just ripping out 
 > > this name-based resolution scheme. I'm thinking that some kind of 
 > > extdir+classpath approach, with shared libraries instead of .jars, would 
 > > solve the problem in a simpler and more Java-compatible way.
 > 
 > Can you give a bit more detail on what you have in mind?

I'm not Bryce, but there are a couple of possible approaches.

Firstly, as Bryce suggested we could just treat shared objects in the
same way as classfiles, put them in the classpath, and resolve their
names in a compatible way.  This will require some internal libgcj
weirdness.

Secondly, a totally interpreter compatible approach is to "attach" .so
files to their corresponding jar files.  We then do a lookaside at
class load time from the .jar file to the .so file.  This has the
disadvantage that we need the .jar file to be installed, but as these
aren't very large it's not an onerous requirement.

As to how we actually do the "attachment", ATM I'm thinking of using
extended file attributes.

The first approach is conceptually cleaner, and perhaps to be
preferred in a gcj only environmant.

Andrew.