solib: URL patch

[Tom Tromey]

>>>>> "Andrew" == Andrew Haley <aph@redhat.com> writes:

>> This restriction will be lifted once we move to the new ABI.  Then we
>> will be able to load .so files privately (instead of RTLD_GLOBAL).

Andrew> Uh, we've decide this for sure, have we?  I mean, it sounds like a
Andrew> good idea, but have you dot a solid design in mind?  We still need a
Andrew> mechanism that will allow us to call .so files from CNI code, I guess.

I don't have a problem saying we've decided to move to the new ABI for
sure.  What else can we do?  With global linkage, I think we can't
satisfy the semantics of class loaders; we've already run into
problems here.

I don't know how to solve the CNI problem with the new ABI.  Modifying
g++ seems like the best bet, but even this has some problems (how do
you tell the runtime what classes and methods are needed by the CNI
code?).

Even a CNI-less new ABI would be a useful improvement, though.  It
would let us provide actually-conforming binary-compatible
precompilation support for all the JNI-using and pure Java
applications.

Too bad there's no way to hook into ld.so's symbol lookup and replace
it with our own...

Tom