This is the mail archive of the
java@gcc.gnu.org
mailing list for the Java project.
Re: CNI changes (Was: Binary Compatibility)
- From: Tom Tromey <tromey at redhat dot com>
- To: Anthony Green <green at redhat dot com>
- Cc: Andrew Haley <aph at redhat dot com>, java at gcc dot gnu dot org
- Date: 04 Aug 2003 21:56:06 -0600
- Subject: Re: CNI changes (Was: Binary Compatibility)
- References: <Pine.LNX.4.44.0307300929080.11078-100000@ops2.one-point.com><87smoot8xw.fsf@fleche.redhat.com><16167.60120.49100.864112@cuddles.cambridge.redhat.com><1059964034.4179.643.camel@escape>
- Reply-to: tromey at redhat dot com
>>>>> "Anthony" == Anthony Green <green@redhat.com> writes:
Anthony> Are there any other CNI changes we should make at this same
Anthony> time?
Something that came up on irc tonight is letting libgcj.so be
dlopen()d without RTLD_GLOBAL. Currently Mozilla does this.
We could support this with the new ABI. First we would have to have a
function vector like JNI does, both for exported CNI functions, and
for functions to which gcj generates direct calls. The latter (and
perhaps the former) could just be part of the class' otable.
Second, I think we'd have to arrange to look up _Jv_RegisterClasses
explicitly, using dlsym(), before registering classes from a .so.
Tom