This is the mail archive of the
java@gcc.gnu.org
mailing list for the Java project.
Re: Solaris redux
- To: Bryce McKinlay <bryce at albatross dot co dot nz>
- Subject: Re: Solaris redux
- From: Jeff Sturm <jsturm at one-point dot com>
- Date: Tue, 27 Mar 2001 19:05:26 -0500 (EST)
- cc: java at gcc dot gnu dot org
On Wed, 28 Mar 2001, Bryce McKinlay wrote:
> I'm not sure it really helps this case though - if the
> C++ static constructors arn't executing however, what about GCJ's class
> registration functions?
Some of them are working. It's hard to say how many. Debugging the
constructors is difficult, since I cannot set a breakpoint until they
load, but once they load they've already run :(
I don't really understand the problem yet, nor do I understand how libgcj
guarantees constructors execute in the right order. What happens if
_Jv_makeUtf8Const runs before _Jv_InitGC, for example?
> Good idea. I'll try that. It also seems conceptually better to have the
> dependencies in libgcj.so itself, rather than relying on the application to
> link them in, given that the application does not use either directly.
It's also more correct. Symbol versioning won't work otherwise, and I
have a hunch that ld.so needs the dependencies to init libraries in the
correct order. I've already seen failures in complex applications that
were "corrected" just by changing the linking order.
Jeff