From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 16373 invoked by alias); 24 May 2005 18:15:48 -0000 Mailing-List: contact java-prs-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: java-prs-owner@gcc.gnu.org Received: (qmail 16349 invoked by uid 48); 24 May 2005 18:15:47 -0000 Date: Tue, 24 May 2005 18:15:00 -0000 Message-ID: <20050524181547.16347.qmail@sourceware.org> From: "pinskia at gcc dot gnu dot org" To: java-prs@gcc.gnu.org In-Reply-To: <20050521024903.21692.kkojima@gcc.gnu.org> References: <20050521024903.21692.kkojima@gcc.gnu.org> Reply-To: gcc-bugzilla@gcc.gnu.org Subject: [Bug libgcj/21692] [4.1 Regression] unexpected java.lang.NoClassDefFoundError X-Bugzilla-Reason: CC X-SW-Source: 2005-q2/txt/msg00550.txt.bz2 List-Id: ------- Additional Comments From pinskia at gcc dot gnu dot org 2005-05-24 18:15 ------- Confirmed, here is some analysis from RTH on IRC: [16:54] < rth> we hashed a plt entry. [16:54] < rth> grrr. [17:30] < tromey> ah [17:30] < tromey> I'm glad you debugged this then [17:32] < rth> so the problem is that we're not marking some methods as local, which means the make_local_function_alias trick fails. [17:35] < tromey> in make_local_function_alias I see: [17:35] < tromey> if (DECL_EXTERNAL (method)) [17:35] < tromey> return method; [17:35] < tromey> the problem here is that all CNI native methods will be DECL_EXTERNAL, I think [17:35] < tromey> is this even fixable? (I have no idea about this code either) [17:35] < rth> oh, is forName a cni method? [17:36] < tromey> there are 2 overloads, one is native [17:36] < tromey> oh, wait, both are native [17:36] < tromey> I didn't realize that [17:37] < rth> it's very tricky. [17:37] < tromey> ok [17:37] < rth> we would have to create hidden aliases for all native methods, and Just Know to use that alias when building the class table. [17:38] < rth> so i guess it's not that tricky. just a bit o coordination between c++ and java front ends. [17:38] < tromey> oh, I see... modify g++ to also emit the hidden aliases [17:38] < rth> It does assume that cni methods are linked together with its java class. [17:38] < rth> yes. [17:39] < tromey> at the moment there is no other way to really use cni methods [17:39] < tromey> so that is fine, I think [17:40] < rth> i'll see what i can come up with. [17:40] < tromey> thank you And this is reason why it works just fine on ppc-darwin, because darwin always use the function address and not a stub/plt entry for the function pointers. -- What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |NEW Ever Confirmed| |1 Last reconfirmed|0000-00-00 00:00:00 |2005-05-24 18:15:46 date| | http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21692