| Summary: | Compaq JRE complains about unresolved symbols | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Stefaan De Roeck (RETIRED) <stefaan> |
| Component: | [OLD] Library | Assignee: | Alpha Porters <alpha> |
| Status: | RESOLVED LATER | ||
| Severity: | normal | CC: | java |
| Priority: | High | ||
| Version: | 2004.0 | ||
| Hardware: | Alpha | ||
| OS: | All | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
| Attachments: | limewire strace log with __libc_accept patch | ||
|
Description
Stefaan De Roeck (RETIRED)
2005-03-06 08:25:34 UTC
alpha, please advise. this one is for you, since we can't reproduce it without access to such a nifty platform :( I resolved the problem on my local system, after some googling. The resolution is a workaround (necessary as the Compaq JRE is binary only) that implements the obsolete __libc_accept function in a seperate shared library, and then LD_PRELOAD's this. I got this from https://www.redhat.com/archives/axp-list/2004-August/msg00019.html Create a source file with the following contents: #include <errno.h> #include <sys/syscall.h> #include <sys/types.h> #include <sys/wait.h> __pid_t __libc_wait (__WAIT_STATUS_DEFN stat_loc) { return __waitpid (WAIT_ANY, (int *) stat_loc, 0); } and call it libcwait.c. Then do gcc -O2 -shared -o libcwait.so libcwait.c. Put this binary in a safe place, and change the java wrapper-script so that it does export LD_PRELOAD=${path}/libcwait.so before loading the actual binary. Hi First of all, and due to the actual situation of Compaq-java (upstream dead long time ago) i don't know if we are going to be able to solve this issue. Well, after the disclaimer, lets going to see what can we do: #1. Yes, we can reproduce the problem easily. I'm using compaq-jdk-1.3.1-r3 and limewire-4.8.1 . Limewire compiles just fine but, when you start it, it shows to you the error reported in the description of this bug about __libc_accept. #2. Problem with __libc_wait appeared on gentoo-alpha mailing list some time ago [1]. Stefaan, I found the same patch than you posted here, and seems to work fine. The problem here is that __libc_wait patch doesn't implement __libc_accept signal (at least for me). I had the same result with java-jdk patched this way. Stefaan, would you mind emerging limewire-4.8.1 (using your current compaq-jre patched), run it and post the results here? [1] http://marc.theaimsgroup.com/?t=112020017200001&r=1&w=2 Hmm... Maybe I'm being too stubborn, but anyway.
I was indeed too quick to think this issue was fixed, there were other
unresolved symbols. I looked them up in older glibc versions. The __libc_*
versions seem to be copies of the versions without the prefixes (apart from some
thing with weak aliases that I didn't completely understand). So I updated my
libcwait.c file:
#include <errno.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <sys/wait.h>
__pid_t
__libc_wait (__WAIT_STATUS_DEFN stat_loc)
{
return __waitpid (WAIT_ANY, (int *) stat_loc, 0);
}
#include <sys/socket.h>
int __libc_accept (int a1, struct sockaddr* a2, socklen_t* a3)
{
return accept (a1, a2, a3);
}
int __libc_recvfrom (int a1, void * a2, size_t a3, int a4,
struct sockaddr * a5, socklen_t * a6)
{
return recvfrom (a1, a2, a3, a4, a5, a6);
}
int __libc_sendto (int a1, void * a2, size_t a3, int a4,
struct sockaddr * a5, socklen_t a6)
{
return sendto (a1, a2, a3, a4, a5, a6);
}
int __libc_recv (int a1, void * a2, size_t a3, int a4)
{
return recv (a1, a2, a3, a4);
}
Now I enjoy the limewire 4.8.1 splash screen!! Hurray... almost
The stdout/err messages I see are:
Warning: Cannot find callback list in XtAddCallback
Warning: Cannot find callback list in XtAddCallback
log4j:WARN No appenders could be found for logger
(com.limegroup.gnutella.gui.Initializer).
log4j:WARN Please initialize the log4j system properly.
I don't think the last two messages matter, as I get them on x86 as well, but
the previous two might constitute a problem. Or maybe my approach extending
libcwait.c was wrong. I'm clueless for the moment. Maybe you know what's going
wrong now?
"Warning: Cannot find callback list in XtAddCallback" this is caused by a bad
openmotif DEPEND in the ebuild:
>=x11-libs/openmotif-2.1.30-r1 should be
~x11-libs/openmotif-2.1.30-r1
because compaq-{jre,jdk} seems to be statically linked to 2.1 openmotif libs.
You can solve this problem with "emerge =x11-libs/openmotif-2.1.30-r9" (stable
2.1 openmotif). Don't worry, you aren't going to lose your 2.2 branch since the
ebuilds use SLOT (2.1 and 2.2).
After did this, i tried limewire. I recieve warnings from log4j, see the splash
screen, but nothing happends after this. No errors, no program just the splash
screen. I'll attach an strace log of limewire.
Maybe libc_accept patch have something wrong. We could know if this is the
problem if find another app with the same issue. Sorry but i can't help you with
this patch because I lack of glibc-ninja-skills :(
I'm planning on bumping revision of compaq-{jre,jdk} to solve this error with
openmotif and the libcwait signal bug. I wait for the moment to see what can we
do with this bug and another issues with xml support.
Stefaan, thanks a lot for your help.
Created attachment 66515 [details]
limewire strace log with __libc_accept patch
The current status of Java on Alpha doesn't let us handle Java bugs properly. We will revisit all Java related bugs once we find a good Java Virtual Machine for the Alpha Architecture. |