You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Mark Phippard <ma...@softlanding.com> on 2005/11/01 19:48:05 UTC

Possible Problem with JavaHL in 1.3.0 RC2

I am in a bit of a jam here in that I cannot figure out how to proceed.

JavaHL passes its JUnit tests just fine, but when I use it in Subclipse I 
get a hard JVM crash every time.  Some things, like status and ls seem to 
work OK, but others like checkout, commit, update always crash. 

Subclipse uses SVNClientSychronized, but if you change the JUnit tests to 
use that class, they still work OK.

I am attaching the JVM dump but there isn't much of value in it that I can 
see.  I have tried using different JVM's.

I would appreciate any ideas anyone has.  I am not a C developer so I 
cannot really debug this.  I could post a special version of Subclipse if 
someone has Eclipse and wants to try this, especially someone that could 
debug it.

I run Win32, the problem could very well be isolated to that environment, 
or even to my machine.

Thanks

Mark





_____________________________________________________________________________
Scanned for SoftLanding Systems, Inc. and SoftLanding Europe Plc by IBM Email Security Management Services powered by MessageLabs. 
_____________________________________________________________________________

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Mark Phippard <ma...@softlanding.com>.
Patrick Mayweg <ma...@qint.de> wrote on 11/02/2005 09:02:29 AM:

> that does not help much. The instruction pointer points into the code of 

> the C++ Outpuuter class. This class is not used by either of the methods 

> which crash.
> Since the normal tests do not crash, I think it code depend on your 
> repository. It was same with the stack overflow on merge. Didyou use a 
> public repository for your tests?
> It is a NULL-Pointer in the native code.

Interesting. 

I was able to successfully checkout JavaHL from the Subversion trunk.

However, if I try to checkout anything from the Subclipse repository it 
crashes. 

My internal repositories (running 1.2.3 on OS/400) are a mixed bag. I have 
an svn:// repository that consistently fails.  One of my http:// 
repositories was failing yesterday but seems to be working now (at least 
for checkout).

If you want to try with Subclipse, here is a very small project to 
checkout:

http://subclipse.tigris.org/svn/subclipse/trunk/subclipse/feature/

Login as guest/guest or your tigris.org account.

Mark


_____________________________________________________________________________
Scanned for SoftLanding Systems, Inc. and SoftLanding Europe Plc by IBM Email Security Management Services powered by MessageLabs. 
_____________________________________________________________________________

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Philip Martin <ph...@codematters.co.uk>.
Mark Phippard <ma...@softlanding.com> writes:

> I seem to have worked around the problem by calling
> getAdminDirectoryName() at startup when I know there is no other
> activity.  I then just cache it (which I was doing anyway for
> performance).  I then had to implement my own isAdminDirectoryName()
> which returns true if the folder name is ".svn" or whatever the
> getAdminDirectoryName() value is.  This avoids doing any library
> calls in the middle of operations.

If you have to go as far as implementing your own isAdminDirectoryName
then that would suggest threading is totally broken, and that it's not
just an initialisation problem.  Given that you now have a thread-safe
call to JNIInit (the first getAdminDirectoryName call) does your code
work if you continue to make that call but stop using your own
isAdminDirectoryName?

If that works the next step is to replace the first, thread-safe, call
to getAdminDirectoryName with some other function that calls JNIInit,
say 'singleStatus("fake", false)'.  If that works it indicates that
the problem is not the SVN_ASP_DOT_NET_HACK code itself, but that it's
an initialisation problem that happens to be triggered by the way you
are using the SVN_ASP_DOT_NET_HACK code.

-- 
Philip Martin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Mark Phippard <ma...@softlanding.com>.
Philip Martin <ph...@codematters.co.uk> wrote on 11/02/2005 03:47:19 PM:

> Mark Phippard <ma...@softlanding.com> writes:
> 
> > Looking at the history of the code, however, wouldn't really explain 
what 
> > has changed between 1.2 and 1.3 that would cause a new problem.  This 
code 
> > was all present in 1.2 and earlier.
> 
> This is low level initialisation code, perhaps the higher level code
> now calls it more often so provoking a race?  Or perhaps the higher
> level code used to have some bottleneck that effectively serialised
> the initialisation and that has now been removed?  I don't do Java at
> all, I'm just guessing.

I think I have found a workaround to this problem.  I basically started 
over with my 1.3 integration and did it one piece at a time.  The problem 
resurfaced when I added the code to Subclipse to call the new methods for 
the SVN_ASP_DOT_NET_HACK feature.  I believe it is some sort of a 
threading problem.  Take checkout as an example.  I think what would 
happen is I would start a checkout and the folder and first file would 
come down.  This would cause the notification framework to notify 
Subclipse of the new file.  We in turn notify Eclipse, which I think 
notifies our status cache (probably in a different thread).  That code 
then sees the ".svn or _svn folder" and calls one of these new library 
methods to see if it is an admin directory.

I seem to have worked around the problem by calling 
getAdminDirectoryName() at startup when I know there is no other activity. 
 I then just cache it (which I was doing anyway for performance).  I then 
had to implement my own isAdminDirectoryName() which returns true if the 
folder name is ".svn" or whatever the getAdminDirectoryName() value is. 
This avoids doing any library calls in the middle of operations.

At least this is all my theory.  I will let you know if any crashes come 
back as I do more.

Thanks

Mark


_____________________________________________________________________________
Scanned for SoftLanding Systems, Inc. and SoftLanding Europe Plc by IBM Email Security Management Services powered by MessageLabs. 
_____________________________________________________________________________

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Philip Martin <ph...@codematters.co.uk>.
Mark Phippard <ma...@softlanding.com> writes:

> Looking at the history of the code, however, wouldn't really explain what 
> has changed between 1.2 and 1.3 that would cause a new problem.  This code 
> was all present in 1.2 and earlier.

This is low level initialisation code, perhaps the higher level code
now calls it more often so provoking a race?  Or perhaps the higher
level code used to have some bottleneck that effectively serialised
the initialisation and that has now been removed?  I don't do Java at
all, I'm just guessing.

-- 
Philip Martin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Mark Phippard <ma...@softlanding.com>.
Philip Martin <ph...@codematters.co.uk> wrote on 11/02/2005 03:15:38 PM:

> JNIInit appears to attempt to do some once-only initialisation (that's
> once per application) using static variables but the code looks very
> dodgy.  I don't do Java at all, and I've barely looked at the javahl
> code, but I suspect that if you can arrange for a single thread to
> call JNIInit, and allow it to complete, before any other thread gets a
> chance to race then the problem might go away.  I don't know what that
> would involve in Java, perhaps it would be sufficient to construct and
> discard a temporary SVNClientSynchronized object.  Do it from the
> debugger if necessary.

Not being a C programmer I cannot really comment on what that code does. 
Looking at the history of the code, however, wouldn't really explain what 
has changed between 1.2 and 1.3 that would cause a new problem.  This code 
was all present in 1.2 and earlier.

Thanks

Mark


_____________________________________________________________________________
Scanned for SoftLanding Systems, Inc. and SoftLanding Europe Plc by IBM Email Security Management Services powered by MessageLabs. 
_____________________________________________________________________________

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Philip Martin <ph...@codematters.co.uk>.
Mark Phippard <ma...@softlanding.com> writes:

> Philip Martin <ph...@codematters.co.uk> wrote on 11/02/2005 02:42:04 PM:
>
>> Do the crashes happen near the start of the operations?
>
> Yes, for certain.  Although in the case of checkout, sometimes it seems 
> like I receive 1 file successfully.  I haven't really looked at the WC too 
> closely, but I see a message in my Eclipse console right before it 
> crashes.

So it probably is an initialisation problem, rather than memory
corruption.

>> As I pointed out earlier, the on-the-fly static initialisation in
>> JNIUtil::JNIInit() looks dodgy to me.  Could you arrange for JNIInit
>> to be called, and for the call to return, by a single thread before
>> any other calls to JNIInit?
>
> In Subclipse we talk to the Java code, so this is all invisible to us.  I 
> can say that Subclipse constructs the SVNClientSynchronized object 
> frequently.  It doesn't just construct it one time and then hang on to it 
> until you close Eclipse.

JNIInit appears to attempt to do some once-only initialisation (that's
once per application) using static variables but the code looks very
dodgy.  I don't do Java at all, and I've barely looked at the javahl
code, but I suspect that if you can arrange for a single thread to
call JNIInit, and allow it to complete, before any other thread gets a
chance to race then the problem might go away.  I don't know what that
would involve in Java, perhaps it would be sufficient to construct and
discard a temporary SVNClientSynchronized object.  Do it from the
debugger if necessary.

-- 
Philip Martin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Mark Phippard <ma...@softlanding.com>.
Philip Martin <ph...@codematters.co.uk> wrote on 11/02/2005 02:42:04 PM:

> Do the crashes happen near the start of the operations?

Yes, for certain.  Although in the case of checkout, sometimes it seems 
like I receive 1 file successfully.  I haven't really looked at the WC too 
closely, but I see a message in my Eclipse console right before it 
crashes.

> As I pointed out earlier, the on-the-fly static initialisation in
> JNIUtil::JNIInit() looks dodgy to me.  Could you arrange for JNIInit
> to be called, and for the call to return, by a single thread before
> any other calls to JNIInit?

In Subclipse we talk to the Java code, so this is all invisible to us.  I 
can say that Subclipse constructs the SVNClientSynchronized object 
frequently.  It doesn't just construct it one time and then hang on to it 
until you close Eclipse.

Mark


_____________________________________________________________________________
Scanned for SoftLanding Systems, Inc. and SoftLanding Europe Plc by IBM Email Security Management Services powered by MessageLabs. 
_____________________________________________________________________________

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Philip Martin <ph...@codematters.co.uk>.
Mark Phippard <ma...@softlanding.com> writes:

> Philip Martin <ph...@codematters.co.uk> wrote on 11/02/2005 11:42:48 AM:
>> 
>> Perhaps getEnv is returning NULL?
>> 
>> The static bool stuff in JNIUtil::JNIInit looks a bit dodgy.
>
> I have done some additional crash scenarios (commit, checkout) and they 
> produce the same strack trace.

Do the crashes happen near the start of the operations?

> As I am not a C programmer I cannot really analyze the code, but I would 
> think you are right.

If you change the code in isJavaExceptionThrown from

   if (env->ExceptionCheck())

to

   if (env && env->ExceptionCheck())

then it won't crash if env is NULL.  However all the other uses of
getEnv assume that NULL is never returned so this isn't really a
solution, it will probably just move the crash somewhere else.

>  Perhaps it has something to do with the 
> SVN_ASP_DOT_NET_HACK code that was added to the initialization.  However, 
> none of that code should be compiled in the Linux version so it would seem 
> odd.

The SVN_ASP_DOT_NET_HACK code looks innocent to me.

> Also, I am still at a loss to explain what is different about Subclipse 
> using this and the test suite, although I imagine it has something to do 
> with why it works for some repositories and not others.  That being said, 
> I hacked up a JUnit test that checks out that URL I have before and it 
> works fine.  So maybe it is something to do with how the library is 
> initialized when it is run from Subclipse?

As I pointed out earlier, the on-the-fly static initialisation in
JNIUtil::JNIInit() looks dodgy to me.  Could you arrange for JNIInit
to be called, and for the call to return, by a single thread before
any other calls to JNIInit?

-- 
Philip Martin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Mark Phippard <ma...@softlanding.com>.
Philip Martin <ph...@codematters.co.uk> wrote on 11/02/2005 11:42:48 AM:

> Mark Phippard <ma...@softlanding.com> writes:
> 
> > Stack: [0xaff57000,0xaffd8000),  sp=0xaffd61e8,  free space=508k
> > Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, 
C=native code)
> > C  [libsvnjavahl-1.so.0.0.0+0xf2b1] 
_ZN7JNIUtil21isJavaExceptionThrownEv+0x21
> > C  [libsvnjavahl-1.so.0.0.0+0x10afc] 
> _ZN7Notify28onNotifyEPK15svn_wc_notify_tP10apr_pool_t+0x32c
> > C  [libsvnjavahl-1.so.0.0.0+0x107c1] 
> _ZN7Notify26notifyEPvPK15svn_wc_notify_tP10apr_pool_t+0x31
> 
> $ echo _ZN7JNIUtil21isJavaExceptionThrownEv | c++filt
> JNIUtil::isJavaExceptionThrown()
> 
> $ echo _ZN7Notify28onNotifyEPK15svn_wc_notify_tP10apr_pool_t | c++filt
> Notify2::onNotify(svn_wc_notify_t const*, apr_pool_t*)
> 
> Perhaps getEnv is returning NULL?
> 
> The static bool stuff in JNIUtil::JNIInit looks a bit dodgy.

I have done some additional crash scenarios (commit, checkout) and they 
produce the same strack trace.

As I am not a C programmer I cannot really analyze the code, but I would 
think you are right.  Perhaps it has something to do with the 
SVN_ASP_DOT_NET_HACK code that was added to the initialization.  However, 
none of that code should be compiled in the Linux version so it would seem 
odd.

Also, I am still at a loss to explain what is different about Subclipse 
using this and the test suite, although I imagine it has something to do 
with why it works for some repositories and not others.  That being said, 
I hacked up a JUnit test that checks out that URL I have before and it 
works fine.  So maybe it is something to do with how the library is 
initialized when it is run from Subclipse?

I can post an installable version of Subclipse for anyone that wants to 
test this.  You need this version because it has the required 1.3 changes 
in it.

Thanks

Mark




_____________________________________________________________________________
Scanned for SoftLanding Systems, Inc. and SoftLanding Europe Plc by IBM Email Security Management Services powered by MessageLabs. 
_____________________________________________________________________________

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Philip Martin <ph...@codematters.co.uk>.
Mark Phippard <ma...@softlanding.com> writes:

> Stack: [0xaff57000,0xaffd8000),  sp=0xaffd61e8,  free space=508k
> Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
> C  [libsvnjavahl-1.so.0.0.0+0xf2b1]  _ZN7JNIUtil21isJavaExceptionThrownEv+0x21
> C  [libsvnjavahl-1.so.0.0.0+0x10afc]  _ZN7Notify28onNotifyEPK15svn_wc_notify_tP10apr_pool_t+0x32c
> C  [libsvnjavahl-1.so.0.0.0+0x107c1]  _ZN7Notify26notifyEPvPK15svn_wc_notify_tP10apr_pool_t+0x31

$ echo _ZN7JNIUtil21isJavaExceptionThrownEv | c++filt
JNIUtil::isJavaExceptionThrown()

$ echo _ZN7Notify28onNotifyEPK15svn_wc_notify_tP10apr_pool_t | c++filt
Notify2::onNotify(svn_wc_notify_t const*, apr_pool_t*)

bool JNIUtil::isJavaExceptionThrown()
{
    JNIEnv *env = getEnv();
    if(env->ExceptionCheck())
    {
        // retrieving the exception removes it
        // so we rethrow it here
        jthrowable exp = env->ExceptionOccurred();
        env->ExceptionDescribe();
        env->Throw(exp);
        env->DeleteLocalRef(exp);
        setExceptionThrown();
        return true;
    }
    return false;
}

Perhaps getEnv is returning NULL?

The static bool stuff in JNIUtil::JNIInit looks a bit dodgy.

-- 
Philip Martin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Mark Phippard <ma...@softlanding.com>.
Patrick Mayweg <ma...@qint.de> wrote on 11/02/2005 09:02:29 AM:

> that does not help much. The instruction pointer points into the code of 

> the C++ Outpuuter class. This class is not used by either of the methods 

> which crash.
> Since the normal tests do not crash, I think it code depend on your 
> repository. It was same with the stack overflow on merge. Didyou use a 
> public repository for your tests?
> It is a NULL-Pointer in the native code.

OK, I was able to recreate this on Ubuntu Linux.  I have attached the JVM 
dump from that system.  In this case, I already had the Subclipse "core" 
project checked out in my workspace and I just tried to do an Update.

Thanks

Mark




_____________________________________________________________________________
Scanned for SoftLanding Systems, Inc. and SoftLanding Europe Plc by IBM Email Security Management Services powered by MessageLabs. 
_____________________________________________________________________________

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Patrick Mayweg <ma...@qint.de>.
Hi Mark,
that does not help much. The instruction pointer points into the code of 
the C++ Outpuuter class. This class is not used by either of the methods 
which crash.
Since the normal tests do not crash, I think it code depend on your 
repository. It was same with the stack overflow on merge. Didyou use a 
public repository for your tests?
It is a NULL-Pointer in the native code.
Regards,
Patrick
Mark Phippard wrote:

>Joseph Galbraith <ga...@vandyke.com> wrote on 11/01/2005 06:36:40 PM:
>
>  
>
>>MSVC can indeed produce a link map (/MAP or /MAP:filename to the
>>linker.)
>>    
>>
>
>OK, I rebuilt the DLL with the MAP file.  I then did a checkout, update 
>and commit, each of which crashed the JVM and produced a dump file.  I 
>have posted a zip file with the 3 dumps and the MAP file.  Hopefully that 
>will provide more clues.
>
>The zip file is 138 Kb and you can get it here:
>
>http://support.softlanding.com/download/javahl.zip
>
>I am going to see if I can get Ubuntu setup to test this.
>
>Thanks
>
>Mark
>
>
>_____________________________________________________________________________
>Scanned for SoftLanding Systems, Inc. and SoftLanding Europe Plc by IBM Email Security Management Services powered by MessageLabs. 
>_____________________________________________________________________________
>
>---------------------------------------------------------------------
>To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
>For additional commands, e-mail: dev-help@subversion.tigris.org
>
>  
>



---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Mark Phippard <ma...@softlanding.com>.
Joseph Galbraith <ga...@vandyke.com> wrote on 11/01/2005 06:36:40 PM:

> MSVC can indeed produce a link map (/MAP or /MAP:filename to the
> linker.)

OK, I rebuilt the DLL with the MAP file.  I then did a checkout, update 
and commit, each of which crashed the JVM and produced a dump file.  I 
have posted a zip file with the 3 dumps and the MAP file.  Hopefully that 
will provide more clues.

The zip file is 138 Kb and you can get it here:

http://support.softlanding.com/download/javahl.zip

I am going to see if I can get Ubuntu setup to test this.

Thanks

Mark


_____________________________________________________________________________
Scanned for SoftLanding Systems, Inc. and SoftLanding Europe Plc by IBM Email Security Management Services powered by MessageLabs. 
_____________________________________________________________________________

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Joseph Galbraith <ga...@vandyke.com>.
Daniel Rall wrote:
> On Tue, 01 Nov 2005, Malcolm Rowe wrote:
> 
>> Well, the crash is a NULL pointer dereference at
>> 'libsvnjavahl-1.dll+0x2ba9', but that isn't much use unless we can
>> isolate the line that's crashing.
> 
> And the stack trace indicates that the crash is probably occurring in
> the checkout() method of org_tigris_subversion_javahl_SVNClient.cpp or
> (more likely) SVNClient.cpp, or -- likely native -- code which they
> call.
> 
> Earlier today I did look over the changes which have been made to the
> JavaHL native code between 1.2.x and 1.3.x, but nothing obvious jumped
> out at me.
> 
>> I think MSVC can produce MAP files that list the offsets of each function
>> within a DLL (or does it just do it for a single object file?  I can't
>> remember, I'm afraid).  It might even be possible to get it to produce a
>> mixed assembly/source file so that we can trace the specific line that's
>> failing, though that might only be available within the debugger, not
>> as a compiler output.
> ...
> 
> Something like this would be really useful.  Any MSVC gurus around?

MSVC can indeed produce a link map (/MAP or /MAP:filename to the
linker.)

It can also produce mixed assembly/source files (/FAs to
compiler.)

If you are building in the GUI (using DSP files), go to
project / settings / C++ tab / Listing Category.

The link map is on the Link tab, General Category,
"Generate mapfile"

(Sorry I don't have .net or .net 2003 installs to tell
you where they are there.)

Thanks,

Joseph


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Daniel Rall <dl...@collab.net>.
On Tue, 01 Nov 2005, Malcolm Rowe wrote:

> Well, the crash is a NULL pointer dereference at
> 'libsvnjavahl-1.dll+0x2ba9', but that isn't much use unless we can
> isolate the line that's crashing.

And the stack trace indicates that the crash is probably occurring in
the checkout() method of org_tigris_subversion_javahl_SVNClient.cpp or
(more likely) SVNClient.cpp, or -- likely native -- code which they
call.

Earlier today I did look over the changes which have been made to the
JavaHL native code between 1.2.x and 1.3.x, but nothing obvious jumped
out at me.

> I think MSVC can produce MAP files that list the offsets of each function
> within a DLL (or does it just do it for a single object file?  I can't
> remember, I'm afraid).  It might even be possible to get it to produce a
> mixed assembly/source file so that we can trace the specific line that's
> failing, though that might only be available within the debugger, not
> as a compiler output.
...

Something like this would be really useful.  Any MSVC gurus around?

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Possible Problem with JavaHL in 1.3.0 RC2

Posted by Malcolm Rowe <ma...@farside.org.uk>.
Well, the crash is a NULL pointer dereference at
'libsvnjavahl-1.dll+0x2ba9', but that isn't much use unless we can
isolate the line that's crashing.

I think MSVC can produce MAP files that list the offsets of each function
within a DLL (or does it just do it for a single object file?  I can't
remember, I'm afraid).  It might even be possible to get it to produce a
mixed assembly/source file so that we can trace the specific line that's
failing, though that might only be available within the debugger, not
as a compiler output.

Anyway, we need to do something like that to trace the problem to source,
I think.

Regards,
Malcolm

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org