Skip to main content
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [cdt-debug-dev] Thread mix-up when debuggingmulti-threaded programs 
Hi Mikhail,
Here are some questions and findings regarding the thread mix-up problem when debugging multi-threaded programs:
A question; if you use CDT 3.0.2 with Eclipse 3.2, will the new asynchronous org.eclipse.debug framework be used or will the old synchronous one be used instead?
I have noticed that when you select a new thread in the Debug view, there is no MI thread-select command issued in the GDB console. Is this wrong or will CDT change to the new thread as soon as a step operation or similar is issued on it?
I have noticed that org.eclipse.debug.internal.ui.viewers.update.ThreadEventHandler.handleResume(DebugEvent) will remove the thread corresponding to the resume event from its list of suspended threads and then take the next thread in its list of suspended threads and call fireDeltaUpdatingTopFrame() on it, which will make the topmost frame of that thread selected if it has one. However, handleResume(DebugEvent) is not called for all threads, which will make fireDeltaUpdatingTopFrame() change the selection to some other thread's topmost frame when singlestepping a newly selected thread. This seems wrong to me. In many systems, all threads in the program are suspended when a breakpoint is hit or a singlestep is completed, and all threads are resumed during the course of a singlestep operation. Are some resume events lost on the way from CDT to org.eclipse.debug or doesn't the new framework support this kind of systems?
I have also noticed that org.eclipse.debug.internal.ui.viewers.update.EventHandlerModelProxy.dispatchResume() will only call ThreadEventHandler.handleResume(DebugEvent) if the event is not a "step start" event, in which case it will start a one-shot PendingSuspendTask timer that will trigger in 500 ms and then call ThreadEventHandler.handleSuspendTimeout(). However, if it does not timeout, nothing seems to be done. That seems to explain why this problem doesn't show up when debugging org.eclipse.debug and CDT and singlestepping the event handling code. Also, if I change the PendingSuspendTask timer to trigger after 1 ms the problem shows up very rarely (however, it has showed up two times...).
This is as far as I can get with this problem. Debugging the new asynchronous org.eclipse.debug framework is extremely difficult. I hope my ramblings are useful for any further investigations.
There are several bug reports on the new asynchronous org.eclipse.debug framework. Is it wise for CDT 3.1 to switch to it at this point or would we have a more stable CDT 3.1 debug environment by using the old synchronous model? 

Kind regards,
Stefan


Mikhail Khodjaiants wrote:


Hi Stefan,

Thank you for looking at it. I'll check why this problem doesn't occur
with CDT 3.0.2, I don't recall if we have made any serious changes in
this area. And it is very difficult to debug it - the new asynchronous
framework for the views have many layers and runs all request in
separate jobs.

Best regards,
Mikhail

-----Original Message-----
From: cdt-debug-dev-bounces@xxxxxxxxxxx
[mailto:cdt-debug-dev-bounces@xxxxxxxxxxx] On Behalf Of Stefan Bylund
Sent: 30 August 2006 13:51
To: CDT Debug developers list
Subject: Re: [cdt-debug-dev] Thread mix-up when debuggingmulti-threaded
programs

Hi Mikhail,

I noticed that several thread-related bugs has been fixed in
org.eclipse.debug in the Eclipse 3.2.1 maintenance branch. So I tried
the M20060828-1036 build of Eclipse 3.2.1 but unfortunately this problem
still exists. I also tried using CDT 3.0.2 with Eclipse
3.2.1-M20060828-1036 and now it works, i.e. the thread switching problem
is gone! This leads me to believe that this bug is really in CDT 3.1 and
not in org.eclipse.debug in Eclipse 3.2. A colleague of mine is
debugging CDT to try to find this problem but without luck so for...

Any help or input on this problem would be much appreciated.

Kind regards,
Stefan


Mikhail Khodjaiants wrote:


Hi Stefan,
I have spent some time trying to figure out why this is happening. The CDT sets the current thread correctly, but it seems that the Debug view
remembers the previous expansion state and restores it replacing the correct one. The cause is hidden deep in the new implementation of the Debug platform views, I haven't found the solution yet. 

Regards,
Mikhail Khodjaiants
ARM Ltd.

-----Original Message-----
From: cdt-debug-dev-bounces@xxxxxxxxxxx [mailto:cdt-debug-dev-bounces@xxxxxxxxxxx] On Behalf Of Stefan Bylund 
Sent: 25 August 2006 21:38
To: cdt-debug-dev@xxxxxxxxxxx
Subject: [cdt-debug-dev] Thread mix-up when debugging multi-threaded programs 

Hi,

I recently switched from Eclipse 3.1.2 / CDT 3.0.2 / GDB 6.3 to Eclipse
3.2 / CDT 3.1 / GDB 6.3 and discovered a very annoying bug in the CDT Debug view when debugging multi-threaded programs. When you select the topmost frame in another thread in the Debug view, that frame will be selected. However, when trying to singlestep the newly selected thread, 


the topmost frame of the previous thread is erroneously selected again.
By selecting, the topmost frame of the intended thread again, you will notice that the singlestep operation really happened and you can from here on singlestep in the new thread without being switched back. However, when selecting a new thread again, the same problem happens again and you have to reselect the new thread to be able to continue. 

There seems to be some inconstistency about which thread is the current
between the Debug view and the CDT debug model or between different parts of the CDT debug model. This bug did not exist in Eclipse 3.1.2 / 


CDT 3.0.2.
I have searched the Eclipse bug database for any bugs about this and found the bug 151253 "Singlestepping a multi-thread program to fast results in unexpected behavior" which I thought could be related to this problem. Bug 151253 has been fixed in the CDT 3.1.1 branch so I downloaded the I200608210800 nightly build of CDT 3.1.1 but unfortunately the thread switching problem problem still exists.
Is this a known problem with CDT 3.1? Otherwise, I can file a bug report. 

Regards,
Stefan


_______________________________________________
cdt-debug-dev mailing list
cdt-debug-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cdt-debug-dev

-- IMPORTANT NOTICE: The contents of this email and any attachments are

confidential and may also be privileged. If you are not the intended
recipient, please notify the sender immediately and do not disclose the
contents to any other person, use it for any purpose, or store or copy
the information in any medium.  Thank you.

_______________________________________________
cdt-debug-dev mailing list
cdt-debug-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cdt-debug-dev




_______________________________________________
cdt-debug-dev mailing list
cdt-debug-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cdt-debug-dev
_______________________________________________
cdt-debug-dev mailing list
cdt-debug-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cdt-debug-dev

Back to the top