Deficiencies in the DRDT tool for C++
Hi there.
I think I found two deficiencies in the DRDT tool for C++.
(1) If there is a "race" on a line of complex expressions, you might get stuck because the tool does not state which variable is causing what it thinks is a race. If there is a way to get the variable name out of the tool, I did not find that way and would be glad if you could tell me.
(2) There are several circumstances under which you do not get any meaningful code view with C++ programs. I experience that for some template classes in my code, when i change to the "Race Source" tab, I only get several red lines with "Function: ..." giving me the full type expression of my class member function, but no race information.
I would provide you with screenshots and probably with parts of the code, if you could tell me how to do that in this forum.
Best regards,
Christian Terboven
[898 byte] By [
cterboven] at [2007-11-26 8:20:21]
# 1
Thank you for your comments, Christian.
(1) is a known limitation. We are working on providing more useful information to help users pin-down the data race.
I'd like to get more information on issue (2). Unfortunately, currently there is no easy way to attach files or screen-shots in the forum. Maybe you could host them somewhere else and post the url here? Or you can send them to me by email, and I can do the host-and-post so to share them with other community members on the forum.
Thanks!
-- Yuan
yuan at 2007-7-6 21:25:21 >
# 2
Hi Yuan.
I think (1) is a very very important issue, the variable name is definitely required. Yesterday we made the express compilers available to our users and mentioned the DRDT tool, I don't think it is suited for novice users if there isn't a big red blinking field ;-) with the variable name in it that is causing the race.
I have two screenshots for issue (2), you can get them from
http://support.rz.rwth-aachen.de/public/sun/issue2_01.jpg ("Races" tab) and
http://support.rz.rwth-aachen.de/public/sun/issue2_02.jpg ("Race Source" tab).
I probably could make the objects and the experiment results available for the community members (I doubt that there is lot of interest in that), but not the source code.
I could send you the source given that it is handled STRICTLY CONFIDENTIAL, or just ask Eric or Nawal. I had to run the instrumented version of DROPS (simplified dataset) for more than 48h on a 16gb machine with 8 threads to produce the experiment result and it did not finish - would you prefer to receive my experiment results with the source?
Best regards,
Christian Terboven
# 4
Hi Yuan.
I repeated the experiment with the compiler options "-xarch=v9b -library=stlport4 -xopenmp -g -xinstrument=datarace" at both compile time and the linking step using the "CC: Sun C++ 5.9 DEV 2006/06/08" compiler. Same results, for some of the races I do not get a race source.
Because I switched off optimization, the program was even slower and it did not finish, therefore it did not give me the same number of races as during the last run.
Best regards,
Christian
# 5
BTW: is there any known problem with the Forum software regarding email notifications of new posts? I checked the "Immediate" option in the Forum Settings dialog, but I never ever received any email with update notifications of this thread.Thanks,Christian
# 6
Well, I've seen similar problem befoe, but after I posted this error through feedback page (top left corner of the page, right under welcome message), it disappeared for me. Maybe it was introduced again?
# 8
Hi Yuan.
As requested, you can find the output of "er_print -rsummary all drdt.er" at
http://support.rz.rwth-aachen.de/public/sun/er_print.rsummary.txt
and further screenshots of the DRDT tool scrolled to the right at
http://support.rz.rwth-aachen.de/public/sun/issue02_03.jpg
http://support.rz.rwth-aachen.de/public/sun/issue02_04.jpg
http://support.rz.rwth-aachen.de/public/sun/issue02_05.jpg
http://support.rz.rwth-aachen.de/public/sun/issue02_06.jpg
http://support.rz.rwth-aachen.de/public/sun/issue02_07.jpg
http://support.rz.rwth-aachen.de/public/sun/issue02_08.jpg
Best regards,
Christian
