On Thu, Aug 5, 2010 at 3:43 AM, Cristian Cadar <c.cadar at imperial.ac.uk> wrote: > > Hi, it would be indeed useful to set up a bigger regression suite which > would make sure that KLEE still finds the bugs in Coreutils. ?So it would be > great if anybody would offer to do this.
Yes, I completely agree (but don't personally have any spare time). In slightly related news, the KLEE buildbot is starting to actually work, so if someone wanted to work on this we have a clear place to integrate it. - Daniel > However, to reproduce the bugs, I would recommend trying the KLEE and LLVM > versions from the time of the thread mentioned by Brent > (http://keeda.stanford.edu/pipermail/klee-dev/2009-October/000137.html), as > it looks like at that time KLEE was still able to find the bugs for mkdir > and mkfifo, which the current version of KLEE does not find. ?It would be > also very useful to know what's causing the problem now. > > Best, > Cristian > > On 05/08/10 07:01, Daniel Dunbar wrote: >> >> Hi Brent, >> >> On Wed, Aug 4, 2010 at 11:00 AM, Brent Lim<brentlim at brentlim.net> ?wrote: >>> >>> I have been trying to reproduce the bugs in Coreutils 6.10 mentioned in >>> the >>> 2008 OSDI paper, ?KLEE: Unassisted and Automatic Generation of >>> High-Coverage >>> Tests?, but have not been successful. >>> >>> >>> >>> I have looked at the old thread, >>> http://keeda.stanford.edu/pipermail/klee-dev/2009-October/000139.html and >>> tried the options there, but that did not work for me. Specifically, >>> mkdir >>> and mkfifo caused a segfault. >> >> Sounds like we introduced a bug at some point. You might consider >> trying with llvm-2.6, as that was closer to what we used at the time, >> but of course the KLEE code has changed substantially. Unfortunately, >> we don't have any good regression testing to check that we still find >> the same bugs we used to... :/ >> >>> I am using the latest version of KLEE (version 110223), compiled with >>> llvm-gcc-2.7-front-end, with the following options, ./configure >>> ?with-llvm=/path/to/llvm-2.7 ?with-uclibc=/path/touclibc/ >>> --enable-posix-runtime. I have built the coreutils with llvm-gcc-2.7 with >>> the following options, ./configure ?disable-nls ?CFLAGS=-g >>> >>> >>> >>> I guess I am doing something wrong here, but I have no idea what it is. >>> This >>> brings me to the 2 questions that I have: >>> >>> 1) ? ? ?What svn version of KLEE should I use to reproduce the bugs? >> >> None, really -- the KLEE we used in the paper was before the open >> source release. >> >>> 2) ? ? ?In the thread, there was an option >>> ?exclude-cov-file=./../lib/functions.txt. May I know what does this >>> option >>> do and may I have that file? (functions.txt) >> >> --excude-cov was an option to not count the coverage of specified >> functions (in the coreutils case, we used the functions that are in >> the library not the frontend tools). I never really liked it, and we >> never ran a good experiment to even see if it was a good idea, so I >> pulled it out at some point. >> >> ?- Daniel >> >>> >>> >>> >>> >>> Regards, >>> >>> Brent Lim >>> >>> _______________________________________________ >>> klee-dev mailing list >>> klee-dev at keeda.stanford.edu >>> http://keeda.Stanford.EDU/mailman/listinfo/klee-dev >>> >>> >> _______________________________________________ >> klee-dev mailing list >> klee-dev at keeda.stanford.edu >> http://keeda.Stanford.EDU/mailman/listinfo/klee-dev >
