Quoting Peter Finderup Lund <[EMAIL PROTECTED]>: > > The only problem is that crashes cannot be exactly reproduced, but > if > > why not? the random data can be sent both to the x/xfs server and to > a > file. man tee. there might still be timing differences, of course...
Yes, silly me. Just saving the data sent in the current connection to a file should help a lot, I did not bother about it because at every crash I was able to understand what was wrong by just inspecting the server state and the stack (sometimes it overwritten due to buffer overflows, but some investigation like casting stack addresses to a struct pointer, etc, allowed understanding from where it was overwritten). But besides timing differences, it is also possible that an earlier connection do not crash the server, but changes the server internal state causing a crash in a later connection. Saving a few hours of almost continuous /dev/urandom will needs a large amount of storage :-) This sounds like an interesting project for a testing tool/environment. > Other people have done similar things -- it seems to be almost unused > but > very useful (and easy!): > > http://pages.prodigy.net/dave_christian/gwu/cs256/10_Random%20Testing.ppt > http://216.239.39.100/search?q=cache:FI6soE4R8qUC:pages.prodigy.net/dave_christian/gwu/cs256/10_Random%2520Testing.ppt+crash+testing+random+unix&hl=en&ie=UTF-8 > (same in HTML format) > > http://www.cs.wisc.edu/~smethegj/fuzz-2001.pdf > > http://www.usenix.org/events/usenix-win2000/forrester.html > > http://citeseer.nj.nec.com/koopman97comparing.html > "Comparing Operating Systems Using Robustness Benchmarks" > > -Peter Paulo _______________________________________________ XFree86 mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/xfree86
