Post subject: fuzz testing for finding glitches
Player (81)
Joined: 3/11/2005
Posts: 352
Location: Oregon
Zefiris recently asked for a gmv for testing his movie editor, and I generated a one with output from /dev/urandom as the frame data. I had heard of fuzz testing, so this got me curious and I tried playing it from a state I had from Pulseman. Within a very short time, it managed to jump through a diagonal in a way I hadn't thought possible. (gif) From there, I hacked up Gens with a configurable fuzzer. After much funky dancing, the fuzzer found the first half of the movement glitch (no volteccering/disappearing) and a new glitch that lets me jump about 16 pixels higher than normally possible. Does anyone else use this or have thoughts on it?
ideamagnate| .seen aqfaq <nothing happens> DK64_MASTER| .seen nesvideoagent * DK64_MASTER slaps forehead
Post subject: Re: fuzz testing for finding glitches
Emulator Coder, Skilled player (1310)
Joined: 12/21/2004
Posts: 2687
IdeaMagnate wrote:
Does anyone else use this or have thoughts on it?
Yes and yes. See this thread, for example. It might be a little less random than what you're talking about, but it seems to be the same concept. Your "configurable fuzzer" sounds interesting and useful, depending on how configurable it is...
Player (81)
Joined: 3/11/2005
Posts: 352
Location: Oregon
The fuzzer just randomly generates input given a probability for each button for a certain number of frames. It's very easy to implement. I'm glad you posted that. I was aware of the thread, but I had originally just intended the fuzzer for finding general bugs. It seems to work well as input generation for a robot too.
ideamagnate| .seen aqfaq <nothing happens> DK64_MASTER| .seen nesvideoagent * DK64_MASTER slaps forehead