Compiler Fuzzing With Prog-Fuzz Is Turning Up Bugs In GCC, Clang

[phoronix]

Phoronix: Compiler Fuzzing With Prog-Fuzz Is Turning Up Bugs In GCC, Clang

Vegard Nossum of Oracle has been working on fuzzing different open-source compilers for turning up bugs within these code compiler likes GCC and Clang...

http://www.phoronix.com/scan.php?page=news_item&px=Prog-Fuzz-Compiler-Fuzzing

[microcode]

Sweet, I'll put this to work against glslang, like I had with plain AFL before. It's always nice to see these tools become more accessible, it means that you can get results in the first few hours of a weekend, rather than spending a whole week to find out if it'll be useful.

Granted, if the tools uncover a lot, and you don't know the codebase you're fuzzing (i.e. me vs. Mesa), and you're the only person who cares about these bugs, sometimes it's not all that productive. With GLSL, there are little C-style compilers running on virtually every web-capable computer now, and they're compiling code from complete strangers all the time, without any particular notice to the user.

[carewolf]

Internal compiler eror on invalid code has a low priority in general though, and internal compiler errors on valid code is a much bigger problem (and extremely rare, I have only recently seen it on a few outdated clang versions), but is much harder to test with fuzzing.

[indepe]

I am not familiar with compiler fuzzing, but on reading this it seems that often fuzzing techniques are tried out to see what kind of bugs they willi find. Perhaps approaching it from the other direction would also be interesting: To look at a group of bugs that have already been discovered, but only with difficulties, and think of a fuzzing technique that will be able to find similar bugs.