yarpgen VS csmith

One also may bias values to be similar (or equal) to other recently generated values. This was a trick used in the compiler tester yarpgen, where it was useful in testing optimization of bitwise operators on integers.

https://github.com/intel/yarpgen

https://github.com/intel/yarpgen/blob/main/papers/yarpgen-oo...

(see "Policies for constants", page 196:8.)

This is really impressive! A link to the github if anyone wants to check out the code: https://github.com/intel/yarpgen/blob/main/README.md

Here's a list of bugs found by Yarpgen: https://github.com/intel/yarpgen/blob/main/bugs.rst

Yes! Another well-known program reducer is C-Reduce [0]. When Csmith [1] started churning out very large programs that exhibited errors in C compilers, the compiler maintainers asked the researchers to please reduce the ~81KB files to a more manageable size so they could understand the errors better. C-Reduce was developed specifically to address that need.

[0] https://github.com/csmith-project/creduce

[1] https://github.com/csmith-project/csmith

Is CompCert actually safer in practice? One way to evaluate this is via fuzzing tools like CSmith. CSmith has a list of bugs they have found (hundreds total, in mainstream compilers like llvm and gcc). Here is a quote from the 2011 PLDI paper on CSmith:

If you are actually just going to use C, the easiest way is to just use this existing tool that does exactly what you want and has been used to find many, many existing bugs in many compilers: https://embed.cs.utah.edu/csmith/

Here's a list of bugs found by a similar project, Csmith: https://github.com/csmith-project/csmith/blob/master/BUGS_REPORTED.TXT