Evaluate results not only for errors, but also for mismatch

[mvorisek]

At first, I would you like to thank you for this fuzzing app. It found hundreds of bugs already.

I have an idea. What about running any fuzzing code twice against different php configuration, like JIT vs. non JIT?

Excluding functions that can introduce randomness, the results must be always exactly the same.

[YuanchengJiang]

Hi,

That is indeed a really good idea. Actually, I have tried this and have found some bugs (e.g., php/php-src#16473). I will be releasing this differential testing tool soon.

Instead of excluding functions that bring randomness, I try repetitive executions to exclude unstable results.