3 ms·

Reading and learning about these is like peering in to a world of mystery. I know I'm looking at something interesting and important but it always sits just out

• pfdietz 1y ago
  The big question here is why is the approach so effective. Do random testing on a compiler that has never been subject to it and you will quickly find bugs, unless it's some extreme case like the compiler having been proved correct. Even CompCert, which had correctness guarantees, still had bugs that were outside the scope of those guarantees. Back before this was realized compilers were a buggy mess and people just lived with it. Now, they're still buggy, but less so. Random testing with billions of inputs is practical now and tends to rapidly "mine out" significant chunks of potential bug space. One could view this as an example of the Bitter Lesson, where dumb search coupled with massive amounts of computing power solves a problem that had seemed to require more knowledge. Anyone making a production quality compiler these days needs to be using high volume random testing.
  • anitil 1y ago
    > Anyone making a production quality compiler these days needs to be using high volume random testing How is this done these days, do people just set up CI workflows for this or have dedicated machines running tests 24/7?