Avoid rand() in the legacy tr1 test suite by StephanTLavavej · Pull Request #4921 · microsoft/STL (original) (raw)

Our bosses and boss-like entities heard that rand() is considered harmful and are starting a push to avoid unnecessary use.

Almost all of our mentions of rand are Standard-mandated (or Standard stable name citations), but I found a few unnecessary uses in our legacy tr1 test suite. We avoid cleaning up its code unless there's a specific reason to mess with it, and this easily qualifies.

• tr1/tests/algorithm
  • Give rand_gen an mt19937 data member. (Default construction is fine; we weren't attempting to seed rand() before.) This is purely masking bits with & 0xfffff, so there's no need for a uniform_int_distribution.
  • Replace the frand() wrapper with an rng_func lambda. To avoid /clr:pure problems with init-captures, I'm capturing mt by reference (so the lambda doesn't need to be mutable and doesn't need to be passed via ref()). We need to adapt the half-open interface to uniform_int_distribution's fully-closed range.
• tr1/tests/memory2
  • Construct a local mt19937 mt; within the static void ctors() worker function. (It's important to not use a global engine with multiple threads!) Again, we weren't attempting to seed rand(), so default-constructing mt19937 is fine. And because we're just looking at a bit, we don't need a uniform_int_distribution.