nondeterministic RNG seed in tests makes troubleshooting impossible, should use C++ RNG everywhere

[evaleev]

srand(time()) alone is never a good idea in tests, as it prevents reproduction of bugs. Use of rand() is a bad idea anyway, as it results in antipatterns everywhere. Std C++ RNG is far from perfect.

My attempt to switch to std C++ RNG hints at more lurking bugs in test.cc, so this is likely a sister issue to #28

[evaleev]

@SpaceyLake please have a look into this. I'm not so sure the test.cc/TBLIS bindings are bug free. Use https://github.com/TAPPorg/reference-implementation/tree/31-nondeterministic-rng-seed-in-tests-makes-troubleshooting-impossible-should-use-c-rng-everywhere and enable C++ RNG (set use_cpp_rng constexpr to true) to be able to reproduce failures.

[SpaceyLake]

It's created from the change in randi. The generation of extents relied on that resetting srand with the same seed, yielding the same result. But I am happy revising it anyway. A lot of the contraction generation code could be improved upon.

[evaleev]

It's created from the change in randi. The generation of extents relied on that resetting srand with the same seed, yielding the same result. But I am happy revising it anyway. A lot of the contraction generation code could be improved upon.

Ah, I see, extents were random but matching for each slot occupied by the given index. This was not clear from the code. Thanks!! If you could revise this and we could switch to C++ we could have reproducible tests (to increase amount of randomness in our tests is best not by choosing seed randomly but by increasing the sample size with the fixed seed).