<random>: Fix discrete_distribution result out of range

[statementreply]

Fix #1017.

Mathematically, _Par0._Pcdf.back() should be one. However, when it is actually slightly smaller than one due to rounding error, there is a small probability that _Px > _Par0._Pcdf.back(), and the original code returns the invalid value of _Par0._Pcdf.size().

[statementreply]

Question: do we add a (randomized?) test for low probability <random> bugs?

For this particular bug, the probability of returning an invalid result on each call is on the order of numeric_limits<double>::epsilon() * sqrt(N) for the average case and of numeric_limits<double>::epsilon() * N for the worst case, depending on distribution parameters and implementation details.

[CaseyCarter]

Related: Should we consider making _Pcdf an element shorter - leaving the final element as an implicit 1 - when we next break ABI?

[CaseyCarter]

Question: do we add a (randomized?) test for low probability <random> bugs?

For this particular bug, the probability of returning an invalid result on each call is on the order of numeric_limits<double>::epsilon() * sqrt(N) for the average case and of numeric_limits<double>::epsilon() * N for the worst case, depending on distribution parameters and implementation details.

It would be nice to have at least a simple regression test with the repro from the issue. I wouldn't refuse a randomized test, but it seems like that test wouldn't provide much additional marginal coverage. (If you do choose to write a randomized test it needs to dump enough info to reproduce errors when found a la:

STL/tests/std/tests/P0024R2_parallel_algorithms_set_intersection/test.cpp

Lines 19 to 28 in 550713e

	template <class T>
	void assert_message_vector(const bool b, const char* const msg, const T seedValue) {
	if (!b) {
	cerr << msg << " failed for seed value: " << seedValue << "\n";
	cerr << "This is a randomized test.\n";
	cerr << "DO NOT IGNORE/RERUN THIS FAILURE.\n";
	cerr << "You must report it to the STL maintainers.\n";
	abort();
	}
	}

[CaseyCarter]

Just a couple of style issues.

[StephanTLavavej]

Thanks! This looks great; I'll go ahead and push changes for a couple of conventions that we follow.

[CaseyCarter]

Thanks for the bugfix, @statementreply!