Exposed by the autotester on Linux32_64:
****** FAIL release64 std.random
core.exception.AssertError@std/random.d(3003): 99: 4699 < 4700.
Comment #1 by mxfomin — 2015-10-23T05:41:32Z
Can you point on example?
I see some pulls fail on FreeBSD with:
--- std.socket(502) test fails depending on environment ---
(core.exception.AssertError@std/socket.d(504): Assertion failure
Some (old) pulls fail on different platforms with:
****** FAIL release32 std.numeric
core.exception.AssertError@std/numeric.d(1616): unittest failure
Comment #2 by dlang-bugzilla — 2015-10-27T09:24:06Z
I've seen this a bunch of times. If you read the comment above the failing assert, essentially that entire check is luck-based. That test should be changed to use a RNG seeded with a constant value, or removed entirely.
Comment #3 by dlang-bugzilla — 2015-10-27T09:43:52Z
BTW, I'm not sure how intentional this is, but in a lot of places the std.random tests pass the RNG object by-value to e.g. randomSample, which means that it will query the same values over and over without advancing the RNG.
Comment #4 by dlang-bugzilla — 2015-10-27T09:45:03Z