Test failures since switch to rand 0.8

[plugwash]

I am investigating upgrading the rust packaging in Debian to rand 0.8 and I have noticed that proptest seems to suffer from test failures since it switched to rand 0.8.

cargo test -p proptest --lib

failures:

---- num::test::contract_sanity::f32::range_from stdout ----
thread 'num::test::contract_sanity::f32::range_from' panicked at 'Uniform::new_inclusive: range overflow', /root/.cargo/registry/src/github.com-1ecc6299db9ec823/rand-0.8.4/src/distributions/uniform.rs:996:1
note: run with RUST_BACKTRACE=1 environment variable to display a backtrace

---- num::test::contract_sanity::f32::range_to_inclusive stdout ----
thread 'num::test::contract_sanity::f32::range_to_inclusive' panicked at 'Uniform::new_inclusive: range overflow', /root/.cargo/registry/src/github.com-1ecc6299db9ec823/rand-0.8.4/src/distributions/uniform.rs:996:1

---- num::test::contract_sanity::f64::range_from stdout ----
thread 'num::test::contract_sanity::f64::range_from' panicked at 'Uniform::new_inclusive: range overflow', /root/.cargo/registry/src/github.com-1ecc6299db9ec823/rand-0.8.4/src/distributions/uniform.rs:997:1

---- num::test::contract_sanity::f64::range_to_inclusive stdout ----
thread 'num::test::contract_sanity::f64::range_to_inclusive' panicked at 'Uniform::new_inclusive: range overflow', /root/.cargo/registry/src/github.com-1ecc6299db9ec823/rand-0.8.4/src/distributions/uniform.rs:997:1

failures:
num::test::contract_sanity::f32::range_from
num::test::contract_sanity::f32::range_to_inclusive
num::test::contract_sanity::f64::range_from
num::test::contract_sanity::f64::range_to_inclusive

test result: FAILED. 1421 passed; 4 failed; 0 ignored; 0 measured; 0 filtered out; finished in 783.81s

error: test failed, to rerun pass '-p proptest --lib'

I can reproduce the failures with commit commit bc2d867 (current HEAD) and commit 291908a (the first commit to use rand 0.8) and but not with 37dba3a (the last commit to use rand 0.7)

Ideally I'd like to see a fix for this issue, but if that is not possible, i'd at least like an opinion on how serious these failures are so I can decide whether they should block Debian's transition to rand 0.8.

[ghost]

It's 0.8.4 in particular. The reason seems to be this change: rust-random/rand#1108 .

[plugwash]

Hmm, yes thanks for the link, so it seems that the rand guys didn't consider a change from dodgy results to a panic to be a semver-breaking change.

That still leaves the question of how to address this? is this likely to break actual users of proptest? is there a way to disable/modify the tests only for floating point types?

[plugwash]

For what it's worth, after testing reverse dependencies I decided that these failures were not significant enough to block the rand transition in Debian and disabled the tests in question.

[russelldb]

This is very blocking for me. I get the error all the time. Does #261 fix?

[russelldb]

Looking at the history of Ran, Uniform::new panicked even in version 0.6…I don't understand how this can be attributed to the rand crate