On Thu, 13 Jun 2024 at 08:27, Alexandre Oliva <ol...@adacore.com> wrote:
>
>
> When !_GLIBCXX_USE_C99_MATH_TR1, binomial_distribution doesn't use the
> optimized algorithm that was fixed in response to PR114359. Without
> that optimized algorithm, operator() ends up looping very very long
> for the test, to the point that it would time out by several orders of
> magnitude, without even exercising the optimized algorithm that we're
> testing for regressions. Arrange for the test to be skipped if that
> bit won't be exercised.
>
> Regstrapped on x86_64-linux-gnu, also tested with gcc-13
> x-aarch64-rtems6, where the problem was detected. Ok to install?
OK.
This finding suggests we shouldn't even bother defining
binomial_distribution without the <cmath> functions. It doesn't seem
useful if it's that slow (or is that just because this test uses 1<<30
as the t parameter?)
Either way, the change to the test is fine.
>
>
> for libstdc++-v3/ChangeLog
>
> PR libstdc++/114359
> * testsuite/26_numerics/random/binomial_distribution/114359.cc:
> Require cmath.
> ---
> .../random/binomial_distribution/114359.cc | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
> diff --git
> a/libstdc++-v3/testsuite/26_numerics/random/binomial_distribution/114359.cc
> b/libstdc++-v3/testsuite/26_numerics/random/binomial_distribution/114359.cc
> index c1e4c380bf91d..12d967dcbfd34 100644
> ---
> a/libstdc++-v3/testsuite/26_numerics/random/binomial_distribution/114359.cc
> +++
> b/libstdc++-v3/testsuite/26_numerics/random/binomial_distribution/114359.cc
> @@ -2,6 +2,19 @@
>
> // Bug 114359 - std::binomial_distribution hangs in infinite loop
>
> +// { dg-require-cmath "" }
> +
> +// The requirement above is not strictly true. The test should work
> +// without cmath, and it probably does, but without cmath,
> +// binomial_distribution::operator() skips the optimized algorithm and
> +// calls _M_waiting to loop a gazillion times. On aarch64-rtems6
> +// qemu, that loop takes over 5 minutes to go through a small fraction
> +// of the iteration space (__x at 22k, limited at 1G; __sum at 2e-5,
> +// limited at 0.69). The bug we're regression-testing here was in the
> +// cmath-requiring bit, so even if this could conceivably not time out
> +// on a really fast machine, there's hardly any reason to exercise
> +// this extreme case.
> +
> #include <random>
>
> int main()
>
> --
> Alexandre Oliva, happy hacker https://FSFLA.org/blogs/lxo/
> Free Software Activist GNU Toolchain Engineer
> More tolerance and less prejudice are key for inclusion and diversity
> Excluding neuro-others for not behaving ""normal"" is *not* inclusive
>
