[libcxx] [PATCH] http://llvm.org/bugs/show_bug.cgi?id=18767

[Arthur O'Dwyer]

I posted on the Bugzilla.

Basically, Howard's proposed patch anticipates that DR2168 will be
resolved in some way that still allows
uniform_real_distribution(x,x)() to have undefined behavior.
http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n3893.html#2168

I have no inside information, but naively it seems obvious to me that
DR2168 is more likely to be resolved by making libc++'s *current*
implementation conforming (by fixing the <-for-≤ typo), at which point
no patch will be necessary at all.

–Arthur

On Tue, Feb 11, 2014 at 12:34 PM, Howard Hinnant <hhinnant at apple.com> wrote:
> http://llvm.org/bugs/show_bug.cgi?id=18767
>
> This is a fix & test for this bug in uniform_real_distribution.  I believe the bug is simply due to roundoff error, and the way to fix it is to iterate until the roundoff does not happen.
>
>
> In the test case, which is an extreme, there are about 2.6 iterations / call.  In a more realistic scenario, such as defaulting the range to [0, 1), there is only 1 iteration / call (as before this fix).
>
> Howard
>
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>