gamma(::BigFloat) overflows on Windows

DNF · April 28, 2021, 1:53pm

jl> p = big(3.6e13)
3.6e+13

jl> q = 1/big(0.563483398)+1
2.774675178628776489804929213602311938188089959295862682297487944939957397047059

jl> beta(p, q)
NaN

??

DNF · April 28, 2021, 1:55pm

On my computer:

jl> gamma(big(3.6e13))
Inf

jl> @which gamma(big(3.6e13))
gamma(x::BigFloat) in SpecialFunctions at C:\Users\dnf\.julia\packages\SpecialFunctions\mFAQ4\src\gamma.jl:578

mbauman · April 28, 2021, 1:58pm

Hunh. I’m also Julia 1.6, SpecialFunctions 1.3.0. How did you install Julia? On what platform?

DNF · April 28, 2021, 2:11pm

I have two different installations: Julia 1.6.0 on Windows 10, and version 1.6.1 on WSL2 under Windows 10 (it’s on the same system).

The ordinary Windows installation:

jl> gamma(big(3.6e13))
Inf

while the one under WSL2:

jl> gamma(big(3.6e13))
7.882861104028365451673058070377064242970022825665897715265241972922672893761775e+472392288679098

mbauman · April 28, 2021, 2:16pm

I’ve split this out — would you be willing to file a bug report? You seem uniquely well placed to look at precisely the differences that are causing this.

DNF · April 28, 2021, 2:37pm

Yes, I can do that later today.

DNF · April 28, 2021, 7:51pm

https://github.com/JuliaMath/SpecialFunctions.jl/issues/309

bashonubuntu · April 29, 2021, 2:36pm

Hey @DNF and @mbauman, thanks for taking so much time to reply to the other thread.

When I opened the other post, I hadn’t foreseen it could help in uncovering a related issue.

I’ll be watching the github issue closely – thanks again for taking out so much time for this