<long double> issue

[Bob Carpenter]

Anyone know what's up with this?

"Exception thrown at line 82: Error in function boost::math::lgamma<long double>(long double): Evaluation of lgamma at 0."

It was a message reported by Swupnil Sahai on stan-users.
I followed up, and he says he has a 2014 Macbook Pro, so I
have no idea what a long double is doing in there.

If nobody knows, I'll try to recreate the report on my machine
and file an issue.

- Bob

[Daniel Lee]

No idea. My guess is that we didn't check inputs correctly and let 0 pass through to lgamma.

If you can set up the test case, we can track that down pretty quickly.


Daniel

> --
> You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Bob Carpenter]

Sorry, my issue was with how the type ***long double*** got into
the mix. We shouldn't be using that anywhere. And given recent
speed reports, I'm wondering if this is somehow negatively impacting
the speed of Stan code.

- Bob

[Daniel Lee]

I have no idea how we have the "long double" type.

If you know how to recreate Swupnil's example, that'll be a good start. (normally I'd volunteer, but I'm feeling a little behind now.)

Daniel

[Ben Goodrich]

On Thursday, May 5, 2016 at 12:10:02 PM UTC-4, Bob Carpenter wrote:

Sorry, my issue was with how the type ***long double*** got into
the mix.  We shouldn't be using that anywhere.  And given recent
speed reports, I'm wondering if this is somehow negatively impacting
the speed of Stan code.

Macs have been using libc++ as their C++ library for a while, regardless of whether C++11 is specified as the standard. So, it may just be doing what it would do under C++11 where a long double is well-defined.

Ben

[Bob Carpenter]

I created an issue. But the way things are going, I won't
have time to look into it this year.

- Bob