cephes.stdtr(10, math.huge) = nan

[jucor]

It should be 1.
I thought we had checked for this kind of problems: did I miss something?

[jucor]

The cut-off on my machine is 1e155:

t7> return cephes.stdtr(10, 1e154)
1
t7> return cephes.stdtr(10, 1e155)
0.5
t7>

[d11]

Yikes... it's different for me, too:

return cephes.stdtr(10, 10e153)
1
return cephes.stdtr(10, 10e154)
0.5

On 2 Dec 2013, at 13:58, Julien Cornebise [email protected] wrote:

The cut-off on my machine is 1e155:

t7> return cephes.stdtr(10, 1e154)
1
t7> return cephes.stdtr(10, 1e155)
0.5
t7>
—
Reply to this email directly or view it on GitHub.

[jucor]

And we have the same cutoff: 10e153 = 1e154.

[d11]

Oh, oops! Sorry, I should've spotted that :-)

On 2 Dec 2013, at 15:43, Julien Cornebise [email protected] wrote:

And we have the same cutoff: 10e153 = 1e154.

—
Reply to this email directly or view it on GitHub.

[jucor]

:-) The weird thing is that Cephes does not even generate an error!

[d11]

Yeah. I guess this may qualify as a Cephes bug, then… interestingly, scipy doesn't seem to have the problem! I can't see what they're doing differently right away though...

On 2 Dec 2013, at 15:45, Julien Cornebise [email protected] wrote:

:-) The weird thing is that Cephes does not even generate an error!

—
Reply to this email directly or view it on GitHub.

[jucor]

That's exactly what I'm looking for right now…

On 2 Dec 2013, at 15:56, Dan Horgan [email protected] wrote:

Yeah. I guess this may qualify as a Cephes bug, then… interestingly, scipy doesn't seem to have the problem! I can't see what they're doing differently right away though...

On 2 Dec 2013, at 15:45, Julien Cornebise [email protected] wrote:

:-) The weird thing is that Cephes does not even generate an error!

—
Reply to this email directly or view it on GitHub.

—
Reply to this email directly or view it on GitHub.

[d11]

scipy/scipy@d4e3e60
possibly relevant? Just guessing though...

On 2 Dec 2013, at 15:57, Julien Cornebise [email protected] wrote:

That's exactly what I'm looking for right now…

On 2 Dec 2013, at 15:56, Dan Horgan [email protected] wrote:

Yeah. I guess this may qualify as a Cephes bug, then… interestingly, scipy doesn't seem to have the problem! I can't see what they're doing differently right away though...

On 2 Dec 2013, at 15:45, Julien Cornebise [email protected] wrote:

:-) The weird thing is that Cephes does not even generate an error!

—
Reply to this email directly or view it on GitHub.

—
Reply to this email directly or view it on GitHub.

—
Reply to this email directly or view it on GitHub.

[jucor]

Merging issue with #19 , which has unit tests to reproduce the problem.