Domain error with tgammal when using FreeBSD

[stephen]

I tested sage-5.9.beta2 and got errors like this:

"sage -t devel/sage/sage/plot/plot.py"

**********************************************************************
File "devel/sage/sage/plot/plot.py", line 942, in sage.plot.plot.plot
Failed example:
    plot(gamma, (-3, 4), detect_poles = True).show(ymin = -5, ymax = 5)
Expected nothing
Got:
    <BLANKLINE>
    tgammal domain error
**********************************************************************
File "devel/sage/sage/plot/plot.py", line 946, in sage.plot.plot.plot
Failed example:
    plot(gamma, (-3, 4), detect_poles = 'show').show(ymin = -5, ymax = 5)
Expected nothing
Got:
    <BLANKLINE>
    tgammal domain error

This is because FreeBSD uses the cephes math libraries. These output error messages to stdout when they shouldn't.

The following patch removes all the print statements from files called mtherror.c in cephes.

--- cephes-2.8.p1/spkg-install-orig     2013-03-30 20:48:35.000000000 +0000
+++ cephes-2.8.p1/spkg-install  2013-03-30 20:48:57.000000000 +0000
@@ -24,6 +24,8 @@
     fi
 done

+sed -i .orig 's|.*printf|//&1|' */mtherr.c
+
 echo "Building Cephes..."
 $MAKE
 if [ $? -ne 0 ]; then

[kcrisman]

I like this idea, but I'm not sure whether this is what we should do. It might be better to use a sed script as above to make the changes, then to create a patch we can apply based on that.

[stephen]

I don't mind using an explicit patch. I was just trying to indicate where the error took place.

In some sense it is a bandaid fix. But then the whole cephes library is a bandaid fix.

[kcrisman]

Great, perhaps I will get around to making this patch sometime. You are right about cephes, I suppose, but numerical issues are somewhat beyond my ken.

[jdemeyer]

Replying to kcrisman:

It might be better to use a sed script as above to make the changes, then to create a patch we can apply based on that.

I agree. Document the sed script in SPKG.txt, but apply a patch in spkg-install.