Experimental Physics and
Industrial Control System

Andrew Johnson <[email protected]> · Wed, 27 Apr 2016 10:28:59 -0500

On 04/27/2016 10:06 AM, Michael Davidsaver wrote:
> So strictly speaking Base should never have depended on this behavior. 
> In fact beginning with Base 3.15.0.1, and soon to be included in
> 3.14.12.5, it does not.
> 
> http://bazaar.launchpad.net/~epics-core/epics-base/3.14/revision/12486
> 
> Given that newer versions of Base will build, and that this change has
> already found its way into an Ubuntu LTS release, I don't think it's
> worth pursuing another change to glibc.

According to the Linux manpage, finite() was originally a BSD
floating-point classification function which takes a double. There were
equivalents available for float and long double with different names.
All the epicsMath.h code is trying to do is ensure that we provide a
definition of finite(), using isfinite() where it exists. I see no
reason to check GLIBC versions, we don't need it to be that efficient.

I am committing this change to the 3.14 branch, so it will appear in the
forthcoming 3.14.12.6 release:

 #ifdef isfinite
 #  undef finite
-#  define finite(x) isfinite(x)
+#  define finite(x) isfinite((double)x)
 #endif

- Andrew

-- 
There are only two hard problems in distributed systems:
  2. Exactly-once delivery
  1. Guaranteed order of messages
  2. Exactly-once delivery
 -- Mathias Verraes

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System