Problem with configure isfinite/finite detection

[cppunit-devel mailing list]

Compilation of CppUnit currently fails on both Solaris 10 (CC 5.8) & Suse
Linux 9.3 (g++ 3.3.5).

configure detects either isfinite() or finite(), but compilation fails when
attempting to use that function in CppUnit. Below are the relevant part of
configure/config-auto and compilation failure message. I don't know how
configure detects the presence of those function, but my guess is that it is
not done the same way as we compile CppUnit.

Linux man page state that isfinite() is only available in standard C99 mode
(which does not seem to be combinable with C++).
Solaris man page state that finite() requires include of <ieeefp.h>.
isfinite() is also present but I guess it is also only when compiling C (did
not manage to use it by changing the code).

Are the configure tests compiled in C or C++?

Here is the relevant data:
* Suse Linux 9.3 with g++ 3.3.5:
- configure:
checking for isfinite... yes
checking for finite... yes

- config-auto.h matches configure output:
 /* Define to 1 if you have the `finite' function. */
#ifndef CPPUNIT_HAVE_FINITE
#define CPPUNIT_HAVE_FINITE  1
#endif

/* define if compiler has isfinite */
#ifndef CPPUNIT_HAVE_ISFINITE
#define CPPUNIT_HAVE_ISFINITE  1
#endif

- but compilation fails:

g++ -DHAVE_CONFIG_H -I. -I../../../src/cppunit -I../../config -I../../include
 -                      I../../../include -g -O2 -MT
TestAssert.lo -MD -MP -MF .deps/TestAssert.Tpo -c .
./../../src/cppunit/TestAssert.cpp  -fPIC -DPIC -o .libs/TestAssert.o
In file included from ../../../src/cppunit/TestAssert.cpp:2:
../../../include/cppunit/portability/FloatingPoint.h: In function `bool
   CppUnit::floatingPointIsFinite(double)':
../../../include/cppunit/portability/FloatingPoint.h:38: error: `isfinite'
   undeclared (first use this function)
../../../include/cppunit/portability/FloatingPoint.h:38: error: (Each
   undeclared identifier is reported only once for each function it appears
   in.)

* Solaris 10 (CC 5.8):
- configure:
checking for isfinite... no
checking for finite... yes


- config-auto.h matches configure output:
/* Define to 1 if you have the `finite' function. */
#ifndef CPPUNIT_HAVE_FINITE
#define CPPUNIT_HAVE_FINITE  1
#endif

/* define if compiler has isfinite */
/* #undef CPPUNIT_HAVE_ISFINITE */

- but compilation fails:
CC -DHAVE_CONFIG_H -I. -I. -I../../config -I../../include -I../../include -pta
 -mt -xtarget=generic -g -features=no%transitions -xildoff -c
TestAssert.cpp -o TestAssert.o
"../../include/cppunit/portability/FloatingPoint.h", line 40: Error: The
function "finite" must have a prototype.
1 Error(s) detected.

Baptiste.
---
Baptiste Lepilleur <blep@...>
CppUnit maintainer
OpenTest and CppUnit 2 developer.


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Cppunit-devel mailing list
Cppunit-devel@...
https://lists.sourceforge.net/lists/listinfo/cppunit-devel

[cppunit-devel mailing list]

Hi Baptiste,

The configure test for isfinite() is performed by compiling with C++  
compiler and linking with -lm.  See config/ax_cxx_have_isfinite.m4 for  
the macro itself, and see config.log for the detailed output.

I'm very surprised that it is detected at configure time but the  
function is not defined at compile time.  Can you check carefully  
config.log and to see whether different compiler options might account  
for this?

The only other thing that comes to mind is that the real code is  
inside namespace CppUnit, whereas the configure test is not.  Do we  
need to qualify it to ::isfinite()?


Puzzled,
-Steve



-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Cppunit-devel mailing list
Cppunit-devel@...
https://lists.sourceforge.net/lists/listinfo/cppunit-devel

[mostekcm]

The problem is not that it doesn't find isfinite (it shouldn't on Solaris 10 from what I saw), but that configure does find finite when it should not find that either.  My fix for this was to comment out line 21700 in configure for cppunit-1.12.1:
//char $ac_func();

This line causes the configure test compilation to find finite() even though it is not defined in any Solaris 10 header I can find.  I've tried to understand the comment attempting to describe why that line is there, but I'm still confused as to what was trying to be solved.

Hope that helps,

Carlos

cppunit-devel mailing list wrote:

Hi Baptiste,

The configure test for isfinite() is performed by compiling with C++  
compiler and linking with -lm.  See config/ax_cxx_have_isfinite.m4 for  
the macro itself, and see config.log for the detailed output.

I'm very surprised that it is detected at configure time but the  
function is not defined at compile time.  Can you check carefully  
config.log and to see whether different compiler options might account  
for this?

The only other thing that comes to mind is that the real code is  
inside namespace CppUnit, whereas the configure test is not.  Do we  
need to qualify it to ::isfinite()?


Puzzled,
-Steve



-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Cppunit-devel mailing list
[e-mail]
https://lists.sourceforge.net/lists/listinfo/cppunit-devel

[JamesPiechota]

I've hit this issue on SuSE Linux 9.3. In my case the problem is due to
cmath replacing the isfinite function (see here for a discussion:
http://lists.apple.com/archives/Darwin-development/2002/Aug/msg00485.html).

Updating the include/cppunit/portability/FloatingPoint.h file to call
std::isfinite when CPPUNIT_HAVE_CMATH is defined worked for me. Here's the
diff:



--- FloatingPoint.h.orig        2008-04-11 09:56:04.000000000 -0700
+++ FloatingPoint.h     2008-04-11 09:52:27.000000000 -0700
@@ -2,7 +2,10 @@
 #define CPPUNIT_PORTABILITY_FLOATINGPOINT_H_INCLUDED

 #include <cppunit/Portability.h>
+
+#if not defined(CPPUNIT_HAVE_CMATH)
 #include <math.h>
+#endif

 CPPUNIT_NS_BEGIN

@@ -37,7 +40,9 @@
 /// @return \c true if x is neither a NaN, nor +inf, nor -inf, \c false
otherwise.
 inline int floatingPointIsFinite( double x )
 {
-#if defined(CPPUNIT_HAVE_ISFINITE)
+#if defined(CPPUNIT_HAVE_CMATH)
+   return std::isfinite( x );
+#elif defined(CPPUNIT_HAVE_ISFINITE)
    return isfinite( x );
 #elif defined(CPPUNIT_HAVE_FINITE)
    return finite( x );