RE: isinf and type-pun warning/error on OSF guile 1.8.7

[Jay K]

Andy, your fix quashed the warning for type punning.
memcpy does too.
Your way seems to be the recommended way.
 I lost the link that seems to recommend your way.


isinf is different than I though, it seems to be some gcc builtin.


% cat 4.c
int main(int argc, char** argv)
{
 return isinf((double)argc);
}

% cc 4.c
ld:
Unresolved:
isinf
% gcc 4.c
 => success

http://gcc.gnu.org/onlinedocs/gcc-4.5.0/gcc/Other-Builtins.html#Other-Builtins

% gcc -fno-builtin 4.c
isinf
collect2: ld returned 1 exit status


I don't know what to recommend here for portability and autoconfing.



% gcc -ansi 4.c
isinf
collect2: ld returned 1 exit status

Maybe configure for __builtin_isinf?

bash-4.1$ cat 4.c
int main(int argc, char** argv)
{
 return isinf((double)argc);
}

bash-4.1$ cat 5.c
int main(int argc, char** argv)
{
 return __builtin_isinf((double)argc);
}

bash-4.1$ gcc 4.c  -Wall
4.c: In function 'main':
4.c:3:2: warning: implicit declaration of function 'isinf'
bash-4.1$ gcc 5.c  -Wall
 => success

http://www.gnu.org/software/hello/manual/autoconf/Function-Portability.html

recommend wither gnulib or a little other workaround shown.


Here they just prototype it:
http://mail-index.netbsd.org/pkgsrc-bugs/2007/05/18/0000.html

I'm tempted to say it is a gcc bug.

Thanks,
 - Jay


----------------------------------------
> From: address@hidden
> To: address@hidden
> CC: address@hidden
> Subject: RE: isinf and type-pun warning/error on OSF guile 1.8.7
> Date: Wed, 9 Jun 2010 23:16:28 +0000
>
>
> Andy, I will try it later.
>
> Often what autoconf does is:
>
> echo "int main() { isinf(0); }"> conftest.c
> gcc conftest.c
> => if it links successfully, it is there.
>
>
> and then later on the "real" code uses gcc -Wmissing-prototypes -Werror, boom.
> That's what I saw months/year+ ago on Irix at least.
> I think in the Irix case the protype was under an #if.
> I'll look around more on Tru64. Might get the admin to install newer 
> compiler/headers too.
>
>
>
> You know, every so often I get up the gumption to try gcc/make check, which 
> requires a whole bunch more stuff than just gcc/make ... bdwgc, guile, 
> libunistring, autogen, tcl, dejagnu, expect, pkg-config, ffi....kind of 
> annoying, oh well. It looks like I succeeded this time, phew..
>
>
> Thanks, later,
> - Jay
>
>
>
> ----------------------------------------
>> From: address@hidden
>> To: address@hidden
>> CC: address@hidden
>> Subject: Re: isinf and type-pun warning/error on OSF guile 1.8.7
>> Date: Thu, 10 Jun 2010 00:28:00 +0200
>>
>> Hi Jay,
>>
>> On Wed 09 Jun 2010 17:43, Jay K writes:
>>
>>> libtool: compile: gcc -DHAVE_CONFIG_H -I.. -I/home/jayk/src/guile-1.8.7 
>>> -I.. -mieee -mieee -D_REENTRANT -pthread -g -O2 -Wall -Wmissing-prototypes 
>>> -Werror -MT libguile_la-numbers.lo -MD -MP -MF 
>>> .deps/libguile_la-numbers.Tpo -c 
>>> /home/jayk/src/guile-1.8.7/libguile/numbers.c -DPIC -o 
>>> .libs/libguile_la-numbers.o
>>> cc1: warnings being treated as errors
>>> /home/jayk/src/guile-1.8.7/libguile/numbers.c: In function 'xisinf':
>>> /home/jayk/src/guile-1.8.7/libguile/numbers.c:144: error: implicit
>>> declaration of function 'isinf'
>>
>>> I couldn't find isinf in any header.
>>
>> This call only occurs if configure detected support `isinf', so surely
>> it is there? Can you check again? Barring that can you send a
>> config.log, please.
>>
>>> /home/jayk/src/guile-1.8.7/libguile/numbers.c: In function 
>>> 'guile_ieee_init':
>>> /home/jayk/src/guile-1.8.7/libguile/numbers.c:623: error: dereferencing 
>>> type-punned pointer will break strict-aliasing rules
>>> /home/jayk/src/guile-1.8.7/libguile/numbers.c:654: error: dereferencing 
>>> type-punned pointer will break strict-aliasing rules
>>
>> Does replacing that DINIFINITY block with the following help?
>>
>> /* OSF */
>> extern unsigned int DINFINITY[2];
>> union
>> {
>> double d;
>> int i[2];
>> } alias;
>> alias.i[0] = DINFINITY[0];
>> alias.i[1] = DINFINITY[1];
>> guile_Inf = alias.d;
>>
>> Likewise for DQNAN:
>>
>> {
>> /* OSF */
>> extern unsigned int DQNAN[2];
>> union
>> {
>> double d;
>> int i[2];
>> } alias;
>> alias.i[0] = DQNAN[0];
>> alias.i[1] = DQNAN[1];
>> guile_NaN = alias.d;
>> }
>>
>>> I recall seeing the same problem on Irix, where autoconf does a link check
>>> and compiles without -Wmissing-prototype so it passes, because the function
>>> does exist somewhere. Autoconf checks need to more closely resemble
>>> how later compilation will occur.
>>
>> How does that work though -- does isinf have no header/declaration?
>>
>>> /home/jayk/src/guile-1.8.7/test-suite/standalone/test-conversion.c:859:
>>> error: dereferencing type-punned pointer will break strict-aliasing
>>> rules make[4]: *** [test_conversion-test-conversion.o] Error 1
>>
>> We can copy the above solution if it works. Let us know!
>>
>> Andy
>> --
>> http://wingolog.org/