Re: missing isinf declaration on solaris

From: Oskari Saarenmaa <os(at)ohmu(dot)fi>
To: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: pgsql-hackers(at)postgresql(dot)org
Subject: Re: missing isinf declaration on solaris
Date: 2014-09-24 13:11:50
Message-ID: 5422C316.9080901@ohmu.fi
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

24.09.2014, 15:25, Tom Lane kirjoitti:
> Oskari Saarenmaa <os(at)ohmu(dot)fi> writes:
>> GCC 4.9 build on Solaris 10 shows these warnings about isinf:
>> float.c: In function 'is_infinite':
>> float.c:178:2: warning: implicit declaration of function 'isinf'
>
> Ugh.
>
>> isinf declaration is in <iso/math_c99.h> which is included by <math.h>,
>> but it's surrounded by #if defined(_STDC_C99) || _XOPEN_SOURCE - 0 >=
>> 600 || defined(__C99FEATURES__). A couple of quick Google searches
>> suggests that some other projects have worked around this by always
>> defining __C99FEATURES__ even if compiling in C89 mode. __C99FEATURES__
>> is only used by math.h and fenv.h in /usr/include.
>
>> Should we just add -D__C99FEATURES__ to CPPFLAGS in
>> src/template/solaris, add our own declaration of isinf() or do something
>> else about the warning?
>
> I'm worried that __C99FEATURES__ might do other, not-so-C89-compatible
> things in later Solaris releases. Possibly that risk could be addressed
> by having src/template/solaris make an OS version check before adding the
> switch, but it'd be a bit painful probably.
>
> Based on the #if you show, I'd be more inclined to think about defining
> _XOPEN_SOURCE to get the result. There is precedent for that in
> src/template/hpux which does
> CPPFLAGS="$CPPFLAGS -D_XOPEN_SOURCE_EXTENDED"
> I forget what-all _XOPEN_SOURCE_EXTENDED enables exactly, but if memory
> serves there were a nontrivial number of now-considered-standard features
> turned on by that in ancient HPUX releases. If you want to pursue this
> route, you'd need to do a bit of digging to see what else _XOPEN_SOURCE
> controls in Solaris and if there is some front-end feature macro (like
> _XOPEN_SOURCE_EXTENDED) that you're supposed to set instead of touching
> it directly.

Looking at standards(5) and /usr/include/sys/feature_tests.h it looks
like _XOPEN_SOURCE_EXTENDED enables XPG4v2 environment.
_XOPEN_SOURCE=600 enables XPG6, but feature_tests.h also has this bit:

/*
* It is invalid to compile an XPG3, XPG4, XPG4v2, or XPG5 application
* using c99. The same is true for POSIX.1-1990, POSIX.2-1992, POSIX.1b,
* and POSIX.1c applications. Likewise, it is invalid to compile an XPG6
* or a POSIX.1-2001 application with anything other than a c99 or later
* compiler. Therefore, we force an error in both cases.
*/

so to enable XPG6 we'd need to use C99 mode anyway. Could we just use
-std=gnu99 (with -fgnu89-inline if required) with GCC on Solaris? ISTM
it would be cleaner to just properly enable c99 mode rather than define
an undocumented macro to use a couple of c99 declarations.

/ Oskari

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Abhijit Menon-Sen 2014-09-24 13:16:06 Re: make pg_controldata accept "-D dirname"
Previous Message Heikki Linnakangas 2014-09-24 13:02:29 Re: make pg_controldata accept "-D dirname"