R devel - Oct 2007 - bnlearn package compilation failure on MacOSX

Hi all.

I've recently uploaded a package (bnlearn) to CRAN. It builds fine 
on both Linux (32 and 64 bit) and Windows, but fails on MacOSX ix86
because of C90 vs C99 issues:

http://www.r-project.org/nosvn/R.check/r-patched-macosx-ix86/bnlearn-00install.html

Since I've no MacOSX machine at hand, I would like to ask you:
why is C99 not the default for gcc on MacOSX ix86? Is it safe to 
force gcc to use C99 with either -std=c99 o -std=gnu99?

Thanks in advance.


-- 
Marco Scutari                               
Linux Registered User #341807                  http://counter.li.org
powered by : 
Debian Sid GNU/Linux (SGI-XFS)                       Kernel 2.6.21.3

You've got more serious problems than that: you cannot assume gcc.
For example on the SunPro compiler I got

cc -I/home/ripley/R/R-devel-SunPro/include 
-I/home/ripley/R/R-devel-SunPro/inclu
de  -I/usr/local/include    -Kpic  -xO5 -xc99 -xlibmil -nofstore -c 
mutual.infor
mation.c -o mutual.information.o
"mutual.information.c", line 34: member can not have variably modified
type: n
"mutual.information.c", line 35: member can not have variably modified
type: m
"mutual.information.c", line 41: cannot dereference non-pointer type
"mutual.information.c", line 52: cannot dereference non-pointer type
"mutual.information.c", line 52: cannot dereference non-pointer type
"mutual.information.c", line 53: cannot dereference non-pointer type
"mutual.information.c", line 53: cannot dereference non-pointer type
"mutual.information.c", line 54: cannot dereference non-pointer type
"mutual.information.c", line 54: cannot dereference non-pointer type
"mutual.information.c", line 60: warning: improper pointer/integer 
combination:
arg #5
"mutual.information.c", line 62: cannot recover from previous errors

Note that is in C99 mode, and under Linux.


On Thu, 4 Oct 2007, Marco Scutari wrote:
>
> Hi all.
>
> I've recently uploaded a package (bnlearn) to CRAN. It builds fine
> on both Linux (32 and 64 bit) and Windows, but fails on MacOSX ix86
> because of C90 vs C99 issues:
>
>
http://www.r-project.org/nosvn/R.check/r-patched-macosx-ix86/bnlearn-00install.html
>
> Since I've no MacOSX machine at hand, I would like to ask you:
> why is C99 not the default for gcc on MacOSX ix86? Is it safe to
> force gcc to use C99 with either -std=c99 o -std=gnu99?
>
> Thanks in advance.
>
>
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

Marco,

On Oct 4, 2007, at 2:05 PM, Marco Scutari wrote:
> I've recently uploaded a package (bnlearn) to CRAN. It builds fine  
> on both Linux (32 and 64 bit) and Windows, but fails on MacOSX ix86  
> because of C90 vs C99 issues:
>
> http://www.r-project.org/nosvn/R.check/r-patched-macosx-ix86/ 
> bnlearn-00install.html
>
> Since I've no MacOSX machine at hand, I would like to ask you: why  
> is C99 not the default for gcc on MacOSX ix86?

It was not deliberate. The current R build (R 2.6.0) was modified to  
use -std=gnu99 in the CC section for both architectures instead of  
using CFLAGS, that should solve the problem. A work around for older  
R is to set PKG_CFLAGS=-std=gnu99 when installing bnlearn.

However, note that C99 compiler is not required for R so it may not  
be available. Since you don't really need any advanced functionality  
of C99, it may be a good idea to change it to a more portable C code.

> Is it safe to force gcc to use C99 with either -std=c99 o -std=gnu99?
>
Not in your package as cannot assume gcc in general, but temporarily  
for R 2.5.1 on OS X, yes.

Thanks,
Simon