Compatibility of external C code across platforms

An embedded and charset-unspecified text was scrubbed...
Name: not available
URL: <https://stat.ethz.ch/pipermail/r-devel/attachments/20090926/b72bddac/attachment.pl>

A package I have written makes considerable use of external C code. 
There appear to be no problems building a Windows binary with the 
GNU C compiler, but now I would like to make the package available 
for other platforms.  A stringent check of the code with the gcc 
options '-Wall -pedantic' flags many nested functions and 
variable-length arrays that are not allowed in ISO C.  Is it 
essential that distributed C source code for R packages conforms to 
ISO C in this respect?  I'm working in Windows XP with R2.9.2 and 
Rtools 2.9.
Yes, it is essential.  R does not assume gcc, and it does not even 
assume gcc 4.x (and people are still using gcc 3.x on some platforms, 
e.g. the sunfreeware R builds).  This is why we publish test result on 
CRAN for a completely different compiler (from SunPro).

However, if is not clear which standard you mean by 'ISO C': assuming 
C99 is fairly safe, but GNU extensions from C99 are not -- nested 
functions are one such.
[Incidentally, when I use a makevars.win file to include these gcc 
options in Rcmd build etc. the problem with variable-length arrays 
is not flagged; maybe my command-line gcc is a different version 
from that in Rtools]
Well, you (and only you) can easily check that for yourself, but the 
other flags also matter, especially --std.  Because it has been needed 
for the use of the MinGW headers, R on Windows is built with 
--std=gnu99.  (Linux headers have a similar problem: using --std=c99 
disables some features we test for in configure and wish to use -- and 
there are OS features, not language extensions.)
Murray Efford
University of Otago
Dunedin, New Zealand
murray.efford at otago.ac.nz<mailto:murray.efford at otago.ac.nz>

	[[alternative HTML version deleted]]
Please note what the posting guide has to say about that.

______________________________________________
R-devel at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel

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
Thank you. I'm much clearer on this. I now notice that the messages from gcc distinguish breaches of ISO C and ISO C90, and can see the significance.  My gcc versions were the same - the different behaviour is likely to be explained by the --std flag. And it looks like I have to re-write a lot of C to eliminate nesting of functions.  Point taken about HTML.
Murray Efford
On Sat, 26 Sep 2009, Murray Efford wrote:

> A package I have written makes considerable use of external C code.
> There appear to be no problems building a Windows binary with the
> GNU C compiler, but now I would like to make the package available
> for other platforms.  A stringent check of the code with the gcc
> options '-Wall -pedantic' flags many nested functions and
> variable-length arrays that are not allowed in ISO C.  Is it
> essential that distributed C source code for R packages conforms to
> ISO C in this respect?  I'm working in Windows XP with R2.9.2 and
> Rtools 2.9.

Yes, it is essential.  R does not assume gcc, and it does not even
assume gcc 4.x (and people are still using gcc 3.x on some platforms,
e.g. the sunfreeware R builds).  This is why we publish test result on
CRAN for a completely different compiler (from SunPro).

However, if is not clear which standard you mean by 'ISO C': assuming
C99 is fairly safe, but GNU extensions from C99 are not -- nested
functions are one such.

> [Incidentally, when I use a makevars.win file to include these gcc
> options in Rcmd build etc. the problem with variable-length arrays
> is not flagged; maybe my command-line gcc is a different version
> from that in Rtools]

Well, you (and only you) can easily check that for yourself, but the
other flags also matter, especially --std.  Because it has been needed
for the use of the MinGW headers, R on Windows is built with
--std=gnu99.  (Linux headers have a similar problem: using --std=c99
disables some features we test for in configure and wish to use -- and
there are OS features, not language extensions.)

>
> Murray Efford
> University of Otago
> Dunedin, New Zealand
> murray.efford at otago.ac.nz<mailto:murray.efford at otago.ac.nz>
>
>       [[alternative HTML version deleted]]

Please note what the posting guide has to say about that.

>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>

--
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
An embedded and charset-unspecified text was scrubbed...
Name: not available
URL: <https://stat.ethz.ch/pipermail/r-devel/attachments/20090926/ee0678f4/attachment.pl>