[R-pkg-devel] "Writing R Extensions" manual needs update [was: Error handling in C code]

[Tomas Kalibera]

On 6/24/24 09:45, Agner Fog wrote:
> On 24/06/2024 08.56, Tomas Kalibera wrote:
>>
>> In principle it is the NEWS file that gives hints about important 
>> changes and then one can find the details in Writing R Extensions.  
>> The NEWS file as of mid-April says that "R CMD check --as-cran" will 
>> compile C++ code with -DR_NO_REMAP. And then in Writing R Extensions, 
>> one can read about what is R_NO_REMAP and that this has been planned 
>> to become the default in the future. It is an important thing to do 
>> for the maintainability of the package space: the situation with name 
>> clashes with system headers (typically appearing just on some 
>> platforms) has only been getting worse.
>>
> Thanks, but I couldn't find it when I searched for help, and I still 
> can't find the news you are mentioning. The "Writing R Extensions" 
> guide does not list the required Rf_ function names. This manual 
> definitely has to be changed.

The NEWS file is part of the source tree, e.g. 
https://svn.r-project.org/R/trunk/doc/NEWS.Rd

You can find a rendered version here:

https://cran.r-project.org/doc/manuals/r-release/NEWS.html
https://cran.r-project.org/doc/manuals/r-devel/NEWS.html

>
> I predict chaos when you add R_NO_REMAP to the Windows compilation: A 
> large number of old packages containing C++ code will fail. Mails will 
> be sent to the maintainers. Some maintainers will be uncontactable. 
> Some packages are maintained by somebody less competent than the 
> original programmers. Some maintainers will be too busy to react, or 
> they have forgotten how to submit a package. You will be bombarded 
> with help requests. In the end, somebody will have to help update the 
> packages when the maintainers fail to do so.

I am not the one working specifically on these changes, so I don't have 
the exact details, but in principle, what you describe is daily business 
of maintaining R and CRAN. All of what you describe has been considered 
and changes like this are carefully tuned and planned based on actually 
running checks of all CRAN packages on patched versions of R-devel, 
usually outside the master repository. There is a lot of effort behind 
such changes that may not be immediately visible.

>
> This mess is a consequence of a bad decision many years ago when 
> somebody used non-unique function names like "error" and "warning". 
> But the world looked very different back then, and now we are facing 
> the consequences.
>
> It would be useful to do a test to see how many packages need to be 
> changed.

As above, this is something being done routinely. And it is a moving 
target, too, in this case, as many packages run into naming conflicts 
and are being fixed. Often the package authors receive patches or hints 
how to fix from R Core or the CRAN team.

Tomas
>
> - Agner
>