[R-pkg-devel] cmake, Windows

Tue Mar 15 13:00:46 CET 2022

Dear Matthias,

On 3/13/22 16:31, Matthias Gondan wrote:
>
> Dear Tomas,
>
> ok, I’ve seen it now; the cmake is a bit of a mess. I used the one 
> that is found on r-hub, but it points to gcc.exe in 
> c:/strawberry/perl/ etc., so using cmake just brings trouble at this 
> stage.
>
> I have now followed your advice and prepared an R package that 
> connects to an existing Installation of SWI-Prolog on the system.
>
> https://github.com/mgondan/rologlite
>
> [note that currently, it only works properly with the development tree 
> of SWI-Prolog, but they update quite often]
>
> Is is possible to submit such a package to CRAN at all? What is the 
> next step? Your advice is greatly appreciated.
>
Yes. I would say that ideally your package would work even with older 
versions of Prolog (yet possibly without lacking some features only 
available in recent versions). I don't know whether that makes sense for 
what you are doing, but that would be ideal for R users, who could then 
install SWI Prolog the usual way on their system (e.g. via Debian 
package manager, simply swi-prolog, when on Debian, and similarly on 
other systems). On Windows, they can install SWI Prolog from the 
official installer, which can be recent, but it is not good requiring 
them to update. On Linux distributions it would take longer for the 
updates to propagate.

So the package should be able to use an existing installation of SWI 
Prolog for base functionality at least and pass checks using it. Users 
wanting recent functionality requiring newer SWI Prolog could then be 
instructed to install that manually/pointer to SWI Prolog installation 
manual.

See Writing R Extensions and CRAN Repository policy for more on how to 
specify external dependencies. In short, there is SystemRequirements 
field of the package, it needs to be documented how to install the 
software, the package should not depend on a specific version of SWI 
Prolog, the package once installed should not depend on a specific 
version of SWI Prolog (so that binary R packages can be provided).

If this is not possible and you need a new version of SWI Prolog on all 
systems, newer than normally available, perhaps best to write to CRAN 
and ask there about this specific case.

Best
Tomas

> Best wishes,
>
> Matthias
>
> *Von: *Tomas Kalibera <mailto:tomas.kalibera using gmail.com>
> *Gesendet: *Montag, 7. März 2022 10:45
> *An: *Simon Urbanek <mailto:simon.urbanek using R-project.org>; 
> matthias-gondan <mailto:matthias-gondan using gmx.de>
> *Cc: *r-package-devel using r-project.org
> *Betreff: *Re: [R-pkg-devel] cmake, Windows
>
> On 3/6/22 21:46, Simon Urbanek wrote:
>
> > BTW: one option I forgot to mention would be to add SWI-Prolog to 
> the toolchain instead so you don't need to build it. However, I don't 
> know how tied you are to the development branch that you're using. 
> There is also the official SWI release Windows binary which may be 
> another option, but I didn't look at the details if that fits your 
> bill. I would suggest setting up the UCRT environment locally and 
> testing the different options so you can decide which is the best path 
> forward.
>
> Right, primarily the package should use an existing installation of SWI
>
> Prolog in the system. This is also required by the CRAN repository
>
> policy, see "Using external C/C++/Fortran libraries.").
>
> So if SWI Prolog official release for Windows has the necessary headers
>
> and DLLs, your R package could link to them. One example of R package
>
> doing this is rjags, dynamically linking to JAGS (though when there is a
>
> plain C interface, it is preferable as it is easier to maintain, see
>
> WRE). This would be the simplest option and you could use dynamically
>
> linked SWI Prolog.
>
> Adding a complete language runtime into an R package feels a bit odd to
>
> me, but in principle is possible. If Rtools42 libraries included a
>
> static library of SWI Prolog, your package could link to that. Rtools42
>
> doesn't include it, but in principle it could be added (Rtools42
>
> libraries themselves are built on Linux and indeed cmake is available
>
> there, some of those use it), I could help with that. This wouldn't work
>
> well if you needed to use dynamic linking.
>
> Building SWI Prolog from source included in the R package should not be
>
> needed and it won't be any easier than adding it to Rtools42. Rtools42
>
> currently doesn't have cmake ready for use (adding it is trivial via
>
> pacman, but some work and documentation would be needed to make it find
>
> the libraries).
>
> Best
>
> Tomas
>
> >
>
> > Cheers,
>
> > Simon
>
> >
>
> >
>
> >> On Mar 7, 2022, at 9:05 AM, Simon Urbanek 
> <simon.urbanek using R-project.org> wrote:
>
> >>
>
> >> Matthias,
>
> >>
>
> >> that, that is helpful. So we are talking about UCRT here. The 
> toolchain from https://github.com/kalibera/ucrt3 appears to not 
> include the cmake binary. I suspect the reason is that since it is 
> based on MXE, the cmake binary used for the toolchain is native 
> (Linux) and not the Windows binary. I think there are two options: 
> either use the CMake officially released binary or have the toolchain 
> supply cmake.exe. I'd defer to Tomas who is maintaining the UCRT 
> toolchain to advise which is the better option.
>
> >>
>
> >> Cheers,
>
> >> Simon
>
> >>
>
> >>
>
> >>
>
> >>> On Mar 6, 2022, at 12:37 PM, matthias-gondan 
> <matthias-gondan using gmx.de> wrote:
>
> >>>
>
> >>> Thanks Simon,
>
> >>>
>
> >>> I'll remove the paths and report the results. I'll also check 
> Dirks suggestion (the other response).
>
> >>>
>
> >>> For what it's worth, the error message was simply "cmake not 
> found"; when I added the path it was "/ucrt64/bin/cmake not found"
>
> >>>
>
> >>> 
> https://win-builder.r-project.org/incoming_pretest/rolog_0.9_20220305_172741/Windows/00install.out
>
> >>>
>
> >>> Best wishes,
>
> >>>
>
> >>> Matthias
>
> >>>
>
> >>> -------- Ursprüngliche Nachricht --------
>
> >>> Von: Simon Urbanek <simon.urbanek using R-project.org>
>
> >>> Datum: 05.03.22 23:19 (GMT+01:00)
>
> >>> An: matthias-gondan <matthias-gondan using gmx.de>
>
> >>> Cc: r-package-devel using r-project.org
>
> >>> Betreff: Re: [R-pkg-devel] cmake, Windows
>
> >>>
>
> >>> Matthias,
>
> >>>
>
> >>> You didn't supply any output, so we can only guess. From what I 
> can see in your package you are hard-coding incorrect paths to 
> compilers and tools in your Makevars.{win|ucrt}. Please make sure you 
> use R's compilers - never hard-code paths to tools, you need to use 
> R's settings since the toolchain can be installed in any location.
>
> >>>
>
> >>> Cheers,
>
> >>> Simon
>
> >>>
>
> >>>
>
> >>>> On Mar 6, 2022, at 6:17 AM, matthias-gondan 
> <matthias-gondan using gmx.de> wrote:
>
> >>>>
>
> >>>> Dear package developers, I have submitted a source package to 
> CRAN, and it is rejected because installation fails on 
> Windows.https://github.com/mgondan/rologThe compilation needs cmake, 
> which is available on RTools4+.Is there any chance that cmake is 
> included on the build system? If not, does anyone know a good solution 
> for this problem?Best regards,Matthias
>
> >>>> [[alternative HTML version deleted]]
>
> >>>>
>
> >>>> ______________________________________________
>
> >>>> R-package-devel using r-project.org mailing list
>
> >>>> https://stat.ethz.ch/mailman/listinfo/r-package-devel
>
	[[alternative HTML version deleted]]