[Rd] NAMESPACE file generation issue R 2.14.0 on Debian Squeeze

Gabor Grothendieck ggrothendieck at gmail.com
Wed Nov 9 15:57:38 CET 2011


2011/11/9 Uwe Ligges <ligges at statistik.tu-dortmund.de>:
>>>
>>> Honestly, that (svn revision) is the only part that we do not have on the
>>> front pages but they are given in the log files.
>>
>> The R version is not on the package's CRAN page, e.g.
>> http://cran.r-project.org/web/packages/sqldf/index.html  I was
>> thinking of something like this for the Package source line:
>>
>> Package built source:    sqldf_0.4-3.tar.gz (built with R version 2.14.0
>> Patched (2011-10-10 r12345)
>
> Not relevant for the majority of cases. Version number of the package is
> relevant, not the R used to build it (well, there are cornercases, e.g. for
> knowing which version was used to generate the vignette).

The point is that the answer to the poster's question does not really
resolve it except for that one person.  To really solve the problem
the process needs to be changed so that its unlikely to occur again.

>
> And in addition, we to not have the information. Where do you think we can
> get that from? Even if we change R to provide the information now, we'd only
> know if it was built with R <= 2.14.0 or later.

Yes, I realize that but it would have to be added during the build
process in the same way that certain other information is added
already.

>
>
>> This seems important since that line truly is not the source but is
>> the built source.  The actual source is not uploaded to CRAN or shown.
>
> He? What is the difference between "the source" and "the built source"?  A
> source package has always undergone R CMD build on the submitters machine.
> We do only have only one kind of source packages on CRAN.

The built source has been processed by R CMD build and the true source
has not.  What CRAN labels the source is not the true source but is
really the built source. There has been a transformation applied to
the true source to get the built source and some version of R was used
to do that.  Furthermore, what the result looks like can depend on
what version of R was used for this.  That its different for different
versions of R was one of the things that precipitated this entire
thread.

>
>
>> A transformation has been applied to it
>
> Where and why do you want to apply transformations?

By transformation I am referring to the existing process of
transforming the true source to the built source.  I was not
suggesting other transformations (other than the fact that it implies
that it would be necessary to identify the R version that built any
given built source in the built source itself).

>
>
>> so rightfully that should be
>> tracked back via the R version.
>>
>> This would make it very clear not only which version of the R package
>> you are dealing with but which version of R built it and that can in
>> certain circumstances be important for precisely identifying it.  This
>> could also be done for the .zip file, etc. shown on that page.
>
> The zip file shown *there* is always R-release or R-patched (hence currently
> R-2.14.0 given your request) for ALL packages. We do not list zip files for
> non R-release *there*. I thought you were still talking about the check
> page.

Originally I was referring to the check page but then expanded it to
the main page as well since it seemed relevant in both places.  I find
this whole area is confusing (and clearly I am not the only one hence
this thread) and it could use some clarification right on the web
pages themselves to avoid these sorts of problems.  I think the most
desirable from a user's viewpoint is to actually stick the R version
right there but if its not feasible some other way of addressing this
would be nice.

>
>
>>
>> If that level of detail is not feasible then this would be next best
>> (just showing the R x.y.z version):
>> Package built source:   sqldf_0.4-3.tar.gz (built with R version 2.14.0)
>>
>>
>
> I give up.
>
> uwe
>



-- 
Statistics & Software Consulting
GKX Group, GKX Associates Inc.
tel: 1-877-GKX-GROUP
email: ggrothendieck at gmail.com



More information about the R-devel mailing list