Re: [O] [babel][PATCHES] ob-R patches for review

[Charles C. Berry]

On Fri, 16 May 2014, Rainer M Krug wrote:

> Sorry for coming back to your suggestions so late, but I did some
> thinking about the approach. Comments inline below.

No problem. I am about to leave town and my email for a couple of weeks,
so I might not be able to reply to further correspondence till June.

Comments inline below.

Chuck
> "Charles C. Berry" <address@hidden> writes:
>
> > On Mon, 12 May 2014, Rainer M Krug wrote:
> >
> > > Eric Schulte <address@hidden> writes:

[...]

[Rainer writes]
> > > My idea is to have all R code in one directory and to let ESS load it
> > > upon initialization of ESS (which is a dependency of running R from org
> > > anyway, if I am not mistaken).
[Chuck]
> > Not quite. You can run R src blocks without (require 'ess) if they require
> > no session. (I do not claim that anyone actually runs R src blocks who
> > lacks a working ESS installation, just saying...)
[Rainer again]
> Hm - would it then be possible to add ESS as a requirement for running R
> code, as it is already mentioned on [1] as a requirement?

I'd be surprised if it would have any impact on users. Who uses R and
org-mode without ESS? But as you say it is listed as a requirement.

[...]

> I am now leaning towards the following approach:
>
> New behavior (environment):
> 1) Define the R code in emacs variables, i.e. customizable in emacs.
> 2) Define a function to load the code into an R environment - not customizable.
>
> ## upon creation of the R session (i.e. only once for each R session)

Maybe wait till the first variable transfer is called, if it ever is.

> 3) use the load function to load the code from the emacs variables into
> an R environment called "org:functions"
> 4) load all .R files in the directory ~/.orgFunctions/ into
> "org:functions". This will mean that when functions of the same name as
> loaded earlier will be overwritten, i.e. the functions are customizable
> by R code only.
>
> ## before variable transfer at each execution of a code block

Only if there is a variable transfer, of course.

> 5) source all .R files in the directory RWorkingDirectory/.orgFunctions/
> i.e. they are not loaded into a specific environment. Reasoning: further
> customization possible, kind of an "R package light" approach.
>
> 6) use the load functions (as defied initially in (4), (3) or in () from
> to load the variables. If they are loaded into an environment or not
> depends on the definition of the functions.
>
> Concerning old versus new behavior, I would actually suggest to put this
> into the R functions: one could add an argument "variablesInEnvironment"
> which would, if set, be the name of the environment, i.e. default value
> "org:variables", if NULL it would load into .GlobalEnv, i.e. the old
> behaviour. In org, one could have a variable
> org-babel-R-variable-transfer-type (or whatever the proper name should
> be) and to make it user configurable:
>
> - null :: old behaviour
> - "org:variables" (default) :: variables into environment called
> "org:variables"
> - other string :: name of the environment to be used
>
> As stated earlier, as I think that name clashes are more unlikely with
> the new approach, I would suggest to make "org:variables" the default.

The above sounds good.

If you do not wish to implement the `org-babel-R-assign-elisp-function' 
variable approach I suggested, I'll make it a TINYCHANGE when I return in 
June. It will not impact on what you propose. However, if your 
implementation makes it possible for arbitrary elisp variables to be 
handled in R - I might not bother.

[...]

> > > True - but I like the simplicity of being able to customize the
> > > behavior of org-babel-R by writing an R function without having to thin
> > > about elisp. But maybe there is a way of doing both...
> >
> > noweb will do it. Quote the chunk like this:
> >
> > #+BEGIN_SRC emacs-lisp :noweb yes :tangle elisp-R.el
> >    (setq rlines "
> > <<r-code>>")
> > #+END_SRC
>
> I must admit, I have never used macros and noweb expansion...
>
> If you could give me your opinions on this approach, I would look into
> implementing it.

It is just a labor saving suggestion for you -- it would not show up
in the final elisp. Put all your elisp and R code in src blocks in a
*.org file. You can edit and format the R code in R src blocks and
likewise the elisp. A src block like the one above will import the R
into elisp as a string. You only tangle the elisp. So the benefit is
you have ease of editting R and writing unit tests for it for R code
that appears in elisp.

IMO, noweb is a good way to mix languages in org-mode. I use the R 
`inline' package to write C functions that R can call. R needs the code as 
strings. I write the C code in C src blocks then assign it in as an R 
character string in an R src block. I can write an R wrapper and unit test 
for each little snippet of C code, then when it all looks good join up the 
snippets by putting multiple noweb references on successive lines.


> Footnotes:
> [1]  http://orgmode.org/worg/org-contrib/babel/languages.html
[...]