[R-gui] [Rd] R GUI considerations

Gabor Grothendieck ggrothendieck at gmail.com
Sat Oct 15 16:56:10 CEST 2005


I notice that there is some discussion on this at
http://developer.r-project.org/ under the links involving the
word threading.  Thus using the facilities of another language
seems to be one approach and building such facilities into
R itself another.

On 10/15/05, Philippe Grosjean <phgrosjean at sciviews.org> wrote:
> Gabor Grothendieck wrote:
> > The preferred solutions in the post all seem to involve another language:
> >
> > - tcl to use tk
> > - Python to use wxWidgets
> >
> > and other solutions mentioned also seem to involve other languages:
> >
> > - Visual Basic
> > - Java (Swing?)
> >
> > Is there some key missing feature in R with regards to GUIs
> > that requires interposing another language?
>
> Yes, there is: R is designed for CLI, and this implies a different
> approach as for a GUI (sequential execution of commands versus
> event-driven application). A language more suitable for event-driven way
> of programming is thus often interposed between R and the GUI itself.
> Have you notice that none of "external" GUIs for R do all the job in R
> itself. Perhaps, the closest one from a 100% R coded GUI is R Commander,
> but it takes benefit from the specific event loop of tcltk (otherwise,
> the whole GUI would be completelly frozen during all R computations),
> and partly of Tcl (even if it is indirect). JGR is Java, SciViews-R is
> Visual Basic, Tinn-R is Delphi, RKward is C++, etc... Of course, it is
> possible to implement extensions in C directly within R (that is what
> RGui does), another altenative, not favoured by "external" GUI
> developpers, it seems.
> best,
>
> PhG
>
> > On 10/15/05, Philippe Grosjean <phgrosjean at sciviews.org> wrote:
> >
> >>Hello,
> >>
> >>Following a discussion initiated on r-devel, that mentions SciViews-R
> >>and other GUIs issues for R, I would like to make comments (and would be
> >>happy if these comments would initiate interesting initiatives).
> >>
> >>A big, big problem with SciViews-R is that a part of it is written in
> >>Visual Basic 6, a M$$$$ language, not supported any more, buggy, non
> >>transposable to other platforms, etc, etc. Last year, I started to
> >>rewrite SciViews-R, using much more native R code, which will ultimately
> >>give what I call a "R GUI API", currently partly implemented in the
> >>SciViews bundle available on CRAN. That "R GUI API" is a lot of work:
> >>thousands of lines of code for the full API, and as I said, only a part
> >>of it is currently implemented. That API is developed with reusability
> >>in mind: platform-independent, better basis to rewrite SciViews-R in a
> >>different language, and freely available for use by other GUIs. However,
> >>nobody seems to be interested by this API (may be because it is not
> >>documented enough?). Even simple functions like progress() in the svMisc
> >>package are ignored, although they could be useful to some people. Well,
> >>I regret this situation, but I don't care much more than that: after
> >>all, the main goal is to make it the basis of the future
> >>platform-independent implementation of SciViews-R.
> >>
> >>Now, regarding the rewritting of SciViews-R itself in a
> >>platform-independent environment (i.e., language + graphical widget),
> >>this is not undertaken currently for the reason I am not satisfied
> >>enough with all curerently existing solutions for various reasons. I
> >>give just a few explanations here:
> >>
> >>1) R + Tcl + Tk. The tcltk R package is widely available and largely
> >>debugged. However, there are still problems to make a 100% Tcl console
> >>for R (look at Peter Dalgaard's attempt in the package, not complete,
> >>not 100% operational, currently). Another problem: the Tk widgets are a
> >>little bit old-fashion, and I miss a lot of features provided by more
> >>advanded graphical widgets like wxWidgets, for instance. However, it
> >>seems to be the best bet, currently. So, I explore various kinds of
> >>additions that could make Tcl + Tk a better and more modern GUI
> >>implementation. Look at http://www.sciviews.org/SciViews-R... the
> >>"tcltk2" package. In that package, I add "Tile", a series of themable
> >>and more modern widgets, the famous "tkTable", a tooltip widget, a text
> >>widget with synthax colouring, a tree widget, etc. + some Windows
> >>specific stuff that help regarding problems like focusing on a Tk
> >>window, and communicating with other apps. However, I face the problem
> >>of installing all these additional widgets seemlessly under all
> >>paltforms supported by R. For the pure Tcl widgets, no problems. But for
> >>widgets with compiled code, like Tile or tkTable, this is much more
> >>difficult... And since I mainly work on Windows...
> >>
> >>2) wxWidgets. This is a really great, very capable and
> >>platform-independent solution. I like very much James Wettenhal attemps
> >>of using wxPython (wxWidgets + Python) through RSPython and the
> >>experimental wxPython R package. However, it is still alpha, there are
> >>problems to finalize it, and his author is not continuing its
> >>development for reasons that are personal to him. So, I am not sure we
> >>will have a usable version available soon to integrate wxWidgets with R
> >>and use it for a R GUI.
> >>
> >>3) JAVA. There is a good R GUI written in JAVA: JGR. Moreover, something
> >>like the Eclipse platform is a very promizing environment for a
> >>rich-featured R IDE. JAVA specialists in my University say such a GUI,
> >>written in JAVA will be relatively slow compared to other solutions.
> >>However, JGR is a good actual counter-example.
> >>
> >>4) GTK2. Great widgets... but forgets Windows. I installed and used Gimp
> >>under Windows. This is certainly the best demonstration on what can be
> >>done with GTK2 under Windows. I must admit I am very disappointed as a
> >>Windows user: look&feel is very different, and irritating in several
> >>aspects.
> >>
> >>So, a long mail to conclude that, if I still haven't started to
> >>implement the platform-independent version of SciViews-R, it is because
> >>I am not convinced that any of the currently available R + rich-featured
> >>and platform-independent widgets solutions available is the one that
> >>will make it possible to reimplement SciViews-R in a streamless and
> >>relatively bug-free manner. To summarize, the currently best candidate
> >>is R + tcltk + tcltk2. I am convinced that R + Python + wxWidgets is by
> >>far a much better solution, but it is still in alpha development.
> >>Eclipse + JGR looks promising, and GTK2 is there too, but not enough
> >>integrated under Windows for me to start using such a solution, as a
> >>Windows user.
> >>
> >>I don't have much time to dedicate to SciViews-R and to the Sciviews R
> >>GUI API for the moment, and I will certainly focus on four hot topics:
> >>(1) developing the actual SciViews-R as a better teaching aid (I use and
> >>need it for my own courses!), (2) contributing to the development of
> >>Tinn-R, (3) further integrating R Commander with SciViews-R, and (4)
> >>integrating Rpad with SciViews-R. However, I will have various contracts
> >>in 2006-2007 where the development of SciViews-R is a part of the job.
> >>So, I will have the opportunity to hire one or two developers, and this
> >>will hopefully speed up SciViews-R development in a directly that will
> >>satisfy more users.
> >>
> >>Otherwise, I am open to any suggestion, and more importantly, to any
> >>idea of collaboration with other R GUI developers, as it is currently
> >>the case with John Fox (R Commander integration with SciViews-R),
> >>Jose-Claudio Faria (Tinn-R developement and Tinn-R compatibility with
> >>SciViews-R), and Tom Short (Rpad developments and compatibility with
> >>SciViews-R). I think it is important to insist on this, in a world
> >>populated with a myriad of slowly developed, half finished, half
> >>featured R GUIs, made by people that look too individualist in my view
> >>to be able of working all together and to write a single R GUI that has
> >>any chance to be full-featured, well-documented, reasonnably bug-free,
> >>truly platform-independent, in a near future.



More information about the R-SIG-GUI mailing list