R devel - Aug 2006 - stop R mainloop without calling exit(1)

Hello,

I'm trying to make my Java application work with R, which involves
sending and retrieving data and to run R functions from within the
Java application. I also need to have "live interaction" with R, to
show the R console output (e.g. warnings and print) and to enable the
user to enter input when a function asks for it.

Therefore I created a simple R console in Java using JRI
(http://rosuda.org/JRI/), and it works ok. The only problem is that I
have to start the R main loop which doesn't return. The only way to
stop it is to give R the quit command ('q()'), but then
'exit(1)' is
called from the R code and the Java Virtual Machine (including my
application) also shuts down.

My quick solution to this problem was to modify the R source code so
that the main loop ends without the need to call exit. The 'do_quit'
method in main.c now returns NULL instead of calling exit(1) and
'Rf_ReplIteration' checks for a returned NULL after each eval() and
breaks the main loop by returning -1.

Here are the simple changes to the R source code ('+' are the lines I
added, '//' are the lines I removed):

src/main/main.c : Rf_ReplIteration(...):
------------------------------------------------------------------
value = eval(R_CurrentExpr, rho);
+if(value == NULL)
+return(-1);

src/main/main.c : do_quit():
-------------------------------------
//exit(0);
+return(NULL)

I also removed the call to exit in src/linux/sys-std.c:Rstd_CleanUp
and src/gnuwin32/system.c:R_CleanUp for linux and windows
respectively.

This is obviously a quick hack to make it work (but it works fine,
also the standalone R exits ok this way), but I can imagine more
people want to use the mainloop with the possibility to return from it
without making their application to exit too. I wonder if it's
possible to include this possibility in future versions of R. Or do I
see this wrong and are there other (easier) solutions to achieve what
I want? Thanks in advance for your comments/reactions.

Thomas

This is already covered in the R-devel version of R: please consult
`Writing R Extensions' in that version.

On Thu, 31 Aug 2006, Thomas Kelder wrote:
> Hello,
> 
> I'm trying to make my Java application work with R, which involves
> sending and retrieving data and to run R functions from within the
> Java application. I also need to have "live interaction" with R,
to
> show the R console output (e.g. warnings and print) and to enable the
> user to enter input when a function asks for it.
> 
> Therefore I created a simple R console in Java using JRI
> (http://rosuda.org/JRI/), and it works ok. The only problem is that I
> have to start the R main loop which doesn't return.
But that's not true.  The documented way is

setup_Rmainloop
run a loop (which can return)
end_Rmainloop

The problem was that the only documented way to shut R down cleanly was 
end_Rmainloop (which did call exit), whereas there now is Rf_endEmbeddedR.
> The only way to stop it is to give R the quit command ('q()'), but
then
> 'exit(1)' is called from the R code and the Java Virtual Machine 
> (including my application) also shuts down.
That's because you called end_Rmainloop, possibly by calling 
run_Rmainloop rather than taking control of the loop.

It seems that almost all users of embedded R did not call end_Rmainloop, 
and I met this when creating a common embedding interface for Unix and 
Windows versions of R: the Windows widgets (such as the graphics device) 
do need to be told to shut down.

[...]

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

Dear Professor Ripley,

Thank you for your reply. I have a few questions left:
> > But that's not true.  The documented way is
> >
> > setup_Rmainloop
> > run a loop (which can return)
> > end_Rmainloop
> >
> > The problem was that the only documented way to shut R down cleanly
was
> > end_Rmainloop (which did call exit), whereas there now is
Rf_endEmbeddedR.
In the documentation  "run_Rmainloop" is called to start the loop:

         Rf_initEmbeddedR(argc, argv);
         /* submit some code to R, which is done interactively via
             run_Rmainloop();
         */
         Rf_endEmbeddedR(0);

But that method doesn't return normally, right? It only ends if
ReplIteration returns -1.
Or do you mean I need to re-implement the read-and-evaluate loop in a
way that I can interrupt it (at "run a loop (which can return)" in
your example)? So instead of calling run_Rmainloop, I can copy the
loop in R_ReplConsole, add an interrupted flag and check that every
iteration.
> > It seems that almost all users of embedded R did not call
end_Rmainloop,
> > and I met this when creating a common embedding interface for Unix and
> > Windows versions of R: the Windows widgets (such as the graphics
device)
> > do need to be told to shut down.
> Another idea on Unix-alikes is to replace R_CleanUp by your own callback:
> it is part of the documented interface for writing a front-end.
So for the Unix alikes I could use a customized R_CleanUp for the
non-devel versions of R. For windows I should still call
end_Rmainloop, but if I understand you well, this is only needed if I
use the windows widgets in R. I don't want to, so it shouldn't be a
problem then?

Thank you for your help, I'm going to try to modify the JRI interface
by writing a custom loop instead of calling run_Rmainloop.

By the way, I use R version 2.2.1 in linux, because I got the "Error:
C stack usage is too close to the limit R" error in higher versions (I
saw there was a topic about this already:
https://stat.ethz.ch/pipermail/r-devel/2006-April/037219.html). I
tried to use the R-devel version to make use of the Rf_endEmbeddedR
method, but the error also occurs in this version. Could this be a
bug?

Thomas

2006/8/31, Prof Brian Ripley <ripley at
stats.ox.ac.uk>:
> Another idea on Unix-alikes is to replace R_CleanUp by your own callback:
> it is part of the documented interface for writing a front-end.
>
> On Thu, 31 Aug 2006, Prof Brian Ripley wrote:
>
> > This is already covered in the R-devel version of R: please consult
> > `Writing R Extensions' in that version.
> >
> > On Thu, 31 Aug 2006, Thomas Kelder wrote:
> >
> > > Hello,
> > >
> > > I'm trying to make my Java application work with R, which
involves
> > > sending and retrieving data and to run R functions from within
the
> > > Java application. I also need to have "live
interaction" with R, to
> > > show the R console output (e.g. warnings and print) and to enable
the
> > > user to enter input when a function asks for it.
> > >
> > > Therefore I created a simple R console in Java using JRI
> > > (http://rosuda.org/JRI/), and it works ok. The only problem is
that I
> > > have to start the R main loop which doesn't return.
> >
> > But that's not true.  The documented way is
> >
> > setup_Rmainloop
> > run a loop (which can return)
> > end_Rmainloop
> >
> > The problem was that the only documented way to shut R down cleanly
was
> > end_Rmainloop (which did call exit), whereas there now is
Rf_endEmbeddedR.
> >
> > > The only way to stop it is to give R the quit command
('q()'), but then
> > > 'exit(1)' is called from the R code and the Java Virtual
Machine
> > > (including my application) also shuts down.
> >
> > That's because you called end_Rmainloop, possibly by calling
> > run_Rmainloop rather than taking control of the loop.
> >
> > It seems that almost all users of embedded R did not call
end_Rmainloop,
> > and I met this when creating a common embedding interface for Unix and
> > Windows versions of R: the Windows widgets (such as the graphics
device)
> > do need to be told to shut down.
> >
> > [...]
> >
> >
>
> --
> Brian D. Ripley,                  ripley at stats.ox.ac.uk
> Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
> University of Oxford,             Tel:  +44 1865 272861 (self)
> 1 South Parks Road,                     +44 1865 272866 (PA)
> Oxford OX1 3TG, UK                Fax:  +44 1865 272595
>