R devel - Dec 2007 - Interactiveness

Hi list!

I'm developing R integration for a project called Bioclipse at Uppsala 
University. The current implementation works by simply forking an R and 
sending it text (with some substitutions on it) down a pipe, getting the 
printed output back up another pipe. This of course works fine, except 
it runs into one problem: R finds a pipe on its stdin and decides to be 
"non-interactive", which means that as soon as the user makes a typo
and
causes an error, R exits.

I checked the source, and it's a couple of isatty()'s in the two files 
named system.c that are doing it. They are of course intended to be a 
feature, but in this case it causes us trouble. Would it be possible to 
get a command line switch to control this behaviour? I'm not sure pseudo 
terminals can be used portably, or can they? Are those few lines with 
the isatty's, which set R_Interactive to FALSE, the only places that 
need to be patched (besides adding the switch somewhere)? Should I send 
in a complete patch or is a request for functionality sufficient?

Lots of questions there, grateful for any answers! :-)


Bjarni
-- 

                        INFORMATION WANTS TO BE FREE

On Tue, 11 Dec 2007, Bjarni Juliusson wrote:
> Hi list!
>
> I'm developing R integration for a project called Bioclipse at Uppsala
> University. The current implementation works by simply forking an R and
> sending it text (with some substitutions on it) down a pipe, getting the
> printed output back up another pipe. This of course works fine, except
> it runs into one problem: R finds a pipe on its stdin and decides to be
> "non-interactive", which means that as soon as the user makes a
typo and
> causes an error, R exits.
Actually, not so.  The default error handler for non-interactive use is to 
do that, but you can change it.
> I checked the source, and it's a couple of isatty()'s in the two
files
> named system.c that are doing it. They are of course intended to be a
> feature, but in this case it causes us trouble. Would it be possible to
> get a command line switch to control this behaviour? I'm not sure
pseudo
> terminals can be used portably, or can they?
They can, and are e.g. by ESS (except on Windows, where there is already a 
switch).  I think you need to look a bit more carefully at what other 
projects do.

-- 
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