Dear Yaroslav,
Having an embedded R running in "interactive" mode seems needed (and
other interfaces to R seem to have it that way, or at least it was the
case last time I looked at it).
The recommended use of Rf_initEmbeddedR() is setting R_Interactive to
TRUE (on UNIX systems, the win32 path is a different story).
Because of the limited control over what Rf_initEmbeddedR() was doing
(and issues originating from this), rpy2 is no longer using it and you
can hack R_Interactive to FALSE easily (line 1015 in rinterface.c for
2-1.dev).
I have not experimented much beyond trying to set to FALSE during the R
initialization in rpy2 and see that it would somehow run through few
examples in a console, but cause it to crash when running the suite of
unittests.
It would be helpful if you could experiment a bit with that.
If setting R_interactive to FALSE is needed, I can add it either as a
parameter to rpy2.rinterface.initr() or as a new method (for example
rpy2.rinterface.set_interactive(True | False))
Best,
L.
Yaroslav Halchenko wrote:
> Dear RPy Developers,
>
> Thanks for the product. Within our little PyMVPA (www.pymvpa.org) we use
> it to interface with some machine learning methods available in R. Problem
> comes whenever R code segfaults -- then regardless either python section is
> interactive or not we are presented with R's dialog giving options to save
> dump
> etc... ie the code
>
> 00562 if(R_Interactive) {
> 00563 REprintf("\nPossible actions:\n1: %s\n2: %s\n3: %s\n4: %s\n",
> 00564 "abort (with core dump, if enabled)",
> 00565 "normal R exit",
> 00566 "exit R without saving workspace",
> 00567 "exit R saving workspace");
> 00568 while(1) {
> 00569 if(R_ReadConsole("Selection: ", ConsoleBuf,
> CONSOLE_BUFFER_SIZE,
> 00570 0) > 0) {
> 00571 if(ConsoleBuf[0] == '1') break;
> 00572 if(ConsoleBuf[0] == '2') R_CleanUp(SA_DEFAULT, 0, 1);
> 00573 if(ConsoleBuf[0] == '3') R_CleanUp(SA_NOSAVE, 70, 0);
> 00574 if(ConsoleBuf[0] == '4') R_CleanUp(SA_SAVE, 71, 0);
> 00575 }
> 00576 }
>
> That stalls unittests which are running overnight, and moreover for some
> reason
> it fills up stdout (or stderr... didn't check - stderr gets redirected to
> stdout) with "Selection: " until disk fills up ;)
>
> So, the question is -- since embedded R starts with hardcoded
> R_Interactive, is there some other way to prevent "interaction" with segfault
> 'dialog'?
>
> In my blind attempt I've tried to do something evil like
>
> def input1(*args): return "1" # which is "1: abort (with core
> dump, if enabled)"
> rpy.set_rpy_input(input1)
>
> which seemed to work at first, but whole battery of unittests started to
> puke at some other places with "XXX lineno: 19, opcode: 0" and "SystemError:
> unknown opcode" at other places...
>
> I was using python2.5 and python-rpy 1.0.3-8 (rebuilt with debugging
> symbols) on Debian amd64 system.
>
> Thanks in advance for ideas!
------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
rpy-list mailing list
rpy-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/rpy-list
