Thanks for that. It does confirm what I was starting to suspect: that the R engine initialization in R.NET needs a change to use the R embedding API functions, though mindful that the R.NET author did settle on the current arrangement for a pragmatic reason. I also shamefully missed a recent post on this list, albeit not mentioning MPI, https://stat.ethz.ch/pipermail/r-devel/2013-November/067951.html, by Nigel Delaney, which reports similar and likely related issues.
Thanks for your and Professor Ripley's feedback. J-M ________________________________ From: Gabriel Becker [gmbec...@ucdavis.edu] Sent: Thursday, November 21, 2013 6:29 PM To: Perraud, Jean-Michel (CLW, Black Mountain) Cc: r-devel Subject: Re: [Rd] Running R embedded in an mpiexec spawned process - Fatal error: you must specify '--save', '--no-save' or '--vanilla' I don't have any real answers for you. Perhaps someone with more experience will. This may be an R.NET<http://R.NET> issue though. I'm suprised that from your comment, and a very brief glance at R.NET<http://R.NET>'s sourcecode, it is calling setup_Rmainloop. I've successfully embedded R in C programs and the entrypoint I always used was Rf_initEmbeddedR. My understanding is that setup_Rmainloop is for the non-embedded case (R running itself). Perhaps I'm wrong and using setup_Rmainloop is supported in the embedded case as well as Rf_initEmbeddedR. If not I can't offer any insight to why it usually works anyway other than Dr. Ripleys suggestion about stdin and mpiexec. Sorry I can't be of more help. ~G On Wed, Nov 20, 2013 at 10:39 PM, <jean-michel.perr...@csiro.au<mailto:jean-michel.perr...@csiro.au>> wrote: I'd like someone familiar with the R options initialization to comment on a difference of behavior within/without mpiexec I have a (.NET) application with embedded R that is proven to run in a single process: ./Sample1.exe on a Debian Linux with R 3.0.2 Running the same code with mpiexec, it fails at the R engine initialization: mpiexec -n 1 ./Sample1.exe Fatal error: you must specify '--save', '--no-save' or '--vanilla' ------------------------------------------------------------------------- The behavior is actually reproducible with the straight R mpiexec -n 1 R Fatal error: you must specify '--save', '--no-save' or '--vanilla' ------------------------------------------------------------------------- The following is working: mpiexec -n 1 R --no-save However in my Sample1 application, I do set up R init options that should be suitable AFAIK: rEngine = REngine.CreateInstance("RDotNet"); StartupParameter rStartParams = new StartupParameter { Quiet = true, SaveAction = StartupSaveAction.NoSave, Slave = false, Interactive = true, Verbose = false, LoadInitFile = true, LoadSiteFile = true, RestoreAction = StartupRestoreAction.NoRestore, NoRenviron = false }; rEngine.Initialize(rStartParams); // calls the R API R_SetParams, then setup_Rmainloop I gather that the following is hit in src/R-3.0.2/src/unix/system.c, in the function Rf_initialize_R: if (!R_Interactive && Rp->SaveAction != SA_SAVE && Rp->SaveAction != SA_NOSAVE) R_Suicide(_("you must specify '--save', '--no-save' or '--vanilla'")); I don't understand why it would complain if spawned by mpiexec and fine otherwise. I do not have a suitable debugging environment to step through a R with debug symbols, and no stack trace is given as console output. As an aside, the same code (barring platform specifics) works on Windows. Thanks ______________________________________________ R-devel@r-project.org<mailto:R-devel@r-project.org> mailing list https://stat.ethz.ch/mailman/listinfo/r-devel -- Gabriel Becker Graduate Student Statistics Department University of California, Davis [[alternative HTML version deleted]] ______________________________________________ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
