[Rcpp-devel] R session crashes when largely using a Rcpp sourced function

[Jonathan Olmsted]

Pierre,

Have you used a tool like Valgrind? There is a bit of a learning curve, but
it is exceedingly useful.

A few comments:

1) For cases likes this gctortute(TRUE) in your R script helps those “for
enough iterations in a loop…” bugs to present quickly. With gctorture(TRUE)
I’m producing the crash with fewer than 50 iters.

​2) What you are implementing has everything to do with RNGs, so setting
the seed is useful for reproducibility​
​.​

2) I got useful messages from Valgrind after changing your prefix
increments to postfix increments, but I didn’t before making that switch
​ (if I remember correctly)​
.

3) These msgs points me to these lines

            for(int i = 1; i < kappa+2; i++){
                phi_omega[i-1] = phi_C(arma::trans(omegas.row(i)),
                                       piks,muks,Cks,Gamma);
            }
            accept = all_C((phi_omega < Upsilon));

in main_function_C(). Pretty sure you have an “off by 1” error here. kappa
is an int, phi_omega has kappa elements (in elements 0 through kappa -1),
but you refer to elements 1-1=0 through kappa+1-1=kappa which is the
position of the kappa+1st elem.

I switched your loop to for(int i = 1; i < kappa+1; i++) and it is running
under Valgrind now (i.e. slowly) and it’s gotten farther than any other run
thus far.
​ Without having a conceptual understanding of what you are trying to do, I
have no idea if this is the right fix for you (and I may be mistaken
altogether of course...).

HTH,
Jonathan​

​

On Tue, Oct 7, 2014 at 8:17 PM, Dirk Eddelbuettel <edd at debian.org> wrote:

>
> Hi Pierre,
>
> On 8 October 2014 at 01:59, Pierre.Gloaguen at ifremer.fr wrote:
> | I will do what I can to reduce the code and have to same error occuring.
> | Of course i don't expect you to read all of it :) I was just wondering
> | if that kind of memory problem occured before, and if has some
> | "typical/generic" solutions.
> | I will try to have a shortest code to exhibit.
>
> Really appreciate it.
>
> Many of us found over the years that just by "creating a better example" to
> often drill down close enough to the problem core.
>
> And the easier you make it for folks to help you, the likier it is that
> some
> kind soul from around here can give you a hand.
>
> Just lowering the barriers, be it cognitive or just plain effort ...
>
> Best, Dirk
>
> --
> http://dirk.eddelbuettel.com | @eddelbuettel | edd at debian.org
> _______________________________________________
> Rcpp-devel mailing list
> Rcpp-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
>



-- 

J.P. Olmsted
j.p.olmsted at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/rcpp-devel/attachments/20141008/9ec783b6/attachment.html>