I believe there is a memory leak in isoreg in the current version of R,
as I believe the following shows
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 120405 3.3 350000 9.4 350000 9.4
Vcells 78639 0.6 786432 6.0 392463 3.0
> for(k in 1:100) {
+
+ y <- runif(10000)
+ isoreg(x,y)
+ }
> rm(x)
> rm(y)
> gc()
used (Mb) gc trigger (Mb) max used (Mb)
Ncells 121637 3.3 350000 9.4 350000 9.4
Vcells 578615 4.5 1300721 10.0 1300642 10.0
^^^
Looking at the C code, I believe the problem arises as a consequence of
using SETLENGTH to resize the result near the very end of isoreg.c,
and the solution is to make a copy of iKnots.
SEXP R_isoreg(SEXP y)
{
int n = LENGTH(y), i, ip, known, n_ip;
double tmp, slope;
SEXP yc, yf, iKnots, ans;
const char *anms[] = {"y", "yc", "yf", "iKnots", ""};
/* unneeded: y = coerceVector(y, REALSXP); */
PROTECT(ans = mkNamed(VECSXP, anms));
SET_VECTOR_ELT(ans, 0, y = y);
SET_VECTOR_ELT(ans, 1, yc = allocVector(REALSXP, n+1));
SET_VECTOR_ELT(ans, 2, yf = allocVector(REALSXP, n));
SET_VECTOR_ELT(ans, 3, iKnots= allocVector(INTSXP, n));
... calculation ...
SETLENGTH(iKnots, n_ip);
UNPROTECT(1);
return(ans);
}
But if this is the problem, I am at a bit of a loss as to what SETLENGTH
is actually for in general.
Clearly my understanding of how allocation/gc works is a bit off here,
but I can't see how else the leak may occur. Hope this is more use than
nuisance.
Simon.
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
