R devel - Jan 2004 - unstability when using isoreg() function (PR#6494)

Full_Name: Petr Klasterecky
Version: 1.8.1
OS: Windows XP, Linux
Submission from: (NULL) (195.113.27.212)


The isoreg() function causes R to crash when called repeatedly. Consider the
following simple script:

{
library(modreg)
N <- 10
x <- rnorm(N)
print("Original x values:")
print(x)
for(n in (1:N)){print(y <- isoreg(x[1:n])$yf)}
}        

I am able to run (call) it several times, let say 6 or 10 times, and when trying
once more, R crashes. In Windows a screen with "The R GUI software has
caused
errors and will be terminated by windows. All unsaved information will be
lost."
message appears, using Linux R crashes with a mesage "Segmentation
fault".

On Wed, 28 Jan 2004 17:31:27 +0100 (CET), you wrote:
>Full_Name: Petr Klasterecky
>Version: 1.8.1
>OS: Windows XP, Linux
>Submission from: (NULL) (195.113.27.212)
>
>
>The isoreg() function causes R to crash when called repeatedly. Consider the
>following simple script:
>
>{
>library(modreg)
>N <- 10
>x <- rnorm(N)
>print("Original x values:")
>print(x)
>for(n in (1:N)){print(y <- isoreg(x[1:n])$yf)}
>} 
I get the crash reproducibly in r-devel after I put "set.seed(2)"
ahead of the sample code above.  (Of course, r-devel doesn't need the
library(modreg), but that doesn't affect the bug.)  Other seeds and
the code works without error.

I tried tracing into R_isoreg to try to spot the problem, but the
error occurs after it returns.  It has all the symptoms of some code
writing outside allocated bounds, but I couldn't see where that
happened.

Duncan Murdoch

> >{
> >library(modreg)
> >N <- 10
> >x <- rnorm(N)
> >print("Original x values:")
> >print(x)
> >for(n in (1:N)){print(y <- isoreg(x[1:n])$yf)}
> >} 
> 
> I get the crash reproducibly in r-devel after I put "set.seed(2)"
> ahead of the sample code above.  (Of course, r-devel doesn't need the
> library(modreg), but that doesn't affect the bug.)  Other seeds and
> the code works without error.
Dear sir,
Thank you for a swift reply.

Yes, but I don't think that seed 2 is the only one producing the crash,
since it occured quite randomly in my testing. For this seed=2 the problem
appeared for isoreg(x[1:3]) and nowhere else, I have tried it just now.
But, after restarting R, I
was able to run isoreg(x[1:4]) through isoreg(x[1:10]), then
isoreg(x[1:3]) and only after calling isoreg(x[1:3]) again R crashed - so
even when I knew there was a problem in isoreg(x[1:3]), it worked (once).

Please let me know if I could do anything else/more to help in tracking
the problem.

Best regards,
Petr Klasterecky

iKnots was allocated one element too short.  This only overruns if the 
sequence is monotone, as happens here for set.seed(2), N=3.  Fixed in 
R-devel.

The line 39 in isoreg.c needs to be

    SET_VECTOR_ELT(ans, 3, iKnots= allocVector(INTSXP, n));


Duncan was trying to give a reproducible example, Petr, as we do ask you 
to do in the FAQ.

On Thu, 29 Jan 2004 dmurdoch@pair.com wrote:
> On Wed, 28 Jan 2004 17:31:27 +0100 (CET), you wrote:
> 
> >Full_Name: Petr Klasterecky
> >Version: 1.8.1
> >OS: Windows XP, Linux
> >Submission from: (NULL) (195.113.27.212)
> >
> >
> >The isoreg() function causes R to crash when called repeatedly.
Consider the
> >following simple script:
> >
> >{
> >library(modreg)
> >N <- 10
> >x <- rnorm(N)
> >print("Original x values:")
> >print(x)
> >for(n in (1:N)){print(y <- isoreg(x[1:n])$yf)}
> >} 
> 
> I get the crash reproducibly in r-devel after I put "set.seed(2)"
> ahead of the sample code above.  (Of course, r-devel doesn't need the
> library(modreg), but that doesn't affect the bug.)  Other seeds and
> the code works without error.
> 
> I tried tracing into R_isoreg to try to spot the problem, but the
> error occurs after it returns.  It has all the symptoms of some code
> writing outside allocated bounds, but I couldn't see where that
> happened.
> 
> Duncan Murdoch
> 
> ______________________________________________
> R-devel@stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-devel
> 
> 
-- 
Brian D. Ripley,                  ripley@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