[Rd] garbage collection, "preserved" variables, and different outcome depending on "--verbose" or not

Laurent Gautier lgautier at gmail.com
Sun Jul 20 16:49:28 CEST 2008


2008/7/20 Duncan Murdoch <murdoch at stats.uwo.ca>:
> On 20/07/2008 10:02 AM, Laurent Gautier wrote:
>>
>> I tripped on that while crafting the example.
>>
>> The problem still exists when moving the "releases" in the middle,
>> and removing the last "release".

I also see that the C code contains old/irrelevant comments at the top
of the function
lostobject. Sorry about that, the bug I am chasing is elusive...

> I can't spot any problems in the new version of your code, but I can't
> reproduce the problem, either.  So it appears to be system specific -- I was
> using the identical revision as you, but on Windows, not
> "x86_64-unknown-linux-gnu".  What exact Linux is that?

Ubuntu (hardy heron).
gcc is 4.2.3
The R source was built with "make clean; ./configure --enable-R-shlib"
before running "make; make install"

>  Can others using that system (or similar ones) reproduce it?
>
> Duncan Murdoch
>
>>
>>
>> #include <R.h>
>> #include <Rdefines.h>
>>
>>
>> SEXP createObject(void)
>> {
>>  SEXP x_R;
>>  int len_x = 1000000;
>>  PROTECT(x_R = allocVector(REALSXP, len_x));
>>  Rprintf("Created 'x' at %p\n", x_R);
>>  Rprintf("  (mode is %i, length is %i)\n", TYPEOF(x_R), LENGTH(x_R));
>>  Rprintf("  (first element is %d)\n", REAL(x_R)[0]);
>>  R_PreserveObject(x_R);
>>  UNPROTECT(1);
>>  return x_R;
>> }
>>
>> void printObject(SEXP sexp)
>> {
>>  Rprintf("object at %p\n", sexp);
>>  Rprintf("  (mode is %i, length is %i, named is %i)\n",
>>          TYPEOF(sexp), LENGTH(sexp), NAMED(sexp));
>> }
>>
>> SEXP lostobject(SEXP n_R)
>> {
>>  /*
>>   * This function will:
>>   * 1- create a numerical vector "x" and "preserve it"
>>   * 2- make call "list(x)"
>>   * 3- return "x" to R
>>   */
>>
>>
>>  SEXP x_R;
>>  int i;
>>
>>  int n = INTEGER(n_R)[0];
>>
>>  /* Create a numerical vector "x_R" */
>>
>>  for (i=0; i<n; i++) {
>>    x_R = createObject();
>>    printObject(x_R);
>>    R_ReleaseObject(x_R);
>>    R_gc();
>>  }
>>
>>  x_R = createObject();
>>  printObject(x_R);
>>  R_gc();
>>
>>  Rprintf("Returning 'x' at %p\n", x_R);
>>  Rprintf("  (first element is %d)\n", REAL(x_R)[0]);
>>  return x_R;
>> }
>>
>>
>> 2008/7/20 Duncan Murdoch <murdoch at stats.uwo.ca>:
>>>
>>> On 20/07/2008 9:01 AM, Laurent Gautier wrote:
>>>>
>>>> Dear list,
>>>>
>>>> While trying to identify the root of a problem I am having with
>>>> garbage collected variables,
>>>> I have come across the following oddity: depending on whether --verbose
>>>> is
>>>> set
>>>> or not, I obtain different results.
>>>
>>> You are working with variables without protecting them, so you just get
>>> lucky whenever the code works.
>>>
>>> More below...
>>>
>>>> I have made a small standalone example to demonstrate it.
>>>> The example is very artificial, but I had a hard time reproducing
>>>> reliably the problem.
>>>>
>>>> So when I do: (the content of test.R is at the end of this email)
>>>>
>>>> R --no-save < test.R
>>>>
>>>> [The two last lines of the output are:]
>>>>>
>>>>> x[1:3]
>>>>
>>>> [1] 0 0 0
>>>>
>>>> while with
>>>>
>>>> R --verbose --no-save < test.R
>>>>
>>>> [The two last lines of the output are:]
>>>>>
>>>>> x[1:3]
>>>>
>>>> [1] 3.733188e-317 3.137345e-317 3.137345e-317
>>>>
>>>>
>>>> The C code is compiled with:
>>>> R CMD SHLIB test_lostobject.c
>>>>
>>>>
>>>>> sessionInfo()
>>>>
>>>> R version 2.7.1 Patched (2008-07-19 r46081)
>>>> x86_64-unknown-linux-gnu
>>>>
>>>> locale:
>>>>
>>>>
>>>> LC_CTYPE=en_US.UTF-8;LC_NUMERIC=C;LC_TIME=en_US.UTF-8;LC_COLLATE=en_US.UTF-8;LC_MONETARY=C;LC_MESSAGES=en_US.UTF-8;LC_PAPER=en_US.UTF-8;LC_NAME=C;LC_ADDRESS=C;LC_TELEPHONE=C;LC_MEASUREMENT=en_US.UTF-8;LC_IDENTIFICATION=C
>>>>
>>>> attached base packages:
>>>> [1] stats     graphics  grDevices utils     datasets  methods   base
>>>>
>>>>
>>>>
>>>> ### -- file test.R
>>>>
>>>> dyn.load("test_lostobject.so")
>>>>
>>>> x = .Call("lostobject", as.integer(5))
>>>>
>>>> x[1:3]
>>>>
>>>>
>>>> ### ---
>>>>
>>>> ###--- file lostobject.c
>>>>
>>>> #include <R.h>
>>>> #include <Rdefines.h>
>>>>
>>>>
>>>>
>>>> SEXP createObject(void)
>>>> {
>>>>  SEXP x_R;
>>>>  int len_x = 1000000;
>>>>  PROTECT(x_R = allocVector(REALSXP, len_x));
>>>>  Rprintf("Created 'x' at %p\n", x_R);
>>>>  Rprintf("  (mode is %i, length is %i)\n", TYPEOF(x_R), LENGTH(x_R));
>>>>  Rprintf("  (first element is %d)\n", REAL(x_R)[0]);
>>>>  R_PreserveObject(x_R);
>>>>  UNPROTECT(1);
>>>>  return x_R;
>>>> }
>>>>
>>>> void printObject(SEXP sexp)
>>>> {
>>>>  Rprintf("object at %p\n", sexp);
>>>>  Rprintf("  (mode is %i, length is %i, named is %i)\n",
>>>>         TYPEOF(sexp), LENGTH(sexp), NAMED(sexp));
>>>> }
>>>>
>>>> SEXP lostobject(SEXP n_R)
>>>> {
>>>>  /*
>>>>  * This function will:
>>>>  * 1- create a numerical vector "x" and "preserve it"
>>>>  * 2- make call "list(x)"
>>>>  * 3- return "x" to R
>>>>  */
>>>>
>>>>
>>>>  SEXP x_R;
>>>>  int i;
>>>>
>>>>  int n = INTEGER(n_R)[0];
>>>>
>>>>  /* Create a numerical vector "x_R" */
>>>>
>>>>  for (i=0; i<n; i++) {
>>>>   x_R = createObject();
>>>>   R_ReleaseObject(x_R);
>>>
>>> At this point, the variable is unprotected, i.e. you have declared that
>>>  its
>>> memory is free for the taking.  You should not try to do anything with
>>> it.
>>>  printObject calls several functions, and one of those may have
>>> overwritten
>>> the memory.  It's not surprising that different flags (--verbose or not)
>>> result in different behaviour.
>>>
>>>>   printObject(x_R);
>>>>   R_gc();
>>>>  }
>>>>
>>>>  x_R = createObject();
>>>>  printObject(x_R);
>>>>  R_gc();
>>>>  R_ReleaseObject(x_R);
>>>
>>> Same thing here.  x_R is unprotected now, so you shouldn't use it.
>>>
>>> Duncan Murdoch
>>>
>>>>  Rprintf("Returning 'x' at %p\n", x_R);
>>>>  Rprintf("  (first element is %d)\n", REAL(x_R)[0]);
>>>>  return x_R;
>>>> }
>>>>
>>>> ______________________________________________
>>>> R-devel at r-project.org mailing list
>>>> https://stat.ethz.ch/mailman/listinfo/r-devel
>>>
>>
>> ______________________________________________
>> R-devel at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel
>
>



More information about the R-devel mailing list