Re: [Rd] Issue with memory deallocation/fragmentation on systems which use glibc

[Dmitriy Selivanov]

Dear R-devel mailing list and especially R-core, is there any chance to
receive feedback on issue I described in previous emails? I would consider
such behaviour as a bug.

As a work-around I've created small "clean-up" function:

malloc_trim_finalizer = function(e) {
  res = NULL
  if(R.version$os == "linux-gnu") {
flog.debug("Calling malloc_trim(0L) to trigger glibc to release
memory\n")
res = malloc_trim(0L)
  }
  res
}

And at the end of each function which produce a lot of intermediate small
objects I provide it to reg.finalizer():
some_function = function(...) {
  # do some useful work
  result = TRUE
  # register finalizer
  e = environment()
  reg.finalizer(e, malloc_trim_finalizer)
  return(result)
}

2017-06-22 11:12 GMT+04:00 Dmitriy Selivanov :

> A few additional details. According to Linux Programmer's Manual
>
>1. http://man7.org/linux/man-pages/man3/mallopt.3.html
>2. http://man7.org/linux/man-pages/man3/malloc_trim.3.html
>
> And if I understood everything correctly `free` could trigger
> `malloc_trim` based on value of several environment variables -
> MALLOC_TOP_PAD_ and MALLOC_TRIM_THRESHOLD_. However setting them as low as
> 1 or 0 doesn't have any effect (but as I wrote in previous email manual
> call of `malloc_trim` helps to release memory).
>



-- 
Regards
Dmitriy Selivanov

[[alternative HTML version deleted]]

__
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel

[Rd] MASS:::dropterm.glm() and MASS:::addterm.glm() should use ... for extractAIC()

[Marc Girondot via R-devel]

Here is a change required from MASS:::dropterm.glm() and 
MASS:::addterm.glm().


Thanks

Marc


The stepAIC() function from package MASS uses extractAIC() to get the 
AIC from a model.

Several methods exist:
extractAIC.glm() for example, some in MASS packages and some in stats 
package.


The parameters for extractAIC() are:
fit, scale, k = 2, ...

The ... are not used in most of the extractAIC.xxx() methods, from 
example in stats:::extractAIC.glm() or MASS:::extractAIC.loglm() but its 
presence could be necessary if extractAIC() is changed to use for 
example to use AICc rather than AIC.


Within stepAIC(), extractAIC() uses always the ... parameter. So all is 
ok for that.


However, stepAIC() uses dropterm() or addterm().

Within MASS:::dropterm.glm() and MASS:::addterm.glm(), extractAIC() is 
also used but without the ... parameter.


It prevents to use new version of extractAIC() that could use this 
parameter.


The solution is simple:
In MASS:::dropterm.glm(), line 60 and MASS:::addterm.glm(), line 84:
aic <- aic + (extractAIC(object, k = k)[2L] - aic[1L])
must be changed to
aic <- aic + (extractAIC(object, k = k, ...)[2L] - aic[1L])

Other dropterm.xxx() and addterm.xxx() do not suffer this problem.

__
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel