R devel - Oct 2018 - unexpected memory.limit on windows in embedded R

Dear All,
I'm linking R from another application and embedding it as described in the
R-exts manual, i.e. with initialization done via Rf_initEmbeddedR.
While everything works the same as in standalone R for Linux, under Windows
I found a difference in the default memory.limit, which is fixed to 2GB
(both win32 and win64) - compared to a limit in standalone R of 3.5GB for
win32 and 16GB on win64 (which is the amount of physical memory I have).

So I wonder if this is just undocumented (at least, I couldn't find any
mention of it) or unintentional.
All I could find is the following:
- man page of memory.limit doesn't mention a default 2GB limit, especially
not on win64;
- man page of memory.limit and rw-FAQ Q2.9 say that you can set the limit
via --max-mem-size or R_MAX_MEM_SIZE env var: both work only in standalone
R (Rf_initEmbeddedR ignores them);
- R\src\gnuwin32\system.c defines
    R_size_t R_max_memory = INT_MAX;
  which is the reason of the 2GB default limit on embedded R; however
INT_MAX is not the maximum integer value of an R_size_t variable either on
win32 nor on win64;
- line 879 of the same system.c source says: "/* set defaults for
R_max_memory. This is set here so that embedded applications get no limit
*/" (meaning that that code is executed only for standalone R) which
suggests that, at some point, the intention for embedded R was to impose no
limit by default and maybe the INT_MAX value is a mistake.

Thank you in advance,
Livio

	[[alternative HTML version deleted]]

Dear Livio,

thank you for the report. Fixed in R-devel, 75450.

Best
Tomas

On 10/01/2018 06:29 PM, Livio Bertacco wrote:
> Dear All,
> I'm linking R from another application and embedding it as described in
the
> R-exts manual, i.e. with initialization done via Rf_initEmbeddedR.
> While everything works the same as in standalone R for Linux, under Windows
> I found a difference in the default memory.limit, which is fixed to 2GB
> (both win32 and win64) - compared to a limit in standalone R of 3.5GB for
> win32 and 16GB on win64 (which is the amount of physical memory I have).
>
> So I wonder if this is just undocumented (at least, I couldn't find any
> mention of it) or unintentional.
> All I could find is the following:
> - man page of memory.limit doesn't mention a default 2GB limit,
especially
> not on win64;
> - man page of memory.limit and rw-FAQ Q2.9 say that you can set the limit
> via --max-mem-size or R_MAX_MEM_SIZE env var: both work only in standalone
> R (Rf_initEmbeddedR ignores them);
> - R\src\gnuwin32\system.c defines
>      R_size_t R_max_memory = INT_MAX;
>    which is the reason of the 2GB default limit on embedded R; however
> INT_MAX is not the maximum integer value of an R_size_t variable either on
> win32 nor on win64;
> - line 879 of the same system.c source says: "/* set defaults for
> R_max_memory. This is set here so that embedded applications get no limit
> */" (meaning that that code is executed only for standalone R) which
> suggests that, at some point, the intention for embedded R was to impose no
> limit by default and maybe the INT_MAX value is a mistake.
>
> Thank you in advance,
> Livio
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel