Le 08/06/2021 à 18:32, Martin Maechler a écrit :
Dario Strbenac
on Tue, 8 Jun 2021 09:00:04 +0000 writes:
> Good day, Indeed, changing the logical test is a
> workaround to the problem. However, a segmentation fault
> means that the software tried to access an invalid memory
> location, so I think the original problem should be
> addressed in Matrix package, regardless.
Hmm, you maybe right or not ..
Note we have the situation you (via R) ask your computer
(i.e. the OS system memory allocation routines) to provide
memory.
In a reasonable setup, the OS routine returns, saying
"I cannot provide the memory you asked for",
and the R function stop() s. .. no segfault, all is fine.
The problem that on some platforms that does not work, is a
relatively deep problem and also has happened in base R in some
cases on some platforms (possibly never on Linux based ones
(Ubuntu, Debian, Fedora, CentOS..), but maybe I'm too
optimistic there as well.
Note: I now also tried on our oldish Windows (Terminal) Server,
and it also just gave errors that it could not allocate so much
memory but did not produce a seg.fault.
Currently, I don't see what we should improve in the Matrix
package here.
Is it possible (pure hypothesis) that when such a big piece of memory is
available, some int32 counter is out of bounds?
Here, we have almost 1.e10 non-zero elements. This number is greater
than 2**31-1 (int32 limit) and even greater than unit32 limit (2**32).
Just a thought.
Best,
Serguei.
Martin Maechler
(co-maintainer of 'Matrix')
> --------------------------------------
> Dario Strbenac University of Sydney Camperdown NSW 2050
> Australia
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
