[datatable-help] A new package int64 to support 64-bit integer in R

[Matthew Dowle]

Hi,

Saw that too. Had a look through the history to remind myself.

Seems that Damian was looking for it :
http://r.789695.n4.nabble.com/Merging-where-key-is-large-tp2315272p2315272.html

and Leon :
http://r.789695.n4.nabble.com/Non-integer-key-tp3743732p3743732.html

There doesn't seem to be an FR in data.table related to this though.

int64 indeed looks very useful. It seems to do it internally using 2
regular int vectors, and that causes data.table an issue because it
doesn't currently accept types that are not atomic (such as POSIXlt).
That could be the FR then: allowing multi-column types (for which int64
is worth it, but not POSIXlt)?

There are a few related questions on SO :
http://stackoverflow.com/questions/8175965/multiplication-of-large-integers-in-r/8212452#8212452
http://stackoverflow.com/questions/2053397/long-bigint-decimal-equivalent-datatype-in-r/8212428#8212428

and those mention some alternatives :
* larger and more accurate integers can be stored in numeric (double)
* Package Brobdingnag
* Package gmp

I don't know how int64 compares to those, or if they use multi-columns
internally or not, or what.

Btw, does anyone know what sponsorship by Google actually means in
practice?

Matthew


On Mon, 2011-11-21 at 14:56 -0500, Steve Lianoglou wrote:
> Hi Branson,
> 
> This also hit my radar earlier today.
> 
> Just wanted to say:
> 
> On Mon, Nov 21, 2011 at 2:02 PM, Branson Owen <branson.owen at gmail.com> wrote:
> 
> > A immediate drawback I see is that they use S4 class. I know it's painful to
> > deal with S4.
> 
> S4 isn't so painful ;-)
> 
> This sounds like it could be interesting, though ... will have to
> check out that package.
> 
> -steve
>