Standardise on bit64 package?

[hadley]

I think it's the best of the options because you can easily convert it to doubles (losing some precision) or characters (if you want to treat as ids).

[bquinart]

From a recent thread on R-devel I understand base R will not support 64 bit integers soon. Do we know if there is a specific reason to not try to address this in base R?
I lack any insight regarding appropriateness of bit64 packake. But I would be very happy if the problem is solved within reasonable timing.

[hadley]

We can not address it in base R, and R-core doesn't seem interested, so the question is moot. All we can do is promote the best-of-bad solutions.

[krlmlr]

With bit64 it's very easy to end up with wrong values, just strip the class attribute. I was considering a dedicated S3 class that stores lossless data in an attribute and provides as.character() and perhaps as.integer64() methods.

As for the specs, they will just ensure that conversion to character() and perhaps integer64 is lossless. This gives the implementer some freedom, and we may revisit the specs later.

[hadley]

@krlmlr yes, it's a little dangerous, but a lot of work has already been done with bit64. I think it's the best of a bad set of solutions. I think coming up with our own solution is out of scope.

[krlmlr]

With the current specs, DBI backends can implement 64-bit support with the bit64 package -- all we specify is that as.numeric() gives the (approximated) value and as.character() gives the accurate value. Both is fulfilled by the bit64 implementation.