[Rd] [R] New errors with difftime()-objects in 2.11.1 (was Re: Request: difftime method for cut())

Gustaf Rydevik gustaf.rydevik at gmail.com
Wed Jun 23 10:57:56 CEST 2010


On Wed, Jun 23, 2010 at 7:13 AM, Peter Dalgaard <pdalgd at gmail.com> wrote:
> Gustaf Rydevik wrote:
>
>> Oh, I forgot to mention that the workaround of using as.double (or
>> as.numeric) works fine, and I've done that.
>> It's just that it can take quite a while (as in several hours) to
>> figure out that the reason for the error is that you have to force
>> difftime objects to be numeric in 2.11.1, when the code's been working
>> fine before and the error messages are obscure.
>
> I don't think you realize the problems that could occur by assuming that
> difftime objects are numerics ON ANY PARTICULAR SCALE!
>
> --
> Peter Dalgaard
> Center for Statistics, Copenhagen Business School
> Phone: (+45)38153501
> Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com
>


Ah. Yes, you're right that it would be problematic to say the least to
assume that the difftime object is measured in days and not in, say,
seconds. And I suppose that it makes sense to prioritize avoiding
calculations that give misleading results over forcing changes in old
code.  I was just caught somewhat unprepared, and I know that my
colleagues  who is not quite as R-literate will be even more
unprepared for old stuff no longer working.
Usually, R prepares the user for these kind of things by throwing
warnings a version or two before the change is actually implemented.
But I guess that's not always practical.

I take it that your argument would also work agains implementing
simple difftime-methods of functions as well, where you force difftime
objectws to be numeric? In that case, people can disregard my
suggestion of adding a difftime-method to cut().

Anyhow, I'll stop whining now. Thanks for the good work you're doing
in the R Core team.

Regards,
Gustaf

-- 
Gustaf Rydevik, M.Sci.
tel: +46(0)703 051 451
address:Essingetorget 40,112 66 Stockholm, SE
skype:gustaf_rydevik



More information about the R-devel mailing list