integer64 interaction about nanotime

[lsilvest]

Looks like the last issue has nothing to do with subsetting. Also, it's really an issue with the underlying integer64 implementation:

integer64(0) > 0
## Error in if (n2%%n1) warning("length(e2) not a multiple length(e1)") : 
##  argument is not interpretable as logical

We also have another problem with zero length vectors, and again we just inherit the integer64 issues:

integer64(0)[TRUE]
## integer64
## [1] 9218868437227407266

instead of:

integer(0)[TRUE]
[1] NA

We might be able to fix the above issues with if statements, but it's not a very satisfactory solution.

[eddelbuettel]

Thanks for this -- I had seen the same error message at times when I created zero-row result sets for which I requested a nanotime column.

@lsilvest Do you have a quick idea where this may be from?

[wsrfad]

format.nanotime <- function(x, format="", tz="", ...)
{
    format <- .getFormat(format)
    tz <- .getTz(x, tz)
    bigint <- as.integer64(x)
    secs  <- as.integer64(bigint / as.integer64(1000000000))
    nanos <- bigint - secs * as.integer64(1000000000)
    res <- RcppCCTZ::formatDouble(as.double(secs), as.double(nanos), fmt=format, tgttzstr=tz)
    res[is.na(x)] <- as.character(NA)
    n = names(x)
    if (!is.null(n)) {
        names(res) <- n  						## #nocov
    }
    res
}```    

This error happens when we do binary operations on integer64 with another integer64 which is empty. In R, integer(0) / integer(1) gives numeric(0), while integer64(0) / integer64(1) gives error. This issue is creating problems at many places

[eddelbuettel]

Yes, traceback() lead me there too but I don't quite know yet what I want the format() of an empty nanotime object should be.

[eddelbuettel]

Try the bugfix/issue35 branch I just committed -- format.nanotime() now tests for an empty argument.

[wsrfad]

This is how Date does it ...

## Scenario 1 : Empty vector in constructor
> d = character()
> as.Date(d)
[1] "Date of length 0"

## Scenario 2 : Subsetting based on empty vector of dates should also be supported
> dates = as.Date(c("2017-10-02", "2017-10-03"))
> dates[dates == as.Date("2017-10-04")]
[1] "Date of length 0"
> dates[dates == as.Date("2017-10-04")]
[1] "Date of length 0"

Does your fix support the second scenario also ?

> times = nanotime(integer64())
> times = times[times > nanotime(100) ]
> print(times)

Without your fix, at least, this case is problematic:

> times = nanotime(integer64())
> times = times[times > nanotime(100) ]
Error in callNextMethod() : 
  error in evaluating a 'primitive' next method: Error in if (n2%%n1) warning("length(e2) not a multiple length(e1)") : 
  argument is not interpretable as logical

[eddelbuettel]

Made a second commit.

[wsrfad]

The scenario 2 still fails, with bug fix !

> times = nanotime(integer64())
> times
character(0)
> times = times[times > nanotime(100) ]
Error in callNextMethod() : 
  error in evaluating a 'primitive' next method: Error in if (n2%%n1) warning("length(e2) not a multiple length(e1)") : 
  argument is not interpretable as logical

[eddelbuettel]

Patches welcome.

[wsrfad]

I am not very well versed with all the features in nanotime, as also oldClass etc keywords. I would have loved to contribute...But will take me much more time. I am taking it as a learning exercise. I had tried few things before reaching out. Few workarounds include checking for size of vector, before subsetting. But actual fix (in my opinion) should be around integer64 / RcppTz / (not sure what, though). Hope you guys can help, I will learn and then next time sure I will be able to contribute ! I will be happy to tell you test cases though, for now

[eddelbuettel]

Maybe @lsilvest has an idea on that one; we did get two warts taken care of so some progress made.

[eddelbuettel]

We can leave this open. Mayne we want to retitle this as it just the subsetting from integer64 that bites.

I merged the two pending branches I had -- this one and another one with two added constructor improvements.

Maybe time for a quick 0.2.1 release.

[wsrfad]

I do not wish to raise issues, where people do not appreciate or welcome feedback !

[eddelbuettel]

Can you please calm down over your non-reproducible issue #39?

This issue here was addressed over two weeks ago with this CRAN release as I mentioned.