ARROW-14844 [R] Implement decimal256()

[dragosmg]

No description provided.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-14844

[github-actions]

⚠️ Ticket has not been started in JIRA, please click 'Start Progress'.

[romainfrancois]

in addition: dragosmg#1

[romainfrancois]

Perhaps decimal() and decimal256 could call the same function for argument checking, e.g.

check_decimal_args <- function(precision, scale) {
  precision <- vec_cast(precision, to = integer())
  vec_assert(precision, size = 1L)

  scale <- vec_cast(scale, to = integer())
  vec_assert(scale, size = 1L)

  list(precision = precision, scale = scale)
}

#' @rdname data-type
#' @export
decimal <- function(precision, scale) {
  args <- check_decimal_args(precision, scale)
  Decimal128Type__initialize(args$precision, args$scale)
}

#' @rdname data-type
#' @export
decimal256 <- function(precision, scale) {
  args <- check_decimal_args(precision, scale)
  Decimal256Type__initialize(args$precision, args$scale)
}

[dragosmg]

Great idea! I will implement it.

[romainfrancois]

As with decimal128 this would not allow for creating arrays directly:

library(arrow, warn.conflicts = FALSE)
#> See arrow_info() for available features

Array$create(c(1, 2), type = decimal(4, 3))
#> Error: NotImplemented: Extend

^{Created on 2021-12-03 by the reprex package (v2.0.1.9000)}

The error comes from this class in r_to_arrow.cpp

template <typename T>
class RPrimitiveConverter<T, enable_if_t<is_decimal_type<T>::value>>
    : public PrimitiveConverter<T, RConverter> {
 public:
  Status Extend(SEXP x, int64_t size, int64_t offset = 0) override {
    return Status::NotImplemented("Extend");
  }
};

Maybe we could catch the NotImplemented error, then convert to a guessed type, and then cast.

[romainfrancois]

Perhaps we could have something like this on the R side:

Array$create <- function(x, type = NULL) {
  if (!is.null(type)) {
    type <- as_type(type)
  }
  if (inherits(x, "Scalar")) {
    out <- x$as_array()
    if (!is.null(type)) {
      out <- out$cast(type)
    }
    return(out)
  }
  tryCatch(
    vec_to_Array(x, type),
    error = function(cnd) {
      if (!is.null(type)) {
        # try again and then cast
        vec_to_Array(x, NULL)$cast(type)
      } else {
        signalCondition(cnd)
      }
    }
  )
}

So that:

library(arrow, warn.conflicts = FALSE)
#> See arrow_info() for available features

Array$create(c(1, 2), type = decimal(4, 3))
#> Array
#> <decimal128(4, 3)>
#> [
#>   1.000,
#>   2.000
#> ]

^{Created on 2021-12-03 by the reprex package (v2.0.1.9000)}

[dragosmg]

This is outdated. Work on decimal256() is taking place in a different PR - #11898