尽管存在默认方法，但 S3 通用“没有适用的方法”

Question

这很棘手，因为这个问题只发生在包的上下文中——在全局命名空间中定义时，一切都按预期工作。

我定义了一个名为

coerce_na_range()

的 S3 泛型，它有两个方法：

coerce_na_range.factor()

和

coerce_na_range.default()

。

coerce_na_range()

已导出，但两个方法均未导出。 （该函数的目的是在给定范围内用

NA

替换编码为字符或因子标签的数字。）

在全局命名空间中定义时，如果我将字符向量传递给

coerce_na_range()

，它会将其分派到

coerce_na_range.default()

并按预期工作：

vec <- c("green", "yellow", "-9", "red", "-1")
coerce_na_range(vec)
# [1] "green"  "yellow" NA       "red"    NA   

但是，如果我在新会话中加载包，它似乎会忽略默认方法：

library(lighthouse)

vec <- c("green", "yellow", "-9", "red", "-1")
coerce_na_range(vec)
# Error in UseMethod("coerce_na_range") : 
#  no applicable method for 'coerce_na_range' applied to an object of class "character"

我不认为问题是方法没有导出？例如，

tidyr:::full_seq.Date()

等不会导出，而

tidyr::full_seq()

显然可以工作。

该软件包托管在 https://github.com/ccsarapas/lighthouse。

coerce_na_range()

、其方法以及它们依赖的一些函数的代码是：

#' Suppress NA warning when coercing to numeric
#'
#' Coerces `x` to numeric. If `x` cannot be coerced, returns `NA` and suppresses
#' coercion warning.
#'
#' @export
try_numeric <- function(x) {
  if (is.factor(x)) {
    warning(
      "`x` is a factor and will be converted based on factor codes, not factor labels."
    )
  }
  withCallingHandlers(
    warning = function(w) {
      if (conditionMessage(w) == "NAs introduced by coercion") {
        rlang::cnd_muffle(w)
      }
    },
    as.numeric(x)
  )
}

#' @rdname try_numeric
#'
#' @export
try.numeric <- function(x) try_numeric(x)

#' Generate NA values of appropriate type
#'
#' Returns compatible `NA` based on `x`. This is usually of the same type as `x`
#' (e.g., `NA_real_` if `x` is a double vector). If `x` is a factor, will
#' return `NA_character_` if `factor_as_character = TRUE` (the default) and
#' `NA_integer_` otherwise.
#'
#' @export
na_like <- function(x, factor_as_character = TRUE, match_length = FALSE) {
  stopifnot("`x` must be an atomic vector" = is.atomic(x))
  type_out <- if (factor_as_character && is.factor(x)) "character" else typeof(x)
  length_out <- if (match_length) length(x) else 1L
  rep(methods::as(NA, type_out), length_out)
}


#' Set NA values based on numbers stored as strings.
#'
#' Changes values coercible to numeric in range `range_min`:`range_max` to `NA`.
#' Useful for imported SPSS files.
#'
#' @export
coerce_na_range <- function(x, ...) UseMethod("coerce_na_range")
coerce_na_range.default <- function(x, range_min = -Inf, range_max = -1) {
  coerced <- try.numeric(x)
  dplyr::if_else(
    is.na(coerced) | (coerced < range_min) | (coerced > range_max),
    x,
    na_like(x)
  )
}
coerce_na_range.factor <- function(x, range_min = -Inf, range_max = -1) {
  lvls <- levels(x)
  coerced <- try.numeric(as.character(lvls))
  lvls <- lvls[is.na(coerced) | coerced < range_min | coerced > range_max]
  factor(x, levels = lvls)
}

Answer 1

在

@export

之前添加

generic.fact

标签并运行

roxgenize

应该可以修复它：

# [...]
#' @export
coerce_na_range.factor <- function(x, range_min = -Inf, range_max = -1) {
  lvls <- levels(x)
  coerced <- try.numeric(as.character(lvls))
  lvls <- lvls[is.na(coerced) | coerced < range_min | coerced > range_max]
  factor(x, levels = lvls)
}

然后在控制台运行：

roxygen2::roxygenise()