Bug: --DTC variable coercion to character

[adchan11]

What happened?

In reference to #145 👍

It would be great to have both. --DTC variables coerced to chr when they have an other type and no message when there is no coercion.

Session Information

No response

Reproducible Example

No response

[cpiraux]

There is a coercion issue when a variable is a numeric date in data frame (df) and has a character type in metadata.

The first_class() function assigns a character type to date(time) variables. @elimillera @bms63, do you remember why date variables are considered as characters?

first_class <- function(x) {
  characterTypes <- getOption("xportr.character_types")
  class_ <- tolower(class(x)[1])
}
table_cols_types <- map(.df, first_class)

Original class:

Currently, characterTypes has the following values: "character", "char", "text", "date", "posixct", "posixt", "datetime", "time", "partialdate", "partialtime", "partialdatetime", "incompletedatetime", "durationdatetime", "intervaldatetime". It is used in xport_type() to convert both R data type (type.x) and metadata data type (type.y) to numeric or character.

meta_ordered <- left_join(
  data.frame(variable = names(.df), type = unlist(table_cols_types)),
  metadata,
  by = "variable"
) %>%
  mutate(
    # `_character` is used here as a mask for character, in case someone doesn't
    # want 'character' coerced to character
    type.x = if_else(type.x %in% characterTypes, "_character", type.x),
    type.x = if_else(type.x %in% numericTypes | (grepl("DT$|DTM$|TM$", variable) & !is.na(format)),
      "_numeric",
      type.x
    ),
    type.y = if_else(is.na(type.y), type.x, type.y),
    type.y = tolower(type.y),
    type.y = if_else(type.y %in% characterTypes | (grepl("DTC$", variable) & is.na(format)), "_character", type.y),
    type.y = if_else(type.y %in% numericTypes, "_numeric", type.y)
  )

I don’t think we should use the same list of character types for R data type and metadata type for date(time) variables. When the variable is a numeric date the R data types is "posixct", "posixt", "date" or "time". When the metadata data type is "date", "datetime", "time", then it is a character variable. Considering this difference, we should not use the same values list to convert R data type and metadata data type to numeric or character.
What are your thoughts about that?