# loop over categorical variables
for (var in cat_vars) {
  # get definitions of category-levels
  labels <- ship_meta[get("VAR_NAMES") == var, get("VALUE_LABELS")]
  split_list <- strsplit(
    x = labels,
    split = " | ",
    fixed = TRUE
  )
  
  # extract the allowed values for each categorical dataelement
  allowed_values <- strsplit(
    x = split_list[[1]],
    split = " =",
    fixed = TRUE
  )
  
  # transform allowed values into a "value_set" array, which is stored
  # as a JSON object (required by DQAstats)
  c <- jsonlite::toJSON(
    list(
      "value_set" = sapply(
          X = allowed_values,
          FUN = function(x) {
            return(x[[1]])
          })
    ),
    pretty = TRUE
  )
  # add constraints for this dataelement to the DQAstats MDR
  mdr[get("source_variable_name") == var, ("constraints") := c]
}

# for comparison, show metadata definition from dataquieR ...
# dataquieR: string, split by pipes '|'
ship_meta[get("VAR_NAMES") == "smoking", get("VALUE_LABELS")]
#> [1] "0 = nonsmoker | 1 = former smoker | 2 = smoker"

# ... and DQAstats: JSON
mdr[get("source_variable_name") == "smoking", get("constraints")]
#> [1] "{\n  \"value_set\": [\"0\", \"1\", \"2\"]\n}"
mdr[get("source_variable_name") == "smoking", get("constraints")] %>%
  jsonlite::fromJSON()
#> $value_set
#> [1] "0" "1" "2"

# get names of continuous variables (float/integer)
cont_vars <- mdr[
  get("variable_type") %in% c("integer", "float"),
  unique(get("source_variable_name"))
]

# loop over continuous dataelements
for (var in cont_vars) {
  # extract constraints of continuous dataelement
  labels <- ship_meta[get("VAR_NAMES") == var, get("HARD_LIMITS")]
  split_list <- strsplit(
    x = labels,
    split = ";",
    fixed = TRUE
  )

  # transform these constraints to a list, with entries "min" and "max"
  constraints <- lapply(
    X = split_list,
    FUN = function(x) {

      if (!is.na(x[[1]])) {
        return(
          list(
            "min" = as.numeric(gsub("[[:punct:]]", "", x[[1]])),
            "max" = as.numeric(gsub("[[:punct:]]", "", x[[2]]))
          )
        )
      } else {
        return("error")
      }
    }
  )

  # save list as nested JSON within the key "range" (required by DQAstats)
  if (length(constraints[[1]]) > 1) {
    c <- jsonlite::toJSON(
      list("range" = constraints[[1]]),
      pretty = TRUE,
      auto_unbox = TRUE
    )
    # write constraint for this dataelement to DQAstats MDR
    mdr[get("source_variable_name") == var, ("constraints") := c]
  }
}

# The `id` variable should actually not be treated as a categorical variable,
# since the number of potential categories is (theoretically) unlimited and
# an analysis of the categories in this case is seldom meaningful.
# Nevertheless, we could perform quality checks on the `id` variable, e.g. to
# check for a required formatting of the values.
# This can be achieved with DQAstats by changing the variable type to "string"
# and use a regular expression that checks for the expected format.

# change variable type to "string"
mdr[get("source_variable_name") == "id", ("variable_type") := "string"]

# add regular expression for the dataelement "id"
mdr[get("source_variable_name") == "id", ("constraints") := jsonlite::toJSON(
  list("regex" = "^[[:digit:]]{1,5}$"),
  pretty = TRUE,
  auto_unbox = TRUE
)]

# add plausibilities
# sex and contraception
p <- jsonlite::toJSON(list(
  "atemporal" = list(
    "contraception" = list(
      "name" = "A_present_and_B_vv",
      "description" = "Contracept in males.",
      "join_crit" = "id",
      "filter" = list(
        "ship" = "^1$"
      ),
      "constraints" = list(
        "value_set" = list(
          "ship" = "1"
        )
      )
    )
  )
),
  pretty = TRUE,
  auto_unbox = TRUE
)

# write plausibility to mdr
mdr[
  source_variable_name == "sex" &
    dqa_assessment == 1,
  plausibility_relation := p
]

p <- jsonlite::toJSON(list(
  "atemporal" = list(
    "diab_age" = list(
      "name" = "A_present_and_B_levels_vl",
      "description" = "Diab age but no diab.",
      "join_crit" = "id",
      "filter" = list(
        "ship" = "^(?!.*(99900|99901|99801)).*$"
      ),
      "constraints" = list(
        "value_set" = list(
          "ship" = "1"
        )
      )
    )
  )),
  pretty = TRUE,
  auto_unbox = TRUE
)
mdr[
  source_variable_name == "diab_known" &
    dqa_assessment == 1,
  plausibility_relation := p
]

p <- jsonlite::toJSON(list(
  "uniqueness" = list(
    "sex" = list(
      "name" =  "Every ID is associated with one Sex",
      "description" = paste0("With each distinct value of 'id', ",
      "only one value of 'sex' may be associated."
    )
  ))),
  pretty = TRUE,
  auto_unbox = TRUE
)
mdr[
  source_variable_name == "id" &
    dqa_assessment == 1,
  plausibility_relation := p
]