Merge pull request #98 from mlr-org/feat/custom-hash-info

feat(calculate_hash): allow to customize hash input

NAMESPACE

@@ -10,6 +10,9 @@ S3method(discard,default)
 S3method(distinct_values,default)
 S3method(distinct_values,factor)
 S3method(distinct_values,logical)
+S3method(hash_input,"function")
+S3method(hash_input,data.table)
+S3method(hash_input,default)
 S3method(insert_named,"NULL")
 S3method(insert_named,data.frame)
 S3method(insert_named,data.table)
@@ -82,6 +85,7 @@ export(formulate)
 export(get_private)
 export(get_seed)
 export(has_element)
+export(hash_input)
 export(ids)
 export(imap)
 export(imap_chr)

R/calculate_hash.R

@@ -1,17 +1,11 @@
 #' @title Calculate a Hash for Multiple Objects
 #'
 #' @description
-#' Calls [digest::digest()] to calculate the hash for all objects provided.
-#'
-#' The following operations are performed to make hashing more robust:
-#' * If an object is a [function()], the formals and the body are hashed separately.
-#'   This ensures that the bytecode or parent environment are not be included
-#'   in the hash.
-#' * If an object is a [data.table::data.table()], the data.table is converted to a
-#'   regular list. This ensures that keys and indices are not included in the hash.
-#'
-#' Note that this only applies to top level objects, these transformations are not done
-#' recursively.
+#' Calls [digest::digest()] using the 'xxhash64' algorithm after applying [`hash_input`] to each object.
+#' To customize the hashing behaviour, you can overwrite [`hash_input`] for specific classes.
+#' For `data.table` objects, [`hash_input`] is applied to all columns, so you can overwrite [`hash_input`] for
+#' columns of a specific class.
+#' Objects that don't have a specific method are hashed as is.
 #'
 #' @param ... (any)\cr
 #'   Objects to hash.
@@ -21,13 +15,42 @@
 #' @examples
 #' calculate_hash(iris, 1, "a")
 calculate_hash = function(...) {
-  digest(lapply(list(...), function(x) {
-    if (is.function(x)) {
-      list(formals(x), as.character(body(x)))
-    } else if (is.data.table(x)) {
-      as.list(x)
-    } else {
-      x
-    }
-  }), algo = "xxhash64")
+  digest(lapply(list(...), hash_input), algo = "xxhash64")
+}
+
+#' Hash Input
+#'
+#' Returns the part of an object to be used to calculate its hash.
+#'
+#' @param x (any)\cr
+#'   Object for which to retrieve the hash input.
+#' @export
+hash_input = function(x) {
+  UseMethod("hash_input")
+}
+
+#' @describeIn hash_input
+#' The formals and the body are returned in a `list()`.
+#' This ensures that the bytecode or parent environment are not included.
+#' in the hash.
+#' @export
+hash_input.function = function(x) {
+  list(formals(x), as.character(body(x)))
+}
+
+#' @describeIn hash_input
+#' The data.table is converted to a regular list and `hash_input()` is applied to all elements.
+#' The conversion to a list ensures that keys and indices are not included in the hash.
+#' @export
+#' @method hash_input data.table
+hash_input.data.table = function(x) {
+  lapply(as.list(x), hash_input)
+}
+
+#' @describeIn hash_input
+#' Returns the object as is.
+#' @export
+hash_input.default = function(x) {
+  x
 }
+