Take R types instead to map to JVM types, add check for NA to keep co…

…lumn

R/pkg/R/DataFrame.R

@@ -293,13 +293,22 @@ setMethod("colnames<-",
             dataFrame(sdf)
           })
 
+rToScalaTypes <- new.env()
+rToScalaTypes[["integer"]]   <- "integer" # in R, integer is 32bit
+rToScalaTypes[["numeric"]]   <- "double"  # in R, numeric == double which is 64bit
+rToScalaTypes[["double"]]    <- "double"
+rToScalaTypes[["character"]] <- "string"
+rToScalaTypes[["logical"]]   <- "boolean"
+
 #' coltypes
 #'
 #' Set the column types of a DataFrame.
 #'
 #' @name coltypes
 #' @param x (DataFrame)
-#' @return value (character) A character vector with the target column types for the given DataFrame
+#' @return value (character) A character vector with the target column types for the given
+#'    DataFrame. Column types can be one of integer, numeric/double, character, logical, or NA
+#'    to keep that column as-is.
 #' @rdname coltypes
 #' @aliases coltypes
 #' @export
@@ -309,7 +318,8 @@ setMethod("colnames<-",
 #' sqlContext <- sparkRSQL.init(sc)
 #' path <- "path/to/file.json"
 #' df <- jsonFile(sqlContext, path)
-#' coltypes(df) <- c("string", "integer")
+#' coltypes(df) <- c("character", "integer")
+#' coltypes(df) <- c(NA, "numeric")
 #'}
 setMethod("coltypes<-",
           signature(x = "DataFrame", value = "character"),
@@ -321,7 +331,15 @@ setMethod("coltypes<-",
             }
             newCols <- lapply(seq_len(ncols), function(i) {
               col <- getColumn(x, cols[i])
-              cast(col, value[i])
+              if (!is.na(value[i])) {
+                stype <- rToScalaTypes[[value[i]]]
+                if (is.null(stype)) {
+                  stop("Only atomic type is supported for column types")
+                }
+                cast(col, stype)
+              } else {
+                col
+              }
             })
             nx <- select(x, newCols)
             dataFrame(nx@sdf)

R/pkg/inst/tests/test_sparkSQL.R

@@ -612,14 +612,20 @@ test_that("coltypes() set the column types", {
   expect_equal(dtypes(df), list(c("name", "string"), c("age", "decimal(24,2)")))
 
   df1 <- select(df, cast(df$age, "integer"))
-  coltypes(df) <- c("string", "integer")
+  coltypes(df) <- c("character", "integer")
   expect_equal(dtypes(df), list(c("cast(name as string)", "string"), c("cast(age as int)", "int")))
   value <- collect(df[, 2])[[3, 1]]
   expect_equal(value, collect(df1)[[3, 1]])
   expect_equal(value, 22)
 
-  expect_error(coltypes(df) <- c("string"),
+  coltypes(df) <- c(NA, "numeric")
+  expect_equal(dtypes(df), list(c("cast(name as string)", "string"),
+              c("cast(cast(age as int) as double)", "double")))
+
+  expect_error(coltypes(df) <- c("character"),
                "Length of type vector should match the number of columns for DataFrame")
+  expect_error(coltypes(df) <- c("environment", "list"),
+               "Only atomic type is supported for column types")
 })
 
 test_that("head() and first() return the correct data", {