tricky (edge) case in dcast.data.table fixed. closes #715.

R/fcast.R

@@ -66,6 +66,8 @@ dcast.data.table <- function(data, formula, fun.aggregate = NULL, ..., margins =
     fill.default <- NULL
     if (!is.null(fun.aggregate)) { # construct the 'call'
         fill.default = fun.aggregate(data[[value.var]][0], ...)
+        if (!length(fill.default) && (is.null(fill) || !length(fill)))
+            stop("Aggregating function provided to argument 'fun.aggregate' should always return a length 1 vector, but returns 0-length value for fun.aggregate(", typeof(data[[value.var]]), "(0)).", " This value will have to be used to fill missing combinations, if any, and therefore can not be of length 0. Either override by setting the 'fill' argument explicitly or modify your function to handle this case appropriately.")
         args <- c("data", "formula", "margins", "subset", "fill", "value.var", "verbose", "drop")
         m <- m[setdiff(names(m), args)]
         .CASTfun = fun.aggregate # issues/713
@@ -107,7 +109,7 @@ dcast.data.table <- function(data, formula, fun.aggregate = NULL, ..., margins =
             attr(oo, 'maxgrpn') > 1L
         }
         if (!fun.null && fun_agg_chk(data))
-            stop("Aggregating function provided to argument 'fun.aggregate' should return a length 1 vector for each group, but returns length != 1 for atleast one group. Please have a look at the DETAILS section of ?dcast.data.table ")
+            stop("Aggregating function provided to argument 'fun.aggregate' should always return a length 1 vector for each group, but returns length != 1 for atleast one group. Please have a look at the DETAILS section of ?dcast.data.table ")
     } else {
         if (is.null(subset))
             data = data[, unique(c(ff_, value.var)), with=FALSE] # data is untouched so far. subset only required columns

README.md

@@ -228,6 +228,8 @@ We moved from R-Forge to GitHub on 9 June 2014, including history.
 
   34.  `dcast.data.table` handles `fun.aggregate` argument properly when called from within a function that accepts `fun.aggregate` argument and passes to `dcast.data.table()`. Closes [#713](https://github.com/Rdatatable/data.table/issues/713). Thanks to mathematicalcoffee for reporting [here](http://stackoverflow.com/q/24542976/559784) on SO.
 
+  34.  `dcast.data.table` now returns a friendly error when fun.aggregate value for missing combinations is 0-length, and 'fill' argument is not provided. Closes [#715](https://github.com/Rdatatable/data.table/issues/715)
+
 #### NOTES
 
   1.  Reminder: using `rolltolast` still works but since v1.9.2 now issues the following warning:  

inst/tests/tests.Rraw

@@ -4679,7 +4679,7 @@ test(1313.22, DT[, list(y=max(y, na.rm=TRUE)), by=x], DT[c(5,10)])
 
 # bug git #693 - dcast.data.table error message improvement:
 dt <- data.table(x=c(1,1), y=c(2,2), z = 3:4)
-test(1314, dcast.data.table(dt, x ~ y, value.var="z", fun.aggregate=identity), error="Aggregating function provided to argument 'fun.aggregate' should return a length 1")
+test(1314, dcast.data.table(dt, x ~ y, value.var="z", fun.aggregate=identity), error="Aggregating function provided to argument 'fun.aggregate' should always return a length 1 vector")
 
 # bug #688 - preserving attributes
 DT = data.table(id = c(1,1,2,2), ty = c("a","b","a","b"), da = as.Date("2014-06-20"))
@@ -4855,9 +4855,9 @@ test(1344, fread("A,B\n1,T\n2,NA\n3,"), data.table(A=1:3, B=c(TRUE,NA,NA)))
 # issues/713 - dcast.data.table and fun.aggregate
 DT <- data.table(id=rep(1:2, c(3,4)), k=c(rep(letters[1:3], 2), 'c'), v=1:7)
 foo <- function (tbl, fun.aggregate) {
-    dcast.data.table(tbl, id ~ k, value.var='v', fun.aggregate=fun.aggregate)
+    dcast.data.table(tbl, id ~ k, value.var='v', fun.aggregate=fun.aggregate, fill=NA_integer_)
 }
-test(1345, foo(DT, last), dcast.data.table(DT, id ~ k, value.var='v', fun.aggregate=last))
+test(1345, foo(DT, last), dcast.data.table(DT, id ~ k, value.var='v', fun.aggregate=last, fill=NA_integer_))
 
 # more minor changes to dcast.data.table (subset argument handling symbol - removing any surprises with data.table's typical scoping rules) - test for that.
 DT <- data.table(id=rep(1:2, c(3,4)), k=c(rep(letters[1:3], 2), 'c'), v=1:7)
@@ -4867,6 +4867,11 @@ test(1346.1, dcast.data.table(DT, id ~ k, value.var="v", subset=.(c(TRUE, rep(FA
 DT[, bla := !bla]
 test(1346.2, dcast.data.table(DT, id ~ k, value.var="v", subset=.(bla), fun.aggregate=length), dcast.data.table(DT[(bla)], id ~ k, value.var="v", fun.aggregate=length))
 
+# issues/715
+DT <- data.table(id=rep(1:2, c(3,2)), k=c(letters[1:3], letters[1:2]), v=1:5)
+test(1347.1, dcast.data.table(DT, id ~ k, fun.aggregate=last, value.var="v"), error="Aggregating function provided to argument 'fun.aggregate' should always return a length 1 vector")
+test(1347.2, dcast.data.table(DT, id ~ k, fun.aggregate=last, value.var="v", fill=NA_integer_), data.table(id=1:2, a=c(1L, 4L), b=c(2L,5L), c=c(3L,NA_integer_), key="id"))
+
 ##########################