catch unsupported use of .I in by

Rdatatable · Dec 9, 2021 · e136746 · e136746
1 parent 3a4cfdd
commit e136746
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 6 deletions.
diff --git a/R/data.table.R b/R/data.table.R
@@ -766,10 +766,9 @@ replace_dot_alias = function(e) {
         # may evaluate to NULL | character() | "" | list(), likely a result of a user expression where no-grouping is one case being loop'd through
         bysubl = as.list.default(bysub)
         bysuborig = bysub
-
-        # support .I in by #1732
-        # use %in% to support by=.I, by=.(.I), by=c(.I) and by=list(.I)
-        if (".I" %in% bysubl) {
+        if (".I" %in% bysubl) {  #1732
+          if (!is.symbol(bysub) && (length(bysubl)!=2L || !is.symbol(bysubl[[2L]]) || !(bysubl[[1L]] %chin% c(".","c","list"))))
+            stopf("'by' contains .I but only the following are currently supported: by=.I, by=.(.I), by=c(.I), by=list(.I)")
           bysub = if (is.null(irows)) seq_len(nrow(x)) else irows
           bysuborig = as.symbol("I")
         }

diff --git a/inst/tests/tests.Rraw b/inst/tests/tests.Rraw
@@ -18688,8 +18688,8 @@ ydt[, ret := rnorm(.N)]
 f = shift
 test(2233.33, copy(ydt)[, (ycols) := shift(ret, yn, type = "lead"), by = symbol, verbose=TRUE], copy(ydt)[, (ycols) := f(ret, yn, type = "lead"), by = symbol], output="GForce optimized j to")
 
-# support by=.I #1732
-DT <- data.table(V1=1:5, V2=3:7, V3=5:1)
+# support by=.I; #1732
+DT = data.table(V1=1:5, V2=3:7, V3=5:1)
 test(2234.1, DT[, min(.SD), by=.I], setnames(DT[, min(.SD), by=1:nrow(DT)], "nrow", "I"))
 test(2234.2, DT[, min(.SD), by=.I], data.table(I=1L:5L, V1=c(1L, 2L, 3L, 2L, 1L)))
 # works also with i
@@ -18699,4 +18699,6 @@ test(2234.4, DT[c(4, NA), min(.SD), by=.I], data.table(I=c(4L, NA), V1=c(2L, NA)
 test(2234.5, DT[, min(.SD), by=.(.I)], data.table(I=1L:5L, V1=c(1L, 2L, 3L, 2L, 1L)))
 test(2234.6, DT[, min(.SD), by=list(.I)], data.table(I=1L:5L, V1=c(1L, 2L, 3L, 2L, 1L)))
 test(2234.7, DT[, min(.SD), by=c(.I)], data.table(I=1L:5L, V1=c(1L, 2L, 3L, 2L, 1L)))
+test(2234.8, DT[, min(.SD), by=.I%%2L], error="by.*contains .I.*supported")  # would be nice to support in future; i.e. by odd/even rows, and by=(.I+1L)%/%2L for pairs of rows; i.e. any expression of .I
+test(2234.9, DT[, min(.SD), by=somefun(.I)], error="by.*contains .I.*supported")