rowSums在一列data.frame上中断，原因是在library（reshape）中进行强制转换生成了-为什么？

Question

Here's some code -- I don't understand why rowSums breaks when it does, and why changing the class of test3 fixes it. This seems bizarre. Is it a bug? If not, what am I missing?

library(reshape)

test <- data.frame(x=1:5)
rowSums(test)  # Works
class(test)  # "data.frame"

## Trying to break rowSums; see below
class(test) <- c("data.frame", "cast_df")
rowSums(test)  # Works, but see below

vars <- c("x", "y", "z")
test2 <- data.frame(label=rep(c("A", "B"), c(3, 3)),
                    variable=rep(vars, 2),
                    value=1:6)

test2.cast <- cast(test2, label ~ variable, value="value")

rowSums(test2.cast[, vars])  # Works
rowSums(test2.cast[, "y"])  # Breaks because of drop=TRUE

test3 <- test2.cast[, "y", drop=FALSE]
rowSums(test3)  # Why does this break?
class(test3)  # c("cast_df", "data.frame")
class(test3) <- "data.frame"
rowSums(test3)  # Works!  Why?

Context: I have a large data frame generated by cast. I then programmatically call rowSums on several different subsets of columns of that data frame -- and I noticed this behavior when the subset had only one column. This is a small reproducible example.

Answer 1

rowSums calls as.matrix . The first two lines of as.matrix.cast_df are:

ids <- attr(x, "idvars")
mat <- as.matrix.data.frame(x[, setdiff(names(x), ids)])

Note the lack of drop = TRUE there. When you reset the class to just data.frame this method isn't called, instead the regular as.matrix.data.frame method is called directly, and there's no problem.