Estimate CV error of R's lda using the package cvTools - r

Question

I'm trying to use the package cvTools to estimate the classification error of a LDA fit. I've used cvTools successfully in the past with other models, but with LDA it doesn't work as expected.
data <- rbind(data.frame(cls='A', x=rnorm(10, mean=5)),
data.frame(cls='B', x=rnorm(10, mean=10)))
cv.fit <- cvFit(
lda,
formula=cls ~ x,
data=data,
cost=function(true, pred) {
mean(true != pred)
},
K=10
)
It seems cvTools uses the generic predict model function internally. predict.lda doesn't return a simple vector of class labels, but rather a list with attributes class, posterior and x.
If I set a debug breakpoint in the cost function I see that true is a vector of class labels but pred is said list converted to a vector.
So my question is how I can still use cvTools with LDA.
My first idea was to provide a new predict method by subclassing lda in the hope that cvTools will then call my predict function which returns a vector of class labels:
lda.fit <- lda(cls ~ x, data)
class(lda.fit) <- c('ldaCV', class(lda.fit))
predict.ldaCV <- function(m, newdata) {
MASS:::predict.lda(m, newdata)$class
}
cv.fit <- cvFit(
lda.fit,
data=data,
y=data$cls,
cost=function(true, pred) {
mean(true != pred)
},
K=10
)
Nothing changes however. pred is still the same messed up list.