Predict.lm not working for matrix data

gistfile1.r

# Make a dataset where Y is cubic in X, plus noise:
n <- 1000
x <- runif(n)
x <- sort(x)
eps <- rnorm(n)
beta <- c(1, 2, 3)
X <- poly(x, 3, raw = TRUE) 
y <- X %*% beta + eps

# Fit the data
data <- data.frame(X = X, y = y)

# This formula will not work for predicting, predictions will be wrong length.
lm.fit  <- lm(y ~ X, data = data) 

# This explicit formula works for predictions.
lm.fit2 <- lm(y ~ X.1 + X.2 + X.3, data = data)

# This will also work for predictions, but can handle variable number of columns in the
# predictor matrix:
predictor.names <- grep('X', names(data), value = TRUE)
lm.fit.formula <- sprintf("y ~ %s", paste(predictor.names, collapse = " + "))
lm.fit3 <- lm(lm.fit.formula, data = data)

# Plot the data
#quartz()
plot(x, y)
lines(x, predict(lm.fit), col = "red", lwd = 3)

log.out <- function(msg) {
  cat(msg, "\n")
}

make.predict <- function(lm.model, new.x) {
 new.X <- poly(new.x, 3, raw = TRUE)
 new.data <- data.frame(X = new.X)
 new.prediction <- predict(lm.model, newdata = new.data)
 if ( length(new.x) != length(new.prediction) ) {
  log.out("Warning: length(new.x) != length(new.prediction)\n")
 } else {
  log.out("Prediction length is okay!\n")
 }
}


# This will not work, prediction is wrong length
log.out("Predicting with lm.fit: ")
make.predict(lm.fit, new.x)

# This works, prediction is correct length
log.out("Predicting with lm.fit2: ")
make.predict(lm.fit2, new.x)

log.out("Prediction with lm.fit3: ")
make.predict(lm.fit3, new.x)