Cooley-Tukey FFT in R radix-2 DIT case - r

Question

So I've been trying to (manually) implement the Cooley-Turkey FFT algorithm in R (for Inputs with size N=n^2). I tried:
myfft <- function(s){
N <- length(s)
if (N != 1){
s[1:(N/2)] <- myfft(s[(1:(N/2))*2-1])
s[(N/2+1):N] <- myfft(s[(1:(N/2))*2])
for (k in 1:(N/2)){
t <- s[k]
s[k] <- t + exp(-1i*2*pi*(k-1)/N) * s[k+N/2]
s[k+N/2] <- t - exp(-1i*2*pi*(k-1)/N) * s[k+N/2]
}
}
s
}
This compiles, but for n>1, N=2^n it does not compute the right values. I implemented a DFT-function and used the fft() function to compare, both compute, when normalized, give the same values, but seem to disagree with my algorithm above.
If anyone feels interested and sees where I went wrong, help would be greatly appreciated, I'm going mad searching for the mistake and am starting to question, if I even ever understood this FFT algorithm.
UPDATE: I fixed it, I'm not 100% sure where the problem exactly was, but here is the working implementation:
myfft <- function(s){
N <- length(s)
if (N != 1){
t <- s
t[1:(N/2)] <- myfft(s[(1:(N/2))*2-1]) # 1 3 5 7 ...
t[(N/2+1):N] <- myfft(s[(1:(N/2))*2]) # 2 4 6 8 ...
s[1:(N/2)] <- t[1:(N/2)] + exp(-1i*2*pi*(0:(N/2-1))/N) * t[(N/2+1):N]
s[(N/2+1):N] <- t[1:(N/2)] - exp(-1i*2*pi*(0:(N/2-1))/N) * t[(N/2+1):N]
}
return(s)
}

Answer 1

The problem was with the following line
s[1:(N/2)] <- myfft(s[(1:(N/2))*2-1])
which was overwriting part of the untransformed values that were needed on the subsequent line:
s[(N/2+1):N] <- myfft(s[(1:(N/2))*2])
For example, when N=4, the second call to myfft uses s[2] and s[4], but the assignment from the first call to myfft writes into s[1] and s[2] (thus overwriting the required original value in s[2]).
Your solution of copying the entire array prevents this overwrite.
An alternate solution commonly used is to copy the even and odd parts separately:
myfft <- function(s){
N <- length(s)
if (N != 1){
odd <- s[(1:(N/2))*2-1]
even <- s[(1:(N/2))*2]
s[1:(N/2)] <- myfft(odd)
s[(N/2+1):N] <- myfft(even)
s[1:(N/2)] <- t[1:(N/2)] + exp(-1i*2*pi*(0:(N/2-1))/N) * t[(N/2+1):N]
s[(N/2+1):N] <- t[1:(N/2)] - exp(-1i*2*pi*(0:(N/2-1))/N) * t[(N/2+1):N]
}
return(s)
}