Hi! I was looking at the code for bytevector-ref, vector-ref, struct-ref and it's pretty conclusive that there is a hugh overhead of implementing those. We are talking of 30-40 instructions that has to be put into native code. Now most of this code is working with registers so it's not that bad. But The code size will be hugh for vector oriented and struct oriented code. On the other hand the call overhead will not be that large compared to all those instructions so we could simply use the C corresponding C-functions directly. What's interesting though for a more elaborate native translation is if the compiler (or the user demands it) for unsafe versions of these accessors (I saw that we have class-of as a dengerous version of struct-ref). The idea here would be to add similar unsafe instructions for later use. They will then most certainly lead to a very efficient execution of many algorithms in guile.
Regards Stefan
