support more int types

[UnixJunkie]

Hello,

Supporting all integer types from the OCaml world would be nice:
e.g. Char.t, Int32.t, Int64.t plus all the integer types in Bigarray.

[UnixJunkie]

What is the kind of compression rates that you get by the way?

[barko]

I'm not sure that overloading the interface with more types is really worth it, given that more input types can be easily accommodated through a one-line use of Seq.map. For example, to encode a list of Int32.t's, do:

let i32_lst = [1l; 22l; 333l; 444l] in
let i32_seq = List.to_seq i32_lst in
let i_seq = Seq.map Int32.to_int i32_seq in
let encoded_word_seq = Simple63.encode_to_seq i_seq in 
...

[UnixJunkie]

But, isn't the compression less efficient in that case?
Also, this approach would not work for Int64.t, unless you want to cut the high bits in your integers...

[barko]

The compression rate will vary depending on the distribution of the input sequence of integers. On one extreme, if the input sequence consists of just 0's and 1's, the compression rate will be fabulous: 1/59 . At the other extreme, if the input integers are big (such that their 59-th most significant bit is set to 1), the compression ratio will be a mediocre 1. This technique should in general be able achieve better compression than byte-oriented schemes for sequences of integers that are smaller than a 256.

Right, the approach does not work for integers bigger than (1 lsl 59) - 1 . The encoding functions raise Invalid in such cases.