package decompress

compress ?level ?dynamic ~w ~q ~refill ~flush i o is Zlib.compress (with ~header:true) provided by camlzip package.

• w is the window used by LZ77 compression algorithm.
• q is shared-queue between compression algorithm and DEFLATE encoder.
• i is input buffer.
• o is output buffer.

When compress wants more input, it calls refill with i. The client returns how many bytes he wrote into i. If he returns 0, he signals end of input.

When compress has written output buffer, it calls flush with o and how many bytes it wrote. Bytes into o must be copied and they will be lost at the next call to flush.

A simple example of how to use such interface is:

let deflate_string ?(level= 4) str =
  let i = De.bigstring_create De.io_buffer_size in
  let o = De.bigstring_create De.io_buffer_size in
  let w = De.Lz77.make_window ~bits:15 in
  let q = De.Queue.create 0x1000 in
  let r = Buffer.create 0x1000 in
  let p = ref 0 in
  let refill buf =
    let len = min (String.length str - !p) De.io_buffer_size in
    Bigstringaf.blit_from_string str ~src_off:!p buf ~dst_off:0 ~len ;
    p := !p + len ; len in
  let flush buf len =
    let str = Bigstringaf.substring buf ~off:0 ~len in
    Buffer.add_string r str in
  Zl.Higher.compress ~level ~dynamic:true
    ~w ~q ~refill ~flush i o ;
  Buffer.contents r

As De.Higher.compress, several choices was made in this code and decompress don't want to be responsible of them. It's why such function exists only as example when lengths of buffers (such as i, o or q) changes the speed/compression ratio/memory consumption.

uncompress ~allocate ~refill ~flush i o is Zlib.uncompress (with ~header:true) provided by camlzip package.

• allocate is the allocator of window used by LZ77 uncompression algorithm
• i is input buffer.
• o is output buffer.

When uncompress wants more input, it calls refill with i. The client returns how many bytes he wrote into i. If he returns 0, he signals end of input.

When uncompress has written output buffer, it calls flush with o and how many bytes it wrote. Bytes into o must be copied and they will be lost at the next call to flush.

A simple example of how to use such interface is:

let inflate_string str =
  let i = De.bigstring_create De.io_buffer_size in
  let o = De.bigstring_create De.io_buffer_size in
  let allocate bits = De.make_window ~bits in
  let r = Buffer.create 0x1000 in
  let p = ref 0 in
  let refill buf =
    let len = min (String.length str - !p) De.io_buffer_size in
    Bigstringaf.blit_from_string str ~src_off:!p buf ~dst_off:0 ~len ;
    p := !p + len ; len in
  let flush buf len =
    let str = Bigstringaf.substring buf ~off:0 ~len in
    Buffer.add_string r str in
  match Zl.Higher.uncompress ~allocate ~refill ~flush i o with
  | Ok () -> Ok (Buffer.contents r)
  | Error _ as err -> err

As you can see, several allocations appear. As long as you want to uncompress several contents for example, you can re-use the same window instead of an allocation of one per uncompression. Then, the throughput is mostly limited by i and o (even bigger, even faster but it requires memories). decompress don't want to be responsible about these choices, it's why such function exists only as an example.