Implementation of Zlib and GZip in OCaml






Decompress - Pure OCaml implementation of decompression algorithms

decompress is a library which implements:

The library

The library is available with:

$ opam install decompress

It provides three sub-packages:

  • to handle RFC1951 stream

  • decompress.zl to handle Zlib stream

  • decompress.gz to handle Gzip stream

  • decompress.lzo to handle LZO contents

Each sub-package provide 3 sub-modules:

  • Inf to inflate/decompress a stream

  • Def to deflate/compress a stream

  • Higher as a easy entry point to use the stream

How to use it

Link issue

decompress uses checkseum to compute CRC of streams.
checkseum provides 2 implementations:

  • a C implementation to be fast

  • an OCaml implementation to be usable with js_of_ocaml (or, at least, require
    only the caml runtime)

When the user wants to make an OCaml executable, it must choose which implementation
of checkseum he wants. A compilation of an executable with decompress.zl is:

$ ocamlfind opt -linkpkg -package checkseum.c,decompress.zl

Otherwise, the end-user should have a linking error (see

With dune

checkseum uses a mechanism integrated into dune which solves the link issue.
It provides a way to silently choose the default implementation of checkseum:

By this way (and only with dune), an executable with decompress.zl is:

 (name main)
 (libraries decompress.zl))

Of course, the user still is able to choose which implementation he wants:

 (name main)
 (libraries checkseum.ocaml decompress.zl))


decompress proposes to the user a full control of:

  • the input/output loop

  • the allocation

Input / Output

The process of the inflation/deflation is non-blocking and it does not require
any syscalls (as an usual MirageOS project). The user can decide how to get the
input and how to store the output.

An usual loop (which can fit into lwt or async) of decompress.zl is:

let rec go decoder = match Zl.Inf.decode decoder with
  | `Await decoder ->
    let len = input itmp 0 (Bigstringaf.length tmp) in
    go (Zl.Inf.src decoder itmp 0 len)
  | `Flush decoder ->
    let len = Bigstringaf.length otmp - Zl.Inf.dst_rem decoder in
    output stdout otmp 0 len ;
    go (Zl.Inf.flush decoder)
  | `Malformed err -> invalid_arg err
  | `End decoder ->
    let len = Bigstringaf.length otmp - Zl.Inf.dst_rem decoder in
    output stdout otmp 0 len in
go decoder

Then, the process does not allocate large objects but it requires at the
initialisation these objects. Such objects can be re-used by another
inflation/deflation process - of course, these processes can not use same
objects at the same time.

val decompress : window:De.window -> in_channel -> out_channel -> unit

let w0 = De.make_windows ~bits:15

(* Safe use of decompress *)
let () =
  decompress ~window:w0 stdin stdout ;
  decompress ~window:w0 (open_in "file.z") (open_out "file")

(* Unsafe use of decompress,
   the second process must use an other pre-allocated window. *)
let () = @@
    Lwt.join [ (decompress ~window:w0 stdin stdout |> Lwt.return)
             ; (decompress ~window:w0 (open_in "file.z") (open_out "file") |> Lwt.return) ]

This ability can be used on:

  • the input buffer given to the encoder/decoder with src

  • the output buffer given to the encoder/decoder

  • the window given to the encoder/decoder

  • the shared-queue used by the compression algorithm and the encoder


An example exists into bin/ where you can see how to use
decompress.zl and

Higher interface

However, decompress provides a higher interface close to what camlzip provides
to help newcomers to use decompress:

val compress : refill:(bigstring -> int) -> flush:(bigstring -> int -> unit) -> unit
val uncompress : refill:(bigstring -> int) -> flush:(bigstring -> int -> unit) -> unit


decompress has a benchmark about inflation to see if any update has a performance
implication. The process try to inflate a stream and stop at N second(s) (default is 30),
The benchmark requires libzlib-dev, cmdliner and bos to be able to compile zpipe
and the executable to produce the CSV file. To build the benchmark:

$ dune build bench/output.csv

On linux machines, /dev/urandom will generate the random input for piping to zpipe. To
run the benchmark:

$ cat /dev/urandom | ./build/default/bench/zpipe | ./_build/default/bench/bench.exe

The output file is a CSV file which can be processed by a plot software. It records
input bytes, output bytes and memory usage at each second.

Build Requirements

  • OCaml >= 4.07.0

  • dune to build the project

  • base-bytes meta-package

  • bigarray-compat

  • checkseum

  • optint