package ezirmin

Ezirmin is an easy interface on top of the Irmin library. It comes with set of mergeable datatypes, instantiated to specific backends to quickly get going. For example,

$ utop
utop # #require "ezirmin";;
utop # module M = Ezirmin.FS_queue(Tc.String);; (* Mergeable queue of strings *)
utop # open M;;
utop # open Lwt.Infix;;
utop # let m = Lwt_main.run (init ~root:"/tmp/ezirminq" ~bare:true () >>= master);;
val m : branch = <abstr>
utop # push m ["home"; "todo"] "buy milk";;
- : unit = ()
utop # push m ["work"; "todo"] "publish ezirmin";;
- : unit = ()

This mergeable queue is saved in the git repository at /tmp/ezirminq. In another terminal,

$ utop
utop # #require "ezirmin";;
utop # module M = Ezirmin.FS_queue(Tc.String);; (* Mergeable queue of strings *)
utop # open M;;
utop # open Lwt.Infix;;
utop # let m = Lwt_main.run (init ~root:"/tmp/ezirminq" ~bare:true () >>= master);;
val m : branch = <abstr>
utop # pop m ["home"; "todo"];;
- : string option = Some "buy milk"

For concurrency control, use branches. In the first terminal,

utop # let wip_t1 = Lwt_main.run @@ clone_force m "wip_t1";;
utop # push wip_t1 ["home"; "todo"] "walk dog";;
- : unit = ()
utop # push wip_t1 ["home"; "todo"] "take out trash";;
- : unit = ()

The changes are not visible until the branches are merged.

utop # to_list m ["home"; "todo"];;
- : string list = []
utop # merge wip_t1 ~into:m;;
- : unit = ()
utop # to_list m ["home"; "todo"];;
- : string list = ["walk dog"; "take out trash"]

Working with history

In addition to the data structures being mergeable, they are also persistent. In particular, every object stored in Irmin has complete provenance. Ezimrin simplifies working with object history.

let run = Lwt_main.run
let repo = run @@ init ()
let path = ["Books"; "Ovine Supply Logistics"]
let push_msg = push m ~path;;

run begin
  push_msg "Baa" >>= fun () ->
  push_msg "Baa" >>= fun () ->
  push_msg "Black" >>= fun () ->
  push_msg "Camel"
end;;

run @@ to_list m path;;
- : string list = ["Baa"; "Baa"; "Black"; "Camel"]

Clearly this is wrong. Let's fix this by reverting to earlier version:

let m_1::_ = run @@ predecessors repo m;; (** HEAD~1 version *)
run @@ to_list m_1 path;;
- : string list = ["Baa"; "Baa"; "Black"]
run @@ update_branch m ~set:m_1;;
run @@ to_list m path;;
- : string list = ["Baa"; "Baa"; "Black"]

Now that we've undone the error, we can do the right thing.

run @@ push_msg "Sheep";;
run @@ to_list m path;;
- : string list = ["Baa"; "Baa"; "Black"; "Sheep"]

What's in the box

The mergeable datatypes currently available are:

• Counter: A counter with increment and decrement operations.
• Blog_log: An append-only log stored as blobs. Append is an O(n) operation where n is the size of the log.
• Log: An append-only write-optimized log with support for paginated reads. Append is an O(1) operation
• Lww_register: Last-writer-wins register.
• Queue: An efficient queue with O(1) push and pop operations.
• Rope: An efficient data structure for storing and manipulating int-indexed sequence of contents. Requires a mergeable content implementation.
• Rope_string: A rope specialized to strings with intent-preserving merge implemented using operational transformation.

Since these datatypes are defined such that a merge is always possible and that merges in different orders converge, they are CRDTs.

The backends available are:

• Git file system backend
• Git in-memory backend

Ezirmin is distributed under the ISC license.