Persistent store with Git semantics: lazy reads, delayed writes, content-addressing
ocaml-merkle-sync: generic Merkle DAG sync; irmin Sync.S module type
ocaml-merkle-sync: reusable sync primitives for any content-addressed DAG.
- Layer 1: anti-entropy gossip (branch head exchange, O(log n))
- Layer 2: merkle descent (DAG diff by hash, dependency-order transfer)
- Bloom filter (FNV-1a, ~1% FP rate)
15 tests passing. Fuzz harness for bloom properties.
irmin/lib/sync.ml: Sync.S module type — discover/locate/fetch/push.
Each backend provides its own implementation.
Also: dune fmt across affected packages.
+39
-42
+11
-9
lib/sync.ml
+11
-9
lib/sync.ml
···
1
-
module type TRANSPORT = sig
1
+
module type S = sig
2
2
type t
3
3
type hash
4
4
5
-
val fetch : t -> branch:string -> (hash, [ `Msg of string ]) result
6
-
val push : t -> branch:string -> hash -> (unit, [ `Msg of string ]) result
7
-
end
5
+
val discover : t -> (string * hash) list
6
+
val locate : t -> local_has:(hash -> bool) -> hash -> hash list
7
+
val fetch : t -> hash list -> (hash * string) list
8
8
9
-
type resolver =
10
-
| Fail
11
-
| Ours
12
-
| Theirs
13
-
| Custom of (string list -> string option -> string option -> string)
9
+
val push :
10
+
t ->
11
+
(hash * string) list ->
12
+
branch:string ->
13
+
hash ->
14
+
(unit, [ `Msg of string ]) result
15
+
end
+28
-33
lib/sync.mli
+28
-33
lib/sync.mli
···
1
-
(** Distributed sync: transport + merge composition.
1
+
(** Sync backend interface.
2
2
3
-
Sync composes a backend-specific {!TRANSPORT} (block transfer) with Schema's
4
-
merge (conflict resolution). Three modes:
3
+
Each backend provides its own implementation of the four sync operations.
4
+
Irmin composes them: [discover] → [locate] → [fetch] → [Schema.merge] →
5
+
(optionally) [push].
5
6
6
-
- {b Explicit}: user-triggered pull/push (like Git).
7
-
- {b Continuous}: automatic gossip between peers (eventually consistent).
8
-
- {b DTN}: delay-tolerant via space-dtn bundle envelopes.
7
+
- Git: packfile negotiation, [git fetch] / [git push]
8
+
- ATProto: CAR download/upload via XRPC
9
+
- Local: direct heap access via [Git.Fetch.fetch_local]
10
+
- Generic: [ocaml-merkle-sync] (gossip + merkle descent + bloom) *)
9
11
10
-
The transport handles "get blocks from there to here." The merge handles
11
-
"reconcile concurrent modifications." They compose in [pull]:
12
-
[transport.fetch] + [Schema.merge] + [update_branch]. *)
12
+
module type S = sig
13
+
type t
14
+
(** A connection to a remote store. *)
13
15
14
-
(** {1 Transport}
16
+
type hash
15
17
16
-
Backend-specific block transfer. Each backend implements this: Git uses
17
-
fetch_local / git-fetch, ATProto uses CAR import/export, OCI uses registry
18
-
pull/push. *)
18
+
val discover : t -> (string * hash) list
19
+
(** [discover t] returns the remote's branch heads. *)
19
20
20
-
module type TRANSPORT = sig
21
-
type t
22
-
type hash
21
+
val locate : t -> local_has:(hash -> bool) -> hash -> hash list
22
+
(** [locate t ~local_has remote_head] returns hashes reachable from
23
+
[remote_head] that are missing locally. Result is in dependency order:
24
+
children before parents. *)
23
25
24
-
val fetch : t -> branch:string -> (hash, [ `Msg of string ]) result
25
-
(** [fetch t ~branch] ensures all blocks reachable from the remote's [branch]
26
-
tip are available locally. Returns the remote's HEAD hash. *)
26
+
val fetch : t -> hash list -> (hash * string) list
27
+
(** [fetch t hashes] retrieves blocks by hash from the remote. *)
27
28
28
-
val push : t -> branch:string -> hash -> (unit, [ `Msg of string ]) result
29
-
(** [push t ~branch hash] pushes [hash] and its reachable blocks to the
30
-
remote. Fast-forward only — fails if the remote has diverged. *)
29
+
val push :
30
+
t ->
31
+
(hash * string) list ->
32
+
branch:string ->
33
+
hash ->
34
+
(unit, [ `Msg of string ]) result
35
+
(** [push t blocks ~branch head] sends [blocks] to the remote and updates
36
+
[branch] to [head]. Fast-forward only. *)
31
37
end
32
-
33
-
(** {1 Resolver}
34
-
35
-
Conflict resolution strategy for pull/merge. *)
36
-
37
-
type resolver =
38
-
| Fail (** Abort on any conflict. *)
39
-
| Ours (** Keep our version for all conflicts. *)
40
-
| Theirs (** Keep their version for all conflicts. *)
41
-
| Custom of (string list -> string option -> string option -> string)
42
-
(** [Custom f] calls [f path ours theirs] for each conflict. *)