Portable OCaml runtime utilities from skip-ocaml — works with stock toolchains on macOS and Linux.
skip-lite provides high-performance utilities that work with standard OCaml toolchains (no special linker scripts, no objcopy, no fixed-address loading). It's designed for build tools and language servers that need efficient file caching.
Memory-mapped cache for marshalled OCaml files with automatic invalidation.
(* Read a marshalled file with automatic caching *) Marshal_cache.with_unmarshalled_file "/path/to/file.marshal" (fun (data : my_type) -> (* Process data - mmap stays valid during callback *) process data ) (* Reactive/incremental: only process if file changed *) match Marshal_cache.with_unmarshalled_if_changed path process with | Some result -> (* file changed, got new result *) | None -> (* file unchanged, use cached result *)
Features:
- Zero-copy unmarshalling (OCaml 5+
caml_input_value_from_block) - Automatic cache invalidation (mtime + size + inode)
- LRU eviction with configurable limits
- Off-heap storage (not scanned by GC)
- Thread-safe (mutex released during callbacks)
- Reactive API for incremental systems (22-32x speedup)
A reactive collection that maps file paths to processed values with delta-based updates. Designed for use with file watchers.
(* Create collection with processing function *) let coll = Reactive_file_collection.create ~process:(fun (data : cmt_infos) -> extract_types data) (* Initial load *) List.iter (Reactive_file_collection.add coll) (glob "*.cmt") (* On file watcher events *) match event with | Created path -> Reactive_file_collection.add coll path | Deleted path -> Reactive_file_collection.remove coll path | Modified path -> Reactive_file_collection.update coll path (* Or batch updates *) Reactive_file_collection.apply coll [Added "new.cmt"; Modified "changed.cmt"] (* Iterate over all values *) Reactive_file_collection.iter (fun path value -> ...) coll
Features:
- Delta-based API (add/remove/update) - no full rescans
- Batch event application
- Backed by Marshal_cache for efficient file access
- ~950x speedup vs re-reading all files on each change
opam pin add skip-lite https://github.com/rescript-lang/skip-lite.git
Or add to your dune-project:
(depends
(skip-lite (>= 0.1)))
With 10,000 files ×ばつ 20KB each (195MB total):
| Approach | Time | Speedup |
|---|---|---|
Marshal.from_channel (baseline) |
1276 ms | 1x |
with_unmarshalled_file (warm cache) |
742 ms | 1.7x |
with_unmarshalled_if_changed (scan all) |
22 ms | 58x |
| Operation | Time | Speedup vs baseline |
|---|---|---|
| Update 10 files | 1.26 ms | - |
| Iterate all 10k values | 0.09 ms | - |
| Total per iteration | 1.34 ms | 950x |
Once loaded, iterating over all 10,000 processed values takes 0.1ms (9,393x faster than re-reading files).
| Platform | Status |
|---|---|
| macOS 10.13+ | ✅ Fully supported |
| Linux (glibc) | ✅ Fully supported |
| Linux (musl) | |
| FreeBSD/OpenBSD | |
| Windows | ❌ Not supported |
- OCaml 5.0+ — the library uses multicore-safe primitives and tests use
Domain.spawn - dune 3.0+
- C++17 compiler (clang or gcc)
dune build dune runtest
MIT License - see LICENSE file.
- skip-ocaml - Full skip runtime (requires special toolchain setup)