SysTutorials

OCaml Array Length Limits Explained

· Q A · updated April 12, 2026

[md] The maximum array length in OCaml is platform-dependent and can be queried at runtime using Sys.max_array_length:

# Sys.max_array_length;;
- : int = 4611686018427387903

## Values by Platform

**64-bit systems:**

# Sys.max_array_length;;
- : int = 4611686018427387903

**32-bit systems:**

# Sys.max_array_length;;
- : int = 1073741823

Note that 32-bit OCaml support has been deprecated in recent versions. Modern OCaml distributions primarily target 64-bit architectures, and OCaml 5.0+ dropped official support for 32-bit platforms entirely.

## Why These Limits Exist

OCaml’s array length limit stems from how the runtime represents values in memory. Arrays store a length field alongside the data, and this length is encoded in a tagged integer. On 64-bit systems, OCaml uses 63 bits for value representation (one bit reserved for tagging), giving the theoretical maximum of 2^62 – 1 elements.

The practical limit accounts for:

– The tag bit reserved by OCaml’s garbage collector – Memory alignment requirements – The need to distinguish between different value types

## Memory Consumption

While the theoretical limits are enormous, you’ll hit memory constraints long before reaching Sys.max_array_length. Here’s the actual memory usage for OCaml arrays:

– int array — 8 bytes per element on 64-bit (tagged integers) – float array — 8 bytes per element (unboxed in dedicated float arrays) – string array — 8 bytes per pointer + string data

A float array with 100 million elements uses ~800MB. A billion elements requires ~8GB — feasible on modern hardware but beyond typical workstation workloads.

## Practical Alternatives for Large Data

For handling datasets that approach memory limits, consider these alternatives:

**Bigarrays** for C-interop and memory-mapped I/O:

let big_arr = Bigarray.Array1.create Bigarray.float64 Bigarray.c_layout 100_000_000

Bigarrays have their own limit (Sys.max_bigarray_length), which is typically larger than regular arrays, and they can be memory-mapped to files for persistent storage.

**Buffers for byte data:**

let buf = Buffer.create 65536
Buffer.add_string buf "data"

**Bytes for mutable byte sequences:**

let b = Bytes.make 1024 '\\000'

**Persistent data structures** from libraries like Core or Base when immutability is needed:

(* Using Core_kernel *)
let arr = Array.of_list [1; 2; 3]

## OCaml 5.0 Parallelism Changes

OCaml 5.0 introduced major changes to the runtime that affect array behavior in parallel contexts:

– **Domains** — OCaml 5’s parallelism unit. Each domain has its own minor heap but shares the major heap. – Arrays in shared memory are accessible from multiple domains, but you need proper synchronization for mutation. – The Domain module enables parallel computation on array segments:

let parallel_map f arr =
  let n = Array.length arr in
  let domain_count = Domain.recommended_domain_count () in
  let chunk_size = n / domain_count in
  let domains = Array.init domain_count (fun i ->
    let start = i * chunk_size in
    let stop = min ((i + 1) * chunk_size) n in
    Domain.spawn (fun () ->
      Array.init (stop - start) (fun j -> f arr.(start + j))
    )
  ) in
  Array.concat_map (Domain.join) domains

## Check Your Actual Limits

In a running OCaml session:

let () =
  Printf.printf "Max array length: %d\n" Sys.max_array_length;
  Printf.printf "Max bigarray length: %d\n" Sys.max_bigarray_length;
  Printf.printf "System word size: %d bits\n" Sys.word_size;
  Printf.printf "OCaml version: %s\n" Sys.ocaml_version

For safe array allocation without exceeding available memory, validate requested sizes against both Sys.max_array_length and available heap space before allocation.

← All posts