Skip to main content

​
Overview

httpz is a zero-allocation HTTP/1.1 parser for OxCaml that achieves exceptional performance through careful architectural choices. The parser processes HTTP requests without any heap allocations by leveraging OxCaml’s unboxed types, stack-allocated data structures, and span-based parsing.

​
Core Design Principles

​
1. Buffer Reuse Strategy

httpz uses a fixed 32KB pre-allocated buffer that is reused across all requests: 
(* buf_read.ml:44 *)
let buffer_size = 32768
let create () = Base_bigstring.create buffer_size
This buffer is allocated once and reused indefinitely, eliminating per-request allocation overhead. The 32KB size accommodates most HTTP requests while staying within reasonable memory bounds.

​
2. Span-Based Parsing

Instead of copying strings from the buffer, httpz uses spans - lightweight references into the buffer: 
(* span.ml:10-13 *)
type t =
  #{ off : int16#
   ; len : int16#
   }
Spans are unboxed records containing only an offset and length (both as int16#). Since the max buffer size is 32KB, int16# is sufficient and avoids the overhead of boxed integers. Example usage: 
(* parser.ml:157-162 *)
let[@inline] parse_target st ~(pos : int16#) : #(Span.t * int16#) =
  let #(sp, pos) = take_while (fun c ->
    Buf_read.( <>. ) c #' ' && Buf_read.( <>. ) c #'\r') st ~pos
  in
  Err.when_ (Span.len sp = 0) Err.Invalid_target;
  #(sp, pos)

​
3. Parser Combinator Approach

httpz implements a parser combinator library where position is threaded explicitly: 
(* parser.ml:10-11 *)
type pstate = #{ buf : Base_bigstring.t; len : int16# }
Every combinator follows the pattern: 
val combinator : pstate -> pos:int16# -> #(result * int16#)
The combinator takes the current position, performs parsing, and returns both the result and the new position. This explicit position threading eliminates the need for mutable state. Example combinators: 
(* parser.ml:43-46 *)
let[@inline] char (c : char#) st ~(pos : int16#) : int16# =
  Err.partial_when @@ at_end st ~pos;
  Err.malformed_when @@ Buf_read.( <>. ) (Buf_read.peek st.#buf pos) c;
  add16 pos one16

(* parser.ml:48-57 *)
let[@inline] string (s : string) st ~(pos : int16#) : int16# =
  let slen = String.length s in
  Err.partial_when (to_int (remaining st ~pos) < slen);
  for i = 0 to slen - 1 do
    let actual = Buf_read.peek st.#buf (add16 pos (i16 i)) in
    let expected = Stdlib_stable.Char_u.of_char (String.unsafe_get s i) in
    Err.malformed_when @@ Buf_read.( <>. ) actual expected
  done;
  add16 pos (i16 slen)

​
4. Request Parsing Flow

The complete request parsing follows this sequence: 
(* httpz.ml:172-188 *)
let parse buf ~(len : int16#) ~limits = exclave_
  let open Buf_read in
  if to_int len > buffer_size || gt16 len limits.#max_header_size then
    error_result Headers_too_large
  else
    try
      let pst = Parser.make buf ~len in
      let #(meth, target, version, pos) = Parser.request_line pst ~pos:(i16 0) in
      let #(body_off, st, headers) =
        parse_headers_loop pst ~pos ~acc:[] initial_header_state ~limits
      in
      (* Only missing Host header needs end-of-parse check *)
      match (version, st.#has_host) with
      | (Version.Http_1_1, false) -> error_result Missing_host_header
      | _ -> build_request ~meth ~target ~version ~body_off st ~headers
    with Err.Parse_error status ->
      error_result status
Step-by-step:
  1. Validate buffer size against limits
  2. Parse request line (method, target, version)
  3. Parse headers in a loop, accumulating to a local list
  4. Validate HTTP/1.1 requirements (Host header)
  5. Build request struct with all parsed data

​
5. Header Parsing State

During header parsing, httpz maintains state in an unboxed record: 
(* httpz.ml:55-64 *)
type header_state =
 #{ count : int16#
  ; content_len : int64#
  ; chunked : bool
  ; conn : conn_value
  ; has_cl : bool
  ; has_te : bool
  ; has_host : bool
  ; expect_continue : bool
  }
This state is updated functionally (not mutated) as headers are parsed: 
(* httpz.ml:136-144 *)
match name with
| Header_name.Content_length ->
  Err.when_ st.#has_te Err.Ambiguous_framing;
  let #(parsed_len, overflow) =
    Span.parse_int64_limited pst.#buf value_span ~max_value:limits.#max_content_length
  in
  Err.when_ overflow Err.Content_length_overflow;
  parse_headers_loop pst ~pos ~acc ~limits
    #{ st with count = next_count; content_len = parsed_len; has_cl = true }

​
6. Local Lists for Headers

Headers are accumulated in a local list that lives on the stack: 
(* httpz.ml:123 *)
let rec parse_headers_loop (pst : Parser.pstate) ~pos ~acc (st : header_state) ~limits
  : #(int16# * header_state * Header.t list) = exclave_
The exclave_ annotation indicates the function returns stack-allocated values. Headers are prepended to the accumulator: 
(* httpz.ml:164-167 *)
| _ ->
  let hdr = { Header.name; name_span; value = value_span } in
  parse_headers_loop pst ~pos ~acc:(hdr :: acc) ~limits
    #{ st with count = next_count }

​
Security Features

httpz includes comprehensive RFC 7230 validation:

​
Configurable Limits

(* buf_read.ml:103-115 *)
type limits =
  #{ max_content_length : int64#  (* Default: 100MB *)
   ; max_header_size : int16#     (* Default: 16KB *)
   ; max_header_count : int16#    (* Default: 100 *)
   ; max_chunk_size : int         (* Default: 16MB *)
   }

let default_limits =
  #{ max_content_length = #104857600L
   ; max_header_size = i16 16384
   ; max_header_count = i16 100
   ; max_chunk_size = 16777216
   }

​
HTTP Smuggling Prevention

(* buf_read.ml:119-135 *)
let[@inline] has_bare_cr (local_ buf) ~(pos : int16#) ~(len : int16#) =
  let pos = to_int pos in
  let len = to_int len in
  let end_pos = pos + len in
  let mutable p = pos in
  let mutable found = false in
  while (not found) && p < end_pos do
    if peek buf (i16 p) =. #'\r' then (
      if p + 1 >= end_pos || peek buf (i16 (p + 1)) <>. #'\n' then
        found <- true
      else
        p <- p + 2
    ) else
      p <- p + 1
  done;
  found

​
Ambiguous Framing Detection

(* httpz.ml:137-138 *)
| Header_name.Content_length ->
  Err.when_ st.#has_te Err.Ambiguous_framing;

(* httpz.ml:145-146 *)
| Header_name.Transfer_encoding ->
  Err.when_ st.#has_cl Err.Ambiguous_framing;

​
Chunked Transfer Encoding

httpz supports chunked encoding with overflow protection: 
(* chunk.ml:55-83 *)
let[@inline] parse_hex_size_limited buf ~off ~len ~max_size =
  let module P = Buf_read in
  let mutable pos = off in
  let mutable size = 0 in
  let mutable valid = true in
  let mutable overflow = false in
  let mutable digit_count = 0 in
  while valid && pos < len do
    let digit = hex_digit_value (P.peek buf (i16 pos)) in
    if digit >= 0 then (
      digit_count <- digit_count + 1;
      if digit_count > to_int max_hex_digits then (
        overflow <- true;
        valid <- false
      ) else (
        let new_size = (size * 16) + digit in
        if new_size > max_size then (
          overflow <- true;
          valid <- false
        ) else (
          size <- new_size;
          pos <- pos + 1
        )
      )
    ) else
      valid <- false
  done;
  #(size, pos, overflow)

​
Performance Characteristics

  • Parsing throughput: 6.5M requests/sec
  • Minimal request (35B): 300ns, 0 heap allocations
  • Simple request (4 headers): 925ns, 0 heap allocations
  • Browser request (10 headers): 3.3μs, 0 heap allocations
  • 50 headers: 11.2μs, 0 heap allocations
The architecture enables true zero-allocation parsing across all request sizes.

Build docs developers (and LLMs) love

Get started for freeTalk to us