Easy OCaml scripts


Write programs not scripts. I know… but at least quit your abusive relationship to sh derived linguistic disasters.

Getting started

A b0caml script is not different from an ocaml one, it just restricts the toplevel directives you can use and makes it easy for you to tap into installed third-party modules.

Here's a simple echo script:

cat > echo <<EOCAML
#!/usr/bin/env b0caml

let echo oc ss = output_string oc (String.concat " " ss ^ "\n")
let () = echo stdout (List.tl (Array.to_list Sys.argv))
> chmod +x ./echo
> ./echo grunt
> b0caml ./echo grunt  # for Windows compatible invocations

Except for the shebang line, nothing different.

Using third-party modules

If you have compiled modules in a directory DIR that you want to use in your script, add the following directive after the shebang line, before the source and before any comment because b0caml authors are very lazy:

#directory "DIR"

If DIR is relative it is made absolute with respect to the directory of the script. Using the "+DIR" syntax looks for modules in the DIR directory of the directories mentioned in the OCAMLPATH environment variable.

ocaml also has the #directory directive but b0caml treats it a bit differently. First b0caml errors if DIR or +DIR does not resolve to an existing directory. Second b0caml knows how to load the implementation and dependencies of the modules it finds in that directory.

The following script uses the Ptime and Ptime_clock modules. These modules are installed by the ptime package in the ptime and ptime/os of a directory assumed to be in the OCAMLPATH:

> cat > local-time <<EOCAML
#!/usr/bin/env b0caml
#directory "+ptime"
#directory "+ptime/os"

let to_string () =
  let now = Ptime_clock.now () in
  let tz_offset_s = Ptime_clock.current_tz_offset_s () in
  Format.asprintf "%a" (Ptime.pp_human ?tz_offset_s ()) now

let main () = print_endline (to_string ())
let () = if !Sys.interactive then () else main ()
> chmod +x local-time
> ./local-time
1995-09-12 11:27:13 +02:00

Toplevel (REPL) debugging

Since ocaml does not know how to load the implementation of the interfaces it finds in #directory directives you cannot directly load a b0caml script in the toplevel.

b0caml provide the --top (or --utop) option which loads a script and the module it needs in the OCaml toplevel for interactive testing and debugging:

> b0caml --top local-time  # load script and deps in the toplevel
        OCaml version 4.08.0

# Local_time.to_string ();;
- : string = "1995-09-12 11:27:13 +02:00"

The name of the module for the script is determined by mangling the script filename.

If your script parses command line arguments or uses exit you should properly isolate these computations in a main function and prevent its invocation whenever Sys.interactive is true (see for example the source of local-time above).

Modular scripts

Repeat after me, write a program not a script.

A b0caml script can import an OCaml implementation source SRC with the #mod_use "SRC" directive. SRC must be a regular OCaml implementation, it cannot be a b0caml script. A relative SRC is made absolute with respect to the directory of the script.

These directives should also appear only after the shebang line and before the script source or comments. The file SRC must exist or the scripts errors.

A quick and dirty configuration file for a script screams to #mod_use:

> cat > conf.ml <<EOCAML
let lang = "fr"
> cat > miaow <<EOCAML
#!/usr/bin/env b0caml
#mod_use "conf.ml"

let scream = match Conf.lang with | "fr" -> "Miaou!" | _ -> "Miaow!"
let main () = print_endline scream
let () = if !Sys.interactive then () else main ()
> chmod +x miaow
> ./miaow

#mod_use mangles the filename of the path to define a module name and implementation in which the contents of the file is included litteraly. In the example above the line #mod_use "conf.ml" is simply expanded to:

module Conf = struct
#1 "/absolute/path/to/conf.ml"
let lang = "fr"

Knowing OCaml's scoping rules it should be easy to see that a module you #mod_use can refer to the modules #mod_used before. But it is your duty to provide them in the right order. The relative order between #directory and #mod_use directives doesn't matter, you can consider all #directory directives to be written before the first #mod_use.

A #mod_used implementation is constrained by an interface if there's a side .mli for the file you include. For the full details read here.

If you want to see the #mod_use expansions that are performand by b0caml on a script, the following invocation prints the final script source before it gets compiled:

> b0caml --source miaow

Dependency management

No formal dependency management is provided for the third-party modules you use – WRITE A PROGRAM NOT A SCRIPT !

However the #directory and #mod_use directives of a script and can be resolved with the deps subcommand:

> b0caml deps local-time
> b0caml deps miaow

These invocations error and the program exits with a non-zero exit code if the directives do not resolve. For a +DIR directory directive, resolution checks the directory DIR exists in at least one of the directories mentioned in the OCAMLPATH environment variable.

The --raw option prevents resolution and reports the verbatim directive arguments.

> b0caml deps --raw local-time
> b0caml deps --raw miaow

The --root option also eschews resolution and outputs root directory names of + directory directives:

> b0caml deps --root local-time  # Extract +DIR roots
> opam install $(b0caml deps --root local-time)  # You must be joking...

Script cache and build log

The first time a b0caml script runs it gets compiled and cached. This incurs a small overhead. If you want to avoid it, or simply test that it compiles without running it use the --compile option:

b0caml --compile local-time  # compile and cache the script
b0caml log local-time     # Output build log of [local-time]
b0caml log -l local-time  # More details...

By default the script compilation cache location is determined according to the XDG_CACHE_HOME convention. The actual location of the cache can be printed via

b0caml cache path               # print path to the cache
b0caml cache delete local-time  # delete local-time build
b0caml cache                    # delete the cache
b0caml cache size               # print stats about the cache
b0caml cache trim               # trim the cache to 50% of its size


Configuration is looked up in XDG_CONFIG_DIR/b0caml/config. The configuration file is a sequence of s-expressions. Here's a sample file:

(max-cache-size-mb 500)
(compilation-target byte) ; force use of bytecode
(compilation-env         ; Specify the OCaml compilation environment
  (OCAMLPATH /usr/local/ocaml)
  (PATH /usr/local/bin))

The following keys are defined:

Editor support

We hope to eventually convince ocamlmerlin to understand #directory directives and abide by OCAMLPATH the way b0caml does. This will have merlin work out of the box in your script without having to specify anything. If you are using #mod_use you will be punished accordingly.

One thing that remains is for your editor to treat files with b0caml's shebang line as an OCaml file. Follow the instructions below according to your editor.


Add one of the following line to your .emacs depending on the OCaml mode you are using.

(add-to-list 'interpreter-mode-alist '("b0caml" . caml-mode))
(add-to-list 'interpreter-mode-alist '("b0caml" . tuareg-mode))

Script language reference


A b0caml script is an optional shebang line, followed by white space (no comments) separated directives, followed by an OCaml unit implementation.

Using an RFC 5234 grammar this reads as:

script     = [shebang] *(ws directive) ws unit-implementation
shebang    = "!#" *(%x00-%xFF) nl
directive  = dir-dir / dir-use
dir-dir    = "#directory" ws %x22 dchar *dchar %x22
dir-use    = "#mod_use" ws %x22 dchar *dchar %x22
dchar      = escape / cont / ws / %x21 / %x23-%x5B / %x5D-%x7E / %x80-xFF
escape     = %x5C (%x20 / %x22 / %x5C)
cont       = %x5C nl ws
ws         = *(%x20 / %x09 / %x0A / %x0B / %x0C / %x0D)
nl         = %x0A / %x0D / %x0D %x0A
unit-implementation = ... ; See the syntax in the OCaml manual

This syntax is a subset of ocaml's one. However b0caml attributes slightly different semantics to the directives.


The following parts can be distinguished in a b0caml script:

  1. The optional shebang line. This is ignored by b0caml.
  2. The ordered sequence of directives. These can be separated in two ordered list: those of #directory directives and those of #mod_use directives.
  3. The OCaml unit implementation.

The final source of the script is created by concatenating the expansion of the #mod_use directives followed by the OCaml unit implementation. This source is compiled in a compilation environment defined by the #directory directives to a module or a program to be executed.

#directory directive

The syntax of the #directory directive is:

#directory "DIR"

The semantics is to simply add the file path DIR to the ordered list of directories looked up for third-party modules. From a compilation perspective you can see that as -I DIR options given to the compiler.

If DIR is relative it is made absolute with respect to the directory of the script. The +DIR syntax indicates to add all the existing DIR directories from the directories mentioned in the OCAMLPATH environment variable.

The directories have to resolve to existing directories or the script errors. For +DIR it must exist in at least one of the directories of OCAMLPATH.

#mod_use directive

The syntax of the #mod_use directive is:

#mod_use "PATH"

The semantics is to define a module with the contents of PATH at that location. The name of the module is defined by mangling the file name of PATH.

If PATH is relative it is interpreted relative to the script's directory.

For example assuming the filename of PATH is file.ml, the directive expands to:

module File = struct
#1 "PATH"
(* contents of PATH *)

If PATH has a corresponding .mli file say MLI in the same directory, the directive expands to:

module File : sig
#line 1 "MLI"
(* contents of MLI *)
end = struct
#line 1 "PATH"
(* contents of PATH *)

Filename mangling

b0caml uses the following filename mangling convention to produce OCaml module names from arbitrary filenames:

  1. Remove any trailing .ml or .mli.
  2. Map any dash - (0x2D) or dot . (0x2E) to an underscore _ (0x5F).
  3. Map any byte not allowed in OCaml compilation unit names to its two digit capital hexadecimal encoding.
  4. If the result does not start with an US-ASCII letter, prefix the unit name with 'M'.
  5. Capitalize the first letter.

Note that the transformation is neither injective nor surjective. Here are a few examples:

           filename   Module name
    publish-website   Publish_website
     import-data.ml   Import_data
 import-data.xml.ml   Import_data_xml
 import-data.script   Import_data_script
          mix+match   Mix2Bmatch
        _release.ml   M_release

Execution and exit codes

On script execution b0caml terminates with the exit code of you script. However that code may be determined by b0caml itself in case it doesn't get to execute the script. These code may muddle with your own script's exit codes, here's the list of these:


Given a script script.ml its final source SRC is extracted as defined here.

  1. Create the final source SRC by expanding script.ml's #mod_use directive as described here and appending script.ml's OCaml compilation unit.
  2. SRC is compiled to a byte or native code executable via single invocation to the OCaml compiler, the includes specified via #directory directives and all the library archives that are found in these directories along with their dependencies as determined by Archive dependency lookuparchive dependency lookup

Archive dependency lookup