MDX allows to execute code blocks inside markdown and mli/mld documentation to help keeping them up to date.

Use the dune stanza to enable it on your documentation.

mdx is released on opam and can be installed by running:

$ opam install mdx

If you want to contribute to the project, please see the CONTRIBUTING.md.

You can use MDX with your Markdown or .ml{i,d} documentation, which ensures code in multi-line or verbatim code blocks is correct.

To enable MDX on specific files you must first enable it for your project by adding the following stanza to your dune-project:

(using mdx 0.2)

Note that version 0.2 of the MDX stanza is only available in dune 3.0 or higher. You can use the first, 0.1 version with dune 2.4 or higher.

Then add the following in the relevant dune file:

(mdx)

That enables MDX on all markdown files in the folder. The MDX stanza can be further configured. Please visit the relevant section of dune's manual for more information.

MDX supports various type of code blocks but the most common are OCaml toplevel blocks. We illustrate one in our example below. In a Markdown file, you would write something similar to this:

Let's look at how good OCaml is with integers and strings:
```ocaml
# 1 + 2;;
- : int = 2
# "a" ^ "bc";;
- : string = "ab"
```

or in an mli file:

(** Let's look at how good OCaml is with integers and strings:
    {@ocaml[
    # 1 + 2;;
    - : int = 2
    # "a" ^ "bc";;
    - : string = "ab"
    ]}
*)

The content of the toplevel blocks looks just like an interactive toplevel session. Phrases, i.e., the toplevel "input", start with a # and end with ;;. The toplevel evaluation, or "output" follows each phrase.

Now you probably have noticed that 1 + 2 is not equal to 2 nor is "a" ^ "bc" to "ab". Somebody must have updated the phrases, but then forgot to update the evaluation.

That's exactly why MDX is here!

If you enable MDX for this file and then ran dune runtest, this would be the result:

$ dune runtest
File "README.md", line 1, characters 0-0:
       git (internal) (exit 1)
(cd _build/default && /usr/bin/git --no-pager diff --no-index --color=always -u README.md .mdx/README.md.corrected)
diff --git a/README.md b/.mdx/README.md.corrected
index 181b86f..458ecec 100644
--- a/README.md
+++ b/.mdx/README.md.corrected
@@ -1,13 +1,13 @@
Let's look at how good OCaml is with integers and strings:
```ocaml
# 1 + 2;;
-- : int = 2
+- : int = 3
# "a" ^ "bc";;
-- : string = "ab"
+- : string = "abc"
```

The test run just failed and dune is showing the diff between what we have locally and what should be, according to MDX. This uses dune's promotion workflow so at this point you can either investigate it further if you're surprised by this diff or if you're happy with it, simply accept it by running:

dune promote

Now the documentation is up-to-date and running dune runtest again should be successful!

Note that to use the dune runtest/promote workflow with mli or mld files, you will need to adjust the mdx stanza in the dune file, as by default, Dune only checks markdown files with mdx. E.g.,

(mdx
 (files :standard - *.mli))

The blocks can be parameterized by mdx-specific labels, that will change the way mdx interprets the block.

The markdown syntax is: <!-- $MDX LABELS -->, where LABELS is a list of valid labels separated by a comma. This line has to immediately precede the block it is attached to.

<!-- $MDX LABELS -->
```ocaml
```

The .mli and .mld syntax for this is is slightly different to match the conventions of OCaml documentation comments:

(** This is an documentation comment with an ocaml block
    {@ocaml LABELS [
    ]}
*)

The possible labels are:

• skip -- ignore this block
• ocaml, cram, toplevel, include -- set the block type
• version=VERSION -- set OCaml version
• non-deterministic[=output|command] -- see "Non-deterministic tests" section
• dir=PATH -- set the directory where the tests should be run
• source-tree=PATH -- does nothing?
• file=PATH -- see the "File sync" section
• part=PART -- see the "File sync" section
• env=ENV -- see the "Named execution environments" section
• set-VAR=VALUE -- set an environment variable
• unset-VAR -- unset an environment variable

ocaml-mdx interprets shell scripts inside sh code blocks as cram-like tests. The syntax is the following:

• Lines beginning with a dollar sign and a space are commands and will be run in the shell.

• Multi-lines commands end by \ and continue with two spaces and a > sign on the next line:

   ```sh
   $ <line1> \
   > <line2> \
   > <line3>
   ```
  

• Commands support the heredoc syntax (<<):

   ```sh
   $ cat <<EOF \
   > hello\
   > world\
   > EOF
   hello
   world
   ```
  

• Lines beginning without a dollar sign are considered command outputs.

• Command outputs can contain ellipses: .... These will match any possible outputs (on zero, one or multiple lines).

• Arbitrary padding with whitespace is supported, as long as it is consistent inside a code block.

Here is an example of a markdown file using shell scripts inside code blocks, with a padding of 3:

```sh
   $ for i in `seq 1 10`
   1
   ...
   10
```

MDX will also consider exit codes when the syntax [<exit code>]is used:

```sh
$ exit 1
[1]
```

Note that nothing will be displayed when the exit code is 0 (e.g. in case of success).

MDX interprets OCaml fragments. It understands normal code fragments and toplevel code fragments (starting with a # sign and optionally ending with ;;). Arbitrary whitespace padding is supported, at long as it stays consistent within a code block.

Toplevel fragments interleave OCaml code and their corresponding outputs.

Here is an example of normal OCaml code:

```ocaml
print_endline "42"
```

Here is an examples of toplevel OCaml code:

```ocaml
# print_endline "42"
42
```

MDX is also capable of including content from files in fenced code blocks using the label file. OCaml files can be sliced using named blocks:

(* $MDX part-begin=partName *)
let meaning_of_life () =
  print_endline "42"
(* $MDX part-end *)

These can then be included in the document:

<!-- $MDX file=sync_to_md.ml,part=partName -->
```ocaml
```

Non-OCaml files can also be read and included in a block:

<!-- $MDX file=any_file.txt -->
```
```

However, part splitting is only supported for OCaml files.

Cram tests can be executed and checked with ocaml-mdx test <file.md>.

```sh
 $ for i in `seq 1 10`; do echo $i; done
 1
 ...
 10
 ```

If the output is not consistent with what is expected, <file.md>.corrected is generated.

To execute OCaml code and toplevel fragments, uses ocaml-mdx test <file.md>.

```ocaml
# print_endline "42"
42
```

If the output is not consistent with what is expected <file.md>.corrected is generated.

Non-deterministic Outputs

ocaml-mdx test supports non-deterministic outputs:

<!-- $MDX non-deterministic=output -->
```sh
$ <command>
<output>
```

In that case, ppx test <file> will run the command but will not generate <file>.corrected if the new output differs from the one described in the file. Use ocaml-mdx test --non-deterministic <file> to come back to the default behaviour.

Non-deterministic Commands

ocaml-mdx test supports non-deterministic commands:

<!-- $MDX non-deterministic=command -->
```ocaml
# Random.int 10;;
- : int = 5
```

In that case, ocaml-mdx test <file> will not run the command. Use ocaml-mdx test --non-deterministic <file> to come back to the default behaviour.

Alternatively, instead of passing the option it is also possible to set the environment variable MDX_RUN_NON_DETERMINISTIC to make MDX execute non-deterministic blocks. This is useful when not calling MDX directly but through other commands like dune or Makefiles etc. Use MDX_RUN_NON_DETERMINISTIC=1 ocaml-mdx test in this case.

Separate environments can be defined for blocks:

x holds the value 1 in the environment e1.

<!-- $MDX env=e1 -->
```ocaml
let x = 1;;
```

<!-- $MDX env=e1 -->
```ocaml
module M = struct let k = 42 let f x = x * k end;;
```

x holds the value 3 in the environment e2.

<!-- $MDX env=e2 -->
```ocaml
let x = 3;;
```

We can retrieve the value of x in environment e1:

<!-- $MDX env=e1 -->
```ocaml
# print_int x;;
1
- : unit = ()
# print_int M.k;;
42
- : unit = ()
# M.f;;
- : int -> int = <fun>
```

Blocks can be processed or ignored depending on the current version of OCaml.

For example to have a different outcome whether we are past OCaml 4.06:

<!-- $MDX version<4.06 -->
```ocaml
# let f x = x + 1
val f : int -> int = <fun>
# let f y =
  y^"foo"
val f : bytes -> bytes = <fun>
```

<!-- $MDX version>=4.06 -->
```ocaml
# let f x = x + 1
val f : int -> int = <fun>
# let f y =
  y^"foo"
val f : string -> string = <fun>
```

The available operators are <>, >=, >, <=, < and =. The version number can be of the following forms:

• *
• X
• X.Y
• X.Y.Z

Block can be processed or ignored depending on the current os_type.

For example, different blocks could be enabled depending on whether we are on Windows or not:

```ocaml
#require "unix"
```

<!-- $MDX os_type<>Win32 -->
```ocaml
# Unix.nice 0
- : int = 0
```

<!-- $MDX os_type=Win32 -->
```ocaml
# Unix.nice 0
Exception: Invalid_argument "Unix.nice not implemented".
```

The os_type values should be written in ASCII and are compared case insensitively.

Environment variables can be declared at the beginning of a block:

<!-- $MDX set-FOO=bar,set-BAR=foo -->
```ocaml
# print_endline (Sys.getenv "FOO")
bar
- : unit = ()
# print_endline (Sys.getenv "BAR")
foo
- : unit = ()
```

Those variables are then available in the subsequent blocks

```ocaml
# print_endline (Sys.getenv "FOO")
bar
- : unit = ()
```