Contributing to odoc

Please ask any questions you have about odoc, open any issues, offer feedback, etc. All of these are valued contributions :)

If you'd like specifically to work on the code of odoc, we hope that you will find the information in this file helpful.

Quick start: HTML and CSS

The odoc CSS is found at src/odoc/etc/odoc.css. It still needs work, and PRs are very welcome. You can edit CSS using your browser's developer tools. Then send us a PR for the same changes made to this file.

Working on the HTML is more involved. The main HTML generator is in src/html/ It operates on the types defined in Odoc_document.Types, which is an intermediate representation used by all output renderers. The type that describes an individual HTML page is Odoc_document.Types.Page.t.

To make edits to the HTML generation, run the following commands:

  1. Install requirements:

    • A recent version of HTML tidy (used for HTML validity testing) is required:

      # On MacOS (should be version 5.6.0 by the date of this writing)
      brew install tidy-html5     
      # Debian / Ubuntu
      sudo apt-get install tidy
    • A recent version of jq is required.

      # On MacOS
      brew install jq
      # Debian / Ubuntu
      sudo apt-get install jq
  2. Set up for development:

    git clone
    cd odoc
    opam pin add --no-action odoc .
    opam install --with-test --deps-only odoc
    opam install mdx
  3. Make changes to the code. To compile it,


    and then to run the tests,

    make test

    Changes to the HTML are likely to cause the tests to fail. See the section on testing below to understand how to update them.

  4. To test odoc against your own project, install it

    make clean
    opam install odoc

    Since odoc is pinned, this installs your modified version. Then you can run odoc in your project as normal:

    dune build @doc
  5. If all looks good, send odoc a PR :)


odoc uses a variety of different test types. We are slowly converging on using Dune's cram tests, though we still have many tests that aren't yet converted.

Cram Tests

The tests extensively use these for the model layer and are found in test/xref2. These consist of a directory called something.t, containing a file run.t. This file has shell-like syntax and usually runs odoc on some carefully crafted input files. For tests of the model layer it's often useful to use the binary odoc_print which can dump .odoc and .odocl files as JSON. This output can then be piped through jq to verify that values are as expected.

We try to make these test files describe the test and what's expected, which helps when the output isn’t what the test expected. This also means that the tests can serve as documentation of how things work. As an example, see the file test/xref2/multi_file_module_type_of.t/run.t

The tests work by executing the shell script snippets and then comparing the actual output with those in the run.t files. If these don't match, the difference is rendered as a diff. For example, if I change the way type declarations are printed and run dune runtest, I get the following output:

------ test/xref2/module_type_of.t/run.t
++++++ test/xref2/module_type_of.t/run.t.corrected
File "test/xref2/module_type_of.t/run.t", line 95, characters 0-1:
 |                ]
 |              },
 |              "T"
 |            ]
 |          },
 |          "Z"
 |        ]
 |      }
 |    }
 |  ]
 |Check that the expansion of 'T.Y' contains only 1 type
 |  $ jq ".[0].ModuleType.expr.Some.TypeOf.t_expansion.Some.Signature.items" < T_sig.json > T.Y_sig.json
 |  $ odoc_print m.odocl | jq "map(keys | .[0])" < T.Y_sig.json
 |  [
-|    "Type"
+|    "Toupe"
 |  ]

The intended response to this is:

  1. Check the diff. If the - line is correct, the code is broken. If the + line is correct, the test is broken.
  2. If the test is broken, run dune promote to replace the expected output with the current output.

Other Expect-Tests

Many of odoc's older tests are custom Expect-tests, similar to those run in the Cram test above, but that don't use Dune's promote workflow. As an example the parser tests in test/parser work in the following way:

  1. The tests run some code, e.g., the odoc parser on the string {e foo}.
  2. They take the output, in this case an AST representing "emphasized foo," and convert that output to a string, which will be an S-expression roughly like (emphasis (foo)).
  3. There is an expected copy of this S-expression in a file somewhere in the repo. If the S-expression from the code doesn't match the expected one, the test fails.

When one of these Expect-tests fail, the output is saved, so the developer can choose to replace the now-incorrect expected string. For these custom Expect-tests, the results may look like:

-- bold.000 [basic.] Failed --
in _build/_tests/bold.000.output:

{e foo}

--- expect/bold/basic.txt       2018-04-15 14:42:32.356895400 -0500
+++ _actual/bold/basic.txt      2018-04-15 17:36:26.812747400 -0500
@@ -2,5 +2,5 @@
    ((( (1 0) (1 7))
-      ((( (1 0) (1 7)) (bold ((( (1 3) (1 6)) (word foo)))))))))))
+      ((( (1 0) (1 7)) (emphasis ((( (1 3) (1 6)) (word foo)))))))))))
  (warnings ()))

To replace expected output with actual, run

bash _build/default/test/parser/_actual/

As with the Cram tests, the idea is to examine the diff to see if your code is broken or the test is broken. If the test is broken, the actual results may be promoted to the expected results by running the suggested command. If your code is broken, go and fix it!

We are slowly shifting these custom Expect-tests over to the Dune promote workflow.

Coverage Analysis

The odoc repo is set up for coverage analysis. This is most useful if you're writing new tests, and want to know what they’re actually touching.

To use it,

  1. Run make coverage. This will run the tests as normal, except at the end you’ll get a message like

    See _coverage/index.html

    You can then open _coverage/index.html and see the coverage of the code you’d like your new test to reach. However, it’s possible that it’s already covered "accidentally" by tests that are checking other properties. In which case, coverage analysis won’t be very useful :)

  2. Write new tests.
  3. Check coverage again.

CI Tests

odoc is tested by ocaml-ci and by GitHub workflows. One of these also does a coverage build, so we have up-to-date coverage stats on Coveralls.

The tests cover Esy and Opam builds on Windows, macOS, and Linux. The Linux tests cover all supported versions of OCaml. We strive to retain compatibility back as far as we can (currently 4.02) which is important for supporting

API Reference


The library odoc.loader is responsible for converting from the OCaml Typedtree representations to the internal representation.


The library odoc.model contains definitions of the internal types used to represent OCaml interfaces.

Resolution and Expansion

Resolution of Paths, Fragments and References, and Expansion of Modules and Module Types are handled by the odoc.xref2 library.

Intermediate Representation and Renderers

The generic documentation intermediate format is defined in the odoc.document library.

The three current renderers are implemented within the following libraries odoc.html, odoc.latex, and odoc.manpage.

CLI and Driver

The CLI for odoc and various helper functions for driving the process are contained in the odoc.odoc library.

Test and Internal Libraries

There are a couple of libraries used internally for testing - odoc.xref_test and odoc.model_desc.

Dependency Libraries

There are several dependency libraries that odoc uses, whose functions, type, and module declarations are essential for understanding how odoc works. See the driver page for details on how the documentation for these libraries are included.