Building from source

Cowl can be built and run on Windows, macOS and Linux. It has also been successfully deployed to a wider range of platforms, including tiny microcontrollers, with relatively minor build system setup. It can be compiled either as a static or dynamic library.


In order to compile the library, you will need at a minimum:

There are additional requirements depending on which additional components you would like to build or compile (e.g. readers).

Functional reader:

  • Flex version 2.6.4 or later.

  • Bison version 3.0 or later.


Sphinx is optional as Doxygen will already generate some form of HTML docs, though not as fancy as the ones you are viewing.

Downloading the sources

You can find Cowl’s code on its git repository. Please note that it contains submodules, so it is recommended that you clone it using the --recursive flag.

git clone --recursive


The following commands allow you to build Cowl:

# Generate the build system
cmake -B cmake-build -DCMAKE_BUILD_TYPE=Release

# [Optional] Edit build settings (build type, optimization options, etc.)
ccmake cmake-build

# Build the libraries and copy public headers into the output dir
cmake --build cmake-build --config Release

# [Optional] Build the documentation
cmake --build cmake-build --target cowl-docs


If you’re using CMake as your build system, you can link against Cowl by configuring your CMakeLists.txt file as follows:

# Assuming Cowl's source is under "lib/cowl"
add_subdirectory("lib/cowl" EXCLUDE_FROM_ALL)
target_link_libraries(your-target PRIVATE cowl)

For other build systems or if you are building directly through a compiler, please refer to their respective documentations. A relatively headache-free way to integrate Cowl involves compiling it and then linking against the built library, making sure the build system or compiler is aware of Cowl’s headers. Note that in this case you likely need to link against uLib as well.

Programming with Cowl

The easiest way to get started is by checking out the provided examples. However, in order to understand the principles behind the API, reading this section is strongly recommended.

API initialization

Before making any API call, you must invoke cowl_init(), which is needed in order to initialize the library’s internal state. Calling API members without initializing the API is undefined behavior.

Ontology deserialization

In order to query an ontology you must first deserialize it, which can be done via CowlManager. Cowl can use multiple readers, either built-in or provided by the user. For further information, refer to the related documentation.

OWL ontologies may import other ontologies, which may involve loading them from mass storage or retrieving them from the network. Cowl’s approach to imports reflects its focus on portability, so ontology retrieval is delegated to the end user.

Ontology queries

The core type of the API is CowlOntology, which consists of a set of CowlAxiom instances. The base mechanism for querying a CowlOntology is invoking its iterator member functions, which generally accept CowlIterator instances.

CowlIterator is a wrapper around a function that is called for every element matched by the query. By providing a generic context pointer, you can plug any custom data structure (loggers, collections, etc.), which allows for arbitrarily complex programmatic queries.

Ontology editing and writing

Ontologies can be created from scratch, or existing ontologies can be edited by adding or removing axioms, annotations and other constructs, as allowed by the CowlOntology API. Edited ontologies can then be written in any supported syntax (see the related documentation).

Under the hood

This section illustrates a few important low-level details that you need to know in order to correctly use the library.

Memory management

Cowl uses reference counting for memory management. Reference counts are increased and decreased via CowlObject::cowl_retain() and CowlObject::cowl_release(), respectively. The API docs are very explicit about which functions return retained instances, which you must release. If nothing is specified, then the returned instance is not retained, meaning its lifetime is generally tied to that of some other object. If you need to keep it alive after its owner has been deallocated, you must retain it.


Since the OWL 2 specification is highly hierarchical, the API makes extensive use of pseudo-inheritance for structs. Every data structure pseudo-inherits from CowlObject, whose concrete type can be queried via CowlObject::cowl_get_type(). Pseudo-inheritance allows you, as an example, to cast a CowlClass to CowlClsExp or CowlObject and back. Of course, if the API returns a base pseudo-class such as CowlClsExp or CowlObject, and you are unsure about its concrete subclass, you can check its type via get_type functions (e.g. CowlClsExp::cowl_cls_exp_get_type()) and cast accordingly. The API docs for type enumerations explicitly state the concrete type associated with every enumeration value.