Fork me on GitHub
Edit on GitHub << back to documentation

Celix Version 3 (Proposal)


The celix.h header contains a proposed API for Apache Celix version 3. This API is not fixed, but work in progress and should be considered a starting point for discussion.

There are a few reasons for a API change:

The current Apache Celix API is mapped from the OSGi API.

The problem is that OSGi is designed for Java, a language where classes are present runtime, reflection can be used, resource are controlled through garbage collection and classes are kept available when still needed(garbage collector) even if a bundle is uninstalled. This resulted in an API which is very lenient in object sharing/ownership and no real focus on resource locking/protection. This simply does not map wel to a C language. For a dynamic service framework in C, the focus should be in minimizing object sharing and when needed only for a controlled, well defined and short period. When resource sharing is needed - i.e. services - projecting access to these resources should dictate the design of the API. In this proposed API, when possible, object sharing is done through sharing ids in the form of a primitive (value copied) types instead of pointers. These ids can be used to (safely) retrieve info form the object if still present, which should make it more safe to use and most importantly less complex for implementation (think threads & locks). Also most service or service related updates must be handled with callbacks, to ensure correct usage of locking can be applied more easily.

OSGi API leaks too much details.

The OSGi API is more than 15 years old and has grown overtime, this shows. There are a lot of hoops to take to use a service in a correct way. The proposed Apache Celix API tries to minimize the needed objects, while still providing the same functionality. It was also designed for a more user friendly experience.

Smaller code base

Because the OSGi leaks to much details, it is also difficult to implemented; specifically mapped one to one from Java. A redesign of the API could lead to a smaller code base.

Single framework event thread design.

It would be beneficial to use single framework event thread, this would means that as a user it is ensured that all the callbacks are called from the same thread. Preventing issues that service can be added/removed simultaneously and the locking complexity which comes with that. With the current API this is more difficult (too many entries to shared resources).

Multiple languages

One of the key values of Apache Celix should be to deliver a sophisticated develop / design environment for more complex - i.e. composed out of modules) application/systems. For example a applications with combines some well proven C modules/libraries, with modern C++ modules and a splash of development support with python modules. And although Apache Celix is not there yet, it is one of it’s goals. The current API is not designed to support multiple languages and is difficult to port the different languages due to the “too much details” issue.

More focus on the primary ways to interact with a dynamic service framework.

For Apache Celix developers a focus is how to: work with services (register, tracking & monitor service interest) and how to easily create components which provides and depends on services. The proposed API focus on these issues as built-in support. For OSGi some of these functions where added add a later stage and this shows.

Current api has a strong focus on error codes

Java has exceptions, C has not. This has lead to an initial design that almost all function return a error code. While in principle completely correct, from a developer perspective it become cumbersome. Also taking in to perspective, than in many cases if an error is returned there no real way to handle this, the proposed API focus more on easy of use and lenient and silently accepting invalid input (i.e. NULL pointers or invalid ids).

Integral API

Again because the OSGi API has grown overtime, the API seems to be dispersed and not one integral API. The proposed API tries to remedy that.

Runtime Type Introspection

With OSGi for Java you have runtime type introspection (reflection) build-in the Java language, for C (and most other native languages) this is not the case. As effect is not possible to infer if services are compatible or to automagically make service remote / serialize types. The dynamic function interface library (dfi library) adds type introspection support, by using the extender pattern combined with so called type/interface descriptor files. Although not ideal, this can be used to runtime check service compatibility and automagically serialize types. If possible the dfi support will be optional and opt-out.

Eat your own dog food

A bit strange, but as a service oriented framework the OSGi specification for the core framework specification does not provide framework services. The proposed API moves some of the API for detailed info to framework services.

Integrated event admin

Services are great, but not everything fits perfectly solely with services. The event admin from the OSGi specification is a nice complement to the service oriented paradigm and as result event admin awareness is taken into account in the proposed API.

Missing prefix for symbols and headers

The current usage of prefixes is not sufficient. For example the celix framework has te be included as “framework.h” and has no prefix. This is not wise for a language without namespaces. The proposed API prefixes everything with celix_ and all include paths start with celix/. TODO discuss if a more generic OSGi like prefix is desirable, e.g. nosgi_ and nosgi/

const correctness

The proposed API add the use of const when applicable. The usage of const adds more semantics to the API and can prevent certain calls ( the callbacks) when they should not be used.

Support for a more static approach

Although Apache Celix is a framework for dynamic services, this does not mean that every module should be runtime install/uninstall-able. Supporting static modules and modules as “plain old libraries” can make it easier to use and deploy application using Apache Celix. In this proposal, bundle are still present, but just a way to install modules. Support for installing modules using static and shared libraries is added.

Apache Celix V3 CMake Commands

Apache Celix provides several CMake commands to be able to work with Apache Celix modules, bundles and deployments.

Generic Changes

  • symbolic_name is no longer needed nor expected. A module_name / bundle_name is enough and should be unique.
  • All command are prefixed with celix_.

Modules & Bundles

Modules are self-contained, autonomous and bounded software parts which can be combined to provide a greater/broader functionality.

Modules contain resources and required libraries (self-contained), use a activator to bootstrap itself (autonomous), can provide/require services to share/use functionality (bounded) and can import / export libraries so share types / routines (bounded)

There are two types of modules

  • Library Modules
  • Bundled Modules (aka Bundles)

Library modules are singular modules which do not import nor export libraries, cannot be bundled with additional required (private) libraries, and are build as shared or static libraries. Library modules can contain additional (embedded) resources. Static modules will only auto register on platforms where there is support for the __attribute__((ctor)) compiler attribute.

Bundle modules are modules which can import and/or export other libraries and can embed additional required (private) libraries. Bundle modules will be build as bundle file (ZIP file).


Modules can be build a “plain old libraries” using the CMake add_library command. Resources can be added using the celix_module_files command.

To be able to find the module resources, the module name used in the celix_moduleRegister call should be the same as the module name used in the CMake commnands. If no module name is set using the celix_module_name CMake command, the module name is equal to the name target.

celix_module_name(<module_target> <module_name>)

    files... DESTINATION <dir>
    [FILE_PERMISSIONS permissions...]
    [DIRECTORY_PERMISSIONS permissions...]
    [[PATTERN <pattern> | REGEX <regex>]
    [EXCLUDE] [PERMISSIONS permissions...]] [...])
 * Registers the module for all current framework and future framework that are created.
 * If no name is provided, the module will not be registered and an error will be logged.
 * When initializing the module, the framework will look for a <module_name>_resourceEntry and
 * <module_name>_resourceEntrySize symbol. If found, the content is assumed to be an embedded zip (or tar?) file.
 * the celix_modules_files CMake command will ensure that these symbols are created and linked to the module library.
 * @param moduleName The name of the module.
 * @param moduleVersion Optional, the version of the module.
 * @startFp Optional the start function which will be called if the module is going to be added and ready to start.
 * @startFp Optional the stop function which will be called if the module going to be removed and ready to start.
 * @opts Optional the additional options. The options will only be used during the registerModule call.
 * @return 0 if successful
int celix_moduleRegistration_registerModule(
        const char* moduleName,
        const char* moduleVersion,
        celix_module_registration_startModule_fp startFp,
        celix_module_registration_stopModule_fp stopFp,
        const celix_module_registration_options* opts);


Bundles are ZIP files which bundle one module. The CMake commands for addbundle commands are not changed with exception of the added celix prefix. The other command are renamed to module and a celix_ is added (e.g bundle_name -> celix_module_name)

    SOURCES source1 source2 ...
    [NAME bundle_name]
    [SYMBOLIC_NAME bundle_symbolic_name]
    [DESCRIPTION bundle_description]
    [VERSION bundle_version]
    [PRIVATE_LIBRARIES private_lib1 private_lib2 ...]
    [EXPORT_LIBRARIES export_lib1 export_lib2 ...]
    [IMPORT_LIBRARIES import_lib1 import_lib2 ...]
    [HEADERS "header1: header1_value" "header2: header2_value" ...]

    ACTIVATOR <activator_lib>
    [NAME bundle_name]
    [SYMBOLIC_NAME bundle_symbolic_name]
    [DESCRIPTION bundle_description]
    [VERSION bundle_version]
    [PRIVATE_LIBRARIES private_lib1 private_lib2 ...]
    [EXPORT_LIBRARIES export_lib1 export_lib2 ...]
    [IMPORT_LIBRARIES import_lib1 import_lib2 ...]
    [HEADERS "header1: header1_value" "header2: header2_value" ...]

    [NAME bundle_name]
    [SYMBOLIC_NAME bundle_symbolic_name]
    [DESCRIPTION bundle_description]
    [VERSION bundle_version]
    [PRIVATE_LIBRARIES private_lib1 private_lib2 ...]
    [EXPORT_LIBRARIES export_lib1 export_lib2 ...]
    [IMPORT_LIBRARIES import_lib1 import_lib2 ...]
    [HEADERS "header1: header1_value" "header2: header2_value" ...]


Descriptors are files which describe interfaces or messages which can be parsed by the dynamic function interface (dfi) library. The dfi library can be used for runtime ‘type introspection’. The descriptor files are a bit cryptic for human eyes, but relatively easy to parse.

With these descriptors the Celix framework can runtime compare if services and/or messages provider consumer combinations are compatible and make remote services / serialization of messages possible.

The celix_module_descriptor CMake will register header files for descriptor generation and inclusion in the module/bundle resources (under META-INF/descriptors/interfaces and META-INF/descriptors/messages).

The Celix project will have a dfi-gen target which will be used to generate the descriptors from the header files.

    header_file1 header_file2 ...


The add_deploy will be changed to celix_add_deploy and will accept library modules and bundle modules.

    [GROUP group_name]
    [NAME deploy_name]
    [LAUNCHER launcher]
    [DIR dir]
    [MODULES <module_or_bundle1> <module_or_bundle2> ...]
    [PROPERTIES "prop1=val1" "prop2=val2" ...]