In ongoing work, we advocate the integration of models and code as two equal constituents of software components. This approach can be followed to create flexible and self-adaptive software. We claim to extend this direction as a conceptual and technological basis for combining adaptation and evolution to achieve longevity. As a stimulus for discussions at the workshop, we present our work on model-integrating software components as one enabling technology for supporting a seamless software evolution process. In addition, we propose further enabling technologies and provide an initial set of related challenges.
Copyright c by the paper’s authors. Copying permitted for private and academic purposes.
Adaptation Adaptation Manager
Evolution Software Engineer
Dependency on enabling technologies
Activity Transition
ET1 Architectural
Capabilities
ET2
Modeling Capabilities
ET3
Reengineering
Capabilities … Certainly, there are open challenges related to each cycle and especially to their single activities. For example, the monitoring and analysis activities are crucial when assuming that they need to keep both cycles “in sync”. Hence, one may ask how to automatically decide when to switch from one cycle to the other, e.g., in cases where the adaptation manager cannot control an emerged situation.
While very relevant, such questions are out of the scope of this short paper, though. Instead, we like to point out that there is a need for enabling technologies to support the SSE process. We believe that they must at least facilitate architectural capabilities (ET1) and provide modeling capabilities (ET2) as well as reengineering capabilities (ET3). Next, we discuss these three enabling technologies in more detail and identify an initial set of related challenges.
We have already discussed some of the benefits of combining models and code within model-integrating software components to enhance flexibility and to offer software engineers a spectrum of possibilities to choose from when designing and implementing their systems [DEIE14]. As knowledge shall be shared between activities in the adaptation and evolution cycles, strategies and guidelines will be needed. Hence, we are convinced that providing the right architectural capabilities for software design and development is critical, but still requires some challenges to be tackled. Amongst others, questions related (i) to the distribution of models onto software components as well as (ii) to the role-based accessibility of these models within components must be answered.
Considering that models are used in both cycles – these may be even the very same models – we assume that an elementary set of modeling capabilities is required by both cycles. For example, activities in both cycles might attempt to modify the same model, e.g., during execution and realization. This yields the challenge to provide transaction concepts for models and the sketched knowledge base. Another example for capabilities required by both cycles are the specification and the realization of execution environments for models, since executable models can be used to achieve adaptivity and the evolution cycle must be capable of introducing and changing them. Whenever code of a software system is transformed to a model, either by adaptation or evolution, a corresponding execution environment needs to be instantiated. A related challenge is to find ways for efficiently performing this task.
Furthermore, unforeseen events need to be tackled in the evolution cycle. There exists a large body of knowledge in (model-driven) software reengineering that can be applied here. We assume that a solid set of reengineering capabilities to support this task will also be beneficial for adaptation. For instance, evolution may require to reverse engineer and restructure a software system, or to generate source code. The same capabilities can be beneficial for adaptation purposes at runtime. Challenges include how to specify and implement reengineering methods that are reusable under varying situational context, namely adaptation and evolution.
Indeed, we are convinced that there are many more challenges to be discussed and we hope to identify a more concrete list of them during the workshop.
A decade ago, challenges for software evolution were discussed by Mens et al. [MWD+05]. The topic is still very relevant, today. Existing architectural approaches to develop adaptive software, like MoCos, can be beneficial to achieve longevity. Models (at runtime) – e.g., such as those proposed by Blair et al. [BBF09] – can play a key role at the intersection between adaptation and evolution cycles. However, we believe that these directions alone are not sufficient and claim that a set of enabling technologies that facilitate at least (i) architectural capabilities, (ii) modeling capabilities and (iii) reengineering capabilities is a fundamental prerequisite to achieving the greater vision of long-living software. This short paper intends to provide a starting point and context for discussion. Acknowledgements This ongoing work is a result of the MoSAiC project, supported by the Deutsche Forschungsgemeinschaft (DFG) under grants EB 119/11-1 and EN 184/6-1. [BBF09] [DEIE14] [HHJ+13]
Wilhelm Hasselbring, Robert Heinrich, Reiner Jung, Andreas Metzger, Klaus Pohl, Ralf Reussner, and Eric Schmieders. iobserve: Integrated observation and modeling techniques to support adaptation and evolution of software systems. Technical Report, 1309, 2013. [RB00]