=Paper=
{{Paper
|id=Vol-1554/PD_MoDELS_2015_paper_5
|storemode=property
|title=Reflecting on Model-based Code Generators Using Traceability Information
|pdfUrl=https://ceur-ws.org/Vol-1554/PD_MoDELS_2015_paper_5.pdf
|volume=Vol-1554
|authors=Victor Guana,Eleni Stroulia
|dblpUrl=https://dblp.org/rec/conf/models/GuanaS15a
}}
==Reflecting on Model-based Code Generators Using Traceability Information==
https://ceur-ws.org/Vol-1554/PD_MoDELS_2015_paper_5.pdf
Reflecting on Model-based Code Generators Using
Traceability Information
Victor Guana and Eleni Stroulia
Department of Computing Science
University of Alberta
Edmonton, AB. Canada
{guana, stroulia}@ualberta.ca
Abstract—Model-based code generators use model-to-model system concerns [4]. Lastly, model-to-text transformations take
and model-to-text transformations to systematize the construction this intermediate models, and produce code in a general-
of software systems. They integrate rule-based and template- purpose language, such as Java or C++, that can be compiled
based transformation languages to translate high-level specifica-
tions into executable code and scripts. Given the complexity and and executed [5].
heterogeneity of the underlying transformation languages, flexible Numerous challenges remain open to make the construc-
traceability tools are needed to collect and visualize information tion and maintenance of model-based code-generators less
about their architecture and operational mechanics. In this paper, tedious and less brittle to evolution [6] [7]. Despite of the
we present ChainTracker, a traceability analysis environment multiple research addressing the technical aspects of building
for model-based code generators. ChainTracker helps developers
to reflect on the composition of model transformations during code-generators, relative few researchers investigate the core
the different stages of their construction and maintenance. software-engineering problems behind their complex technol-
Furthermore, we describe a family of software-engineering tasks ogy ecosystems.
that developers have to complete during the construction of In this paper, we reflect on two challenges that affect devel-
model-based code generators, and how ChainTracker makes the opers when building and maintaining code generators, namely,
execution of these tasks less error prone and less cognitively
challenging. (a) the platform evolution of code generators, and (b) the
overwhelming complexity of code-generation architectures.
I. I NTRODUCTION The former refers to the difficulties that developers face when
Model-based code generators use models as a vehicle to the to-be-generated systems must fulfill new requirements and
capture a system’s specification. Domain-specific languages the generation artifacts must also synchronously evolve to
with high-level semantics enable developers to create mod- reflect those changes. The latter stems from the fact that non-
els using well-formed constructs [1]. In turn, these models trivial model transformations are hard to understand, given that
are input into model-transformation compositions to derive they heavily rely on textual scripts with cryptic operational
executable and/or deployment artifacts, such as source-code semantics, and are composed in complex transformation chains
text and configuration scripts [2]. By raising the level of that hard to mentally visualize and study.
abstraction for the software specification and making extensive The contribution of this paper is twofold. We first present
reuse of executable assets, model-based code-generators make ChainTracker, a visualization and trace-analysis environment
the construction of software systems less error prone, less that uses traceability information about model-transformation
expensive, and, potentially, less challenging to non-computer compositions, to support developers maintaining the complex
experts [3]. operational mechanics behind model-based code generators.
The role of model-based code generators is very similar to Second, we discuss how ChainTracker supports a number
that of compilers, translating a general-purpose programming of software-engineering tasks involved in the construction of
language to machine code. In this sense, a code generator model-based code generators, ranging from tracing tangled
for a domain-specific language uses model transformations code-generation artifacts, to summarizing information that
that inject execution semantics to the high-level models, and assesses the quality of the generation process.
produces code based on templates that have been previously
engineered for reuse. Domain-specific languages are typically II. R ELATED W ORK
associated with graphical or textual notations; using these During the life-cycle of a model-based code-generator, the
notations, developers can express system specifications in generated code is often manually improved [7] [8]. The goals
terms with application-domain semantics. Such definitions are of these code modifications are bug fixing, code-performance
ruled by metamodels that constrain how the concepts relate optimization, and functionality extensions [9][10]. This pro-
to each other. In the transformation process, model-to-model cess is known as platform evolution in model-driven en-
transformation languages are used to split, merge, or augment gineering [8]. Due to the high complexity of the model-
the information provided in the initial specification, potentially transformation compositions that live at the core of the code
producing multiple intermediate models that capture different generators, and the complexity of the generated code itself,
Fig. 1. ChainTracker - Main Screen (1) Transformation-composition Branch View; (2) Transformation Script Viewers; (3) Transformation In/Out Bindings
Information Tables; (4.a) Filter and Transformation-script Visualization Options; (4.b) Context-aware menu to isolate artifacts related to metamodel elements
or generated textual artifacts.
backwardly propagating code refinements to the generation specification of model-to-model transformations (such as in
architecture is a challenging task [11]. Some of the questions ATL [20]), and template-based model-to-text transformation
that developers have to answer when facing platform evolution technologies (such as Acceleo [21]) that allow the derivation
include, what portions of code have been changed in the of source-code files [5].
generated codebases? and, assuming that a code change ChainTracker is an integrated analysis environment de-
should indeed be propagated, what elements of the underlying signed to support developers of model-based code genera-
models and transformations should be revised?. tors by making model-driven engineering technologies more
Most state-of-the-art tools that support platform evolution efficient, less error prone, and less cognitively challenging.
in model-based code-generators tackle the problem using ChainTracker provides an integrated analysis environment
traceability information. Van Amstel et al. [12][13] have used to visualize models and heterogeneous model-to-model and
traceability information of model-to-model transformations model-to-text transformation compositions, through interac-
to visualize their operational mechanics. Other researchers tive canvases and source-code viewers. Furthermore, Chain-
[14] [15] [16] have used traceability information to assess Tracker collects and summarizes traceability information
the quality of model-transformation compositions. More re- about the scripts of a code generator, and helps developers
cently, in [17], Santiago et al. introduced iTrace, a tool that to identify maintenance hotspots that answer to platform-
uses a transformation-enrichment strategy, similar to Jouault’s evolution scenarios. In [22], we presented a general conceptual
proposal in [18], to collect traceability models at runtime. framework, generalizable to all rule-based and template-based
Nevertheless, to the best of our knowledge, current proposals model-transformation compositions, that formalizes how to
fail to provide comprehensive tools that support end-to-end collect and model traceability information in model-based
platform-evolution analysis, including analysis of the modi- code generators. In [23], we showcased ChainTracker as
fied generated code and collection of traceability information an integrated tool that implemented our generalizable frame-
from both model-to-model and model-to-text transformations. work using static-analysis techniques. We have now further
Furthermore, current proposals fail to provide effective tools enhanced ChainTracker to deal with complex model-to-model
that present the traceability information in an interactive and model-to-text transformation execution semantics that
and integrated tool, effectively supporting developers in the heavily rely on non-trivial OCL expressions [24]. The most
exploration of generation architectures. recent incarnation of ChainTracker uses natural-language
In summary, most of the existing proposals support very processing techniques to collect traceability information in
specific maintenance and construction scenarios. Furthermore, a lightweight and generalisable fashion for rule-based and
they often provide theoretical frameworks with demonstrative template based transformation languages built-in OCL. The
examples, rather than generic tools than can be shipped purpose of this paper is to showcase how ChainTracker pro-
and used by developers in fully-featured model-based code vides a comprehensible analysis environment to solve real
generators. software-engineering tasks that developers have to complete
during the construction of model-based code generators.
III. C HAIN T RACKER
Model-based code-generators are complex; they are built IV. S OLVING D EVELOPERS ’ TASKS U SING
using multiple technology platforms including language work- C HAIN T RACKER
benches that allow the definition of domain-specific languages ChainTracker is conceived to address three types of
[19], rule-based transformation languages that enable the software-engineering tasks around model-based code gener-
Fig. 2. ChainTracker - Main Screen (1) Transformation-composition Overview; (2) Transformation Script Viewers; (3) Metamodel Information Tables
to-text transformations, but also to project the visualizations
into the analyzed source-scripts using code-highlighters, thus
isolating pieces of code of interest (Figure 1.2)
Information summarization tasks require developers to
measure generic information from the code-generation archi-
tecture to identify branches of a transformation composition,
quantify the coverage of a model transformation, and measure
the size of the underlying metamodels. Summarizing infor-
mation about the code-generator enables developers to assess
and, potentially, optimize its overall design and correctness
[4]. Figure 2 presents ChainTracker’s Overview mode, in
which traceability information is used to provide a bird’s-eye-
view of the transformation composition. This view enables
Fig. 3. ChainTracker - Metamodel Coverage Metrics developers to abstract the complexity of individual and isolated
transformation scripts into a single picture that summarizes
the architecture of a given generation architecture. The blue
ators: 1. information discovery, 2. information summarization, nodes represent source and target artifacts of transformations
and 3. information filtering and isolation. Without Chain- (metamodels, in the case of model-to-model transformations,
Tracker, these tasks have to be performed manually over and text files, in the case of model-to-text transformations) and
transformation-source scripts [11]. the edges represent transformation modules that operate over
Information discovery tasks involve locating individual them.
elements of the code-generator’s architecture, i.e., individual Developers can click on the edges to pinpoint transformation
metamodel elements, transformation rules, and collections of rules that live inside of specific transformation branch or mod-
transformation bindings, inside the generator’s source scripts. ule modules (Figure 2 - yellow "sticky notes"). Furthermore,
The developer’s intent when performing this family of tasks is developers can zoom into a branch of interest (as shown in Fig-
to explore the code generator to identify its major components, ure 1). By clicking on individual nodes, developers can access
and to understand how the underlying transformation scripts summarized information about the coverage of a metamodel
are organized form a static point of view. at different stages of the composition (see Figure 2). Coverage
As a concrete example, let us briefly discuss PhyDSL, a information is important since it allows developers to assess
model-based code generator for physics-based games [25][26]. which parts of the source/target model elements have not been
Figure 1 illustrates how ChainTracker uses the traceability used in the generation of code, thus needing to be removed
information of PhyDSL to enable the investigation of how or included in the scope of the transformation modules [14].
source-metamodel elements are transformed into different Maximizing the coverage of model transformations makes
intermediate metamodels (including how their attributes are transformation modules less convoluted and less error prone
being manipulated by transformation rules) and into final while, at the same time, freeing metamodels from unused
textual files. ChainTracker enables developers to not only semantic constructs. In ChainTracker, coverage metrics are
explore the visualizations of the model-to-model and model- summarized in contextual pie-charts, where the in-coverage
metric reflects on the percentage of elements in a metamodel [2] K. Czarnecki, “Generative programming: Methods, techniques, and
that are effectively targeted by transformation bindings in applications tutorial abstract,” Software Reuse: Methods, Techniques, and
Tools, pp. 477–503, 2002.
the composition, and the out-coverage metric represents the [3] M. Mernik, J. Heering, and A. M. Sloane, “When and how to develop
percentage of elements in a metamodel used as a source in domain-specific languages,” ACM computing surveys (CSUR), vol. 37,
transformation bindings to either generate code, or to create an no. 4, pp. 316–344, 2005.
[4] A. Kleppe, “First european workshop on composition of model trans-
intermediate model by the transformation-composition under formations - cmt 2006,” Technical Report TR-CTIT-06-34, 2006.
analysis. [5] K. Czarnecki and S. Helsen, “Classification of model transformation ap-
Information filtering and isolation tasks involve connecting proaches,” in Proceedings of the 2nd OOPSLA Workshop on Generative
Techniques in the Context of the Model Driven Architecture, vol. 45,
and isolating elements of the code-generation architecture, no. 3. Citeseer, 2003, pp. 1–17.
in order to find relationships between metamodel elements, [6] R. France and B. Rumpe, “Model-driven development of complex
metamodel attributes, transformation bindings, and generated software: A research roadmap,” in 2007 Future of Software Engineering.
IEEE Computer Society, 2007, pp. 37–54.
lines of code. These tasks are performed when developers are [7] J. Hutchinson, J. Whittle, M. Rouncefield, and S. Kristoffersen, “Em-
assessing the impact of platform-evolution scenarios. Chain- pirical assessment of mde in industry,” in Proceedings of the 33rd
Tracker includes a variety of filtering mechanism to isolate International Conference on Software Engineering. ACM, 2011.
[8] A. Van Deursen, E. Visser, and J. Warmer, “Model-driven software evo-
metamodel elements and trace how elements relate across the lution: A research agenda,” in Proceedings 1st International Workshop
end-to-end transformation chains, including domain-specific on Model-Driven Software Evolution, 2007, pp. 41–49.
models, intermediate metamodels, and generated source-code [9] K. Bennett and V. Rajlich, “Software maintenance and evolution: a
roadmap,” in Proceedings of the Conference on the Future of Software
files (see Figure 1, at 4.a and 4.b). We are currently extending Engineering. ACM, 2000, pp. 73–87.
ChainTracker to include symbolic and static analysis of [10] V. Guana and E. Stroulia, “Backward propagation of code refinements
generated code, in order to automatically suggest when and on transformational code generation environments,” in Traceability in
Emerging Forms of Software Engineering (TEFSE), 2013 International
how refined portions of generated code need to be backwardly Workshop on, 2013, pp. 55–60.
propagated to the code-generators, using the traceability in- [11] V. Guana, “Supporting maintenance tasks on transformational code
formation already collected and visualized in ChainTracker. generation environments,” in Proceedings of the 2013 International
Conference on Software Engineering. IEEE Press, 2013.
In http:// hypatia.cs.ualberta.ca/ ~guana/ chaintracker.html you [12] M. van Amstel, A. Serebrenik, and M. van den Brand, “Visualizing
can find a video that showcases a developer conducting a traceability in model transformation compositions,” in Pre-proceedings
variety of the aforementioned tasks using ChainTracker. of the First Workshop on Composition and Evolution of Model Trans-
formations, 2011.
V. C ONCLUSIONS [13] J. von Pilgrim, B. Vanhooff, I. Schulz-Gerlach, and Y. Berbers, “Con-
structing and visualizing transformation chains,” in Model Driven
In this paper, we presented ChainTracker, a traceability Architecture–Foundations and Applications. Springer, 2008.
[14] J. Wang, S.-K. Kim, and D. Carrington, “Verifying metamodel coverage
collection and visualization analysis environment for model- of model transformations,” in Software Engineering Conference, 2006.
based code generators. ChainTracker enables developers to Australian. IEEE, 2006, pp. 10–pp.
solve three types of software-engineering tasks around model- [15] L. Kuzniarz, J. L. Sourrouille, and M. Staron, “Workshop on quality in
modeling,” 2007.
based code generators: 1. information discovery, 2. information [16] J. Gray, Y. Lin, and J. Zhang, “Automating change evolution in model-
summarization, and 3. information filtering and isolation, driven engineering,” Computer, vol. 39, no. 2, pp. 51–58, 2006.
when dealing with the platform evolution of code generators [17] I. Santiago, J. M. Vara, M. V. de Castro, and E. Marcos, “Towards the
effective use of traceability in model-driven engineering projects,” in
and the analysis code-generation architectures. The novelty Conceptual Modeling. Springer, 2013, pp. 429–437.
of our work is twofold. First, in ChainTracker traceability [18] F. Jouault, “Loosely coupled traceability for atl,” in Proceedings of
information is used to reduce the cognitive challenges de- the European Conference on Model Driven Architecture (ECMDA)
workshop on traceability, Nuremberg, Germany, vol. 91. Citeseer, 2005.
velopers face when using disparate model-driven engineering [19] M. Völter and E. Visser, “Language extension and composition with
technologies. Second, ChainTracker is an integrated environ- language workbenches,” in Proceedings of the ACM international con-
ment that uses traceability information to allow developers ference companion on Object oriented programming systems languages
and applications companion. ACM, 2010, pp. 301–304.
interactively explore model-transformation scripts. Our future [20] F. Jouault and I. Kurtev, “Transforming models with atl,” in Satellite
research agenda aims at conducting empirical studies with Events at the MoDELS 2005 Conference. Springer, pp. 128–138.
developers using ChainTracker, to accurately validate how [21] J. Musset, É. Juliot, S. Lacrampe, W. Piers, C. Brun, L. Goubet,
Y. Lussaud, and F. Allilaire, “Acceleo user guide,” 2006.
our tool and traceability collection techniques help developers [22] V. Guana, K. Gaboriau, and E. Stroulia, “Chaintracker: Towards a
when using model-driven engineering technologies in real comprehensive tool for building code-generation environments,” in Pro-
code generators. ceedings of the 2014 International Conference on Software Maintenance
and Evolution (ICSME). IEEE Press, 2014.
ACKNOWLEDGEMENTS [23] V. Guana and E. Stroulia, “Chaintracker, a model-transformation trace
analysis tool for code-generation environments,” in Theory and Practice
This work was supported by The Killam Trust, NSERC (the of Model Transformations. Springer, 2014, pp. 146–153.
Discovery and the IRC program), the GRAND NCE and IBM [24] J. Warmer and A. Kleppe, The object constraint language: getting your
models ready for MDA. Addison-Wesley Professional, 2003.
Canada. [25] V. Guana, E. Stroulia, and V. Nguyen, “Building a game engine: A tale
of modern model-driven engineering,” in Fourth International Workshop
R EFERENCES on Games and Software Engineering (GAS 2015), 2015.
[1] M. Voelter, S. Benz, C. Dietrich, B. Engelmann, M. Helander, L. C. Kats, [26] V. Guana and E. Stroulia, “Phydsl: A code-generation environment for
E. Visser, and G. Wachsmuth, DSL Engineering-Designing, Implement- 2d physics-based games,” in 2014 IEEE Games, Entertainment, and
ing and Using Domain-Specific Languages. dslbook. org, 2013. Media Conference (IEEE GEM), 2014.