Concurrency-aware Executable Domain-Specific Modeling Languages as Models of Concurrency

Abstract

-To deal with the increasing complexity of mod- concerns based on a MoC formalism). The latter are captured ern highly-concurrent systems, the GEMOC approach for in the Semantic Rules, which extend the AS with the data and concurrency-aware eXecutable Domain-Specific Modeling Lan- operational concerns of the semantics. Both are connected by sgeumagaensti(cxsDmS ModLels,) tphreopcoosnecsutroremncaykecoenxpcelircnits, uinsinthge aopMeroadtieolnoafl a third model called the Communication Protocol. Concurrency (MoC). This separation of concerns enables re- The approach from [1], [2], [3] relies on the Event Structures finements (e.g., for sequential or parallel execution platforms) MoC [6], in which partial orders [7] are defined over a set and analyses (e.g., to assess behavioral properties like deadlock- of events. The language-level model, the MoCMapping, is freeness) of the concurrency concerns. This approach initially built using EventType Structures [8]. However, the use of tphreovbideesst ofintlyfoornealMlocoCn:cEurvreenntcSytrpuacrtaudriegsm.Bsuutstehdis inMoxCDSisMnLost, a particular MoC for an xDSML is an important decision: resulting in complex models which are difficult to maintain or it defines which formalism will be used to represent the analyze. Moreover, extending the approach with new MoCs is concurrent aspects of the executable models. Depending on complex: many elements must be integrated, and fit into the the MoC used, different concurrency-aware analyses (usually APIs used by the implementation. We propose to seamlessly relying on existing external tools) may be performed to assess odfefithnee caonndcuirnrteengcrya-taewnaerwe xMDoSCMsLtharpopurgohacah, reencaubrlsiinvge tdheefiunsiteioonf behavioral properties of the models. Moreover, not all MoCs previously-defined xDSMLs as MoCs. This allows xDSMLs to are good fits for all xDSMLs. Depending on the xDSML's always rely on an adequate MoC which also comes tooled with concurrency paradigm, MoCs may be more or less adequate. the generic execution and debugging facilities provided by the Providing only the Event Structures MoC effectively limits concurrency-aware approach. We illustrate this on the definition the approach, or complicates the modeling of some xDSMLs. woforfkUbMenLchi.n the GEMOC Studio, an Eclipse-based language Integrating new MoCs alongside Event Structures is complex, Index Terms-Domain-Specific Modeling Languages; Models since it requires modifying the language workbench in which of Concurrency; Executable Metamodeling the concurrency-aware approach is implemented. In this paper, our contribution consists in a recursive I. INTRODUCTION definition of the concurrency-aware xDSML approach, in Modern complex systems (e.g., Cyber-Physical Systems, which a previously-defined concurrency-aware xDSML can be Internet of Things, etc.) are highly-concurrent and executed used as the MoC of another xDSML, based on the composite on increasingly-parallel platforms (e.g., many-core CPUs, design pattern. This greatly reduces the complexity of defining GPGPU pipelines, etc.). eXecutable Domain-Specific Modeling and integrating new MoCs into the approach. Moreover, the Languages (xDSMLs) can be used to specify such systems, MoC can also benefit from any of the generic execution but in traditional executable metamodeling techniques, the con- and debugging facilities provided for free by the language currency concerns are spread throughout the whole semantics, workbench implementing the concurrency-aware approach. The making difficult its identification, refinement or analysis. A contribution to executable metamodeling techniques is thus novel concurrent executable metamodeling approach has been twofold: we extend the existing concurrency-aware xDSML proposed and refined in [1], [2], [3]: the GEMOC concurrency- approach with the possibility to define and integrate new MoCs, aware xDSML approach. It proposes to make explicit, in the effectively allowing xDSMLs to always rely on an adequate operational semantics of xDSMLs, the concurrency concerns MoC; and we bridge the gap between MoCs (usually defined using a Model of Concurrency (MoC) (e.g., Petri nets [4], the as formalisms dedicated to the modeling of concurrent systems) Actor model [5], Event Structures [6], etc.). In this approach, and software language engineering, effectively providing a comthe concurrency concerns are separated from the data concerns. mon interface to MoCs (i.e., as concurrency-aware xDSMLs). The former are captured in the Model of Concurrency Mapping The rest of this paper is organized as follows. First, (MoCMapping), which specifies how a MoC is systematically in Section II, we illustrate the concurrency-aware xDSML used for models conforming to the Abstract Syntax (AS) approach on an example language, fUML. In Section III, we of the language (i.e., it generates the model's concurrency first show the importance of the adequacy of a MoC for an

-

xDSML. We explain why directly integrating MoCs into the approach is complex and time-consuming. We then describe our recursive definition of the approach, illustrated by defining a new xDSML that we will use as the MoC of fUML. The upsides and limitations of our contribution are presented in Section IV. Section V focuses on our implementation in the GEMOC Studio’s Language Workbench. Finally, we discuss related work in Section VI; then conclude and give perspectives for future work in Section VII.

II. CONCURRENCY-AWARE XDSMLS ILLUSTRATED

We apply the seminal concurrency-aware xDSML ap

proach [ 1 ], [ 2 ], [ 3 ] to the Foundational Subset for Executable UML Models (fUML) [ 9 ]: an executable subset of UML Activities.

A. Structural Elements

In Model-Driven Engineering (MDE), a language’s Abstract

Syntax is usually modeled as a metamodel enhanced with static semantics, expressed using the Meta-Object Facilities (MOF) and Object Constraint Language (OCL) from the OMG. Figure 1 shows an example fUML Activity composed of nodes (ActivityNode) of various natures, connected by edges (ActivityEdge). This example models drinking something while talking. The former consists in checking what is available on the table (“CheckTableForDrinks” returns at random “Coffee”, “Tea” or “Neither”), and then executing one of the three branches depending on what was found on the table. In this example, concurrency takes place in the branches of

B. Separation of Concerns in the Semantics

The concurrency-aware approach introduces a separation of

concerns in the operational semantics [ 10 ] models.

The data and operational concerns are first captured in the Semantic Rules. The Execution Data define the dynamic data in the language (i.e., current state of an automaton, current tokens in a place, current value of a variable, etc.) and how they evolve during the execution (i.e., through fired transitions, executed instructions, etc.). In fUML, Tokens are created and consumed by the nodes. The Execution Data are thus the reference currentTokens weaved in the fUML metamodel, from the ActivityEdge concept to the Token concept. The execution of nodes is performed by an Execution Function weaved onto the ActivityNode concept, implemented by the execute() operation.

The concurrency concerns are then captured in the Model of Concurrency Mapping. So far, the approach only supports Event Structures so the MoCMapping is an EventType Structure model. In fUML, the core of the MoCMapping essentially consists in declaring an EventType representing the execution of an ActivityNode, and in constraining it such that for every ActivityEdge, its source is executed before its target. This language-level model is unfolded for each fUML model to generate the Model of Concurrency Application (MoCApplication, in our case, an Event Structure) representing the model’s concurrency concerns.

Finally, both concerns are connected by another languagelevel model called the Communication Protocol that maps the MoCTriggers (i.e., the abstract actions of the MoCMapping, that is the EventTypes of an EventType Structure) and the Execution Functions of the Semantic Rules. It also includes the Feedback Protocol, required in fUML for DecisionNodes, which was motivated and illustrated in [ 3 ].

Figure 2 shows an overview of the approach we just described, as a Class Diagram. the ForkNode: the “Talk” node can be executed simultaneously with, or interleaved with, any of the nodes of the drinking part of the activity. Implementations usually hard code this decision, or rely upon the underlying execution platform. With the concurrency-aware approach, all the valid possibilities are explicitly specified, thus enabling the use of concurrency-aware analyses, allowing the management of semantic variations [ 2 ] or the refinement of the language for specific execution platforms (e.g., sequential, highly-parallel, etc.).

III. DEFINING AND INTEGRATING ADDITIONAL MOCS

This section first motivates the use of other MoCs besides

Event Structures and shows why integrating new MoCs directly into the approach is complex and time-consuming. Then, we propose to define and seamlessly integrate new MoCs thanks to a recursive definition of the concurrency-aware xDSML approach. This proposal is illustrated by the use of a new MoC dard in both natural language and a reference implementation based on the notion of Threads instead of Event Structures to in Java1. Neither description is particularly adapted to the model fUML. use of Event Structures, so the resulting MoCMapping was A. Adequacy of Models of Concurrency complex. Instead, Petri nets [ 4 ] can be considered, since it originally inspired UML Activity Diagrams. Another practical

Several criteria can be used to assess the adequacy of a alternative is to rely on the notion of Thread: a conceptual unit MoC to model an xDSML’s concurrency concerns. First, there for computation, also known as green threads, or user threads, is the conceptual proximity with the xDSML’s concurrency as opposed to the operating system’s threads (or kernel threads). paradigm, or with the systems modeled with the xDSML. In The mapping between these two depends on the implementation “Why Do Scala Developers Mix the Actor Model with Other of the runtime of the concurrency-aware xDSML approach. Concurrency Models?” [ 11 ], one of the reasons why a Scala This MoC is convenient for fUML because it corresponds to code base integrates other MoCs besides the Actor model the MoC provided in Java by the threading API, used for its promoted by Scala, is because of inadequacies of the actor reference implementation. model. Using an inadequate MoC increases the chance of Let us detail how the MoCMapping of fUML can be specified creating deadlocks and data races. Another important factor to using the notions of threads with a set of instructions, and the account for is the language designer’s experience with MoCs. possibility for a thread to start another thread or to wait for This was also one of the reasons in [ 11 ] for the mixing of MoCs the end of another thread. An Activity is mapped to the main in Scala code bases. This can lead to an antipattern known as thread in charge of the activity’s InitialNode and FinalNode. the Golden Hammer: “if all you have is a hammer, everything The execution of nodes in sequence in an activity can be looks like a nail”. These two criteria may complicate the use represented by a sequence of instructions in a thread. For of a MoC, ultimately making the MoCMapping difficult to every ForkNode, one thread is created per branch. It is in specify or maintain. charge of executing the associated branch’s nodes. Executing

Moreover, one of the main benefits of the concurrency-aware the corresponding JoinNode is possible when all these threads approach lies in being able to formally verify properties of have finished executing their instructions. systems based on their MoCApplication. Which properties and verification technologies remains at the discretion of the MoC Figure 3 shows the application of such a mapping on the used. The formal method community has developed tools and example fUML Activity of Figure 1. The main thread consists methods for such verifications, e.g., Petri nets [ 4 ] are commonly in executing the InitialNode and the ForkNode, which starts the used for model-checking activities. This may be an important two sub-threads. It then waits for the sub-threads to complete, factor in selecting which MoC to use, for instance if particular before allowing the execution of the JoinNode and FinalNode. safety properties must be enforced (e.g., in critical systems). In the first sub-thread, corresponding to the drinking part of the activity, the corresponding branch is executed as a sequence B. Integration Cost of Models of Concurrency of instructions. After the DecisionNode and the evaluation of

To integrate new MoCs into the concurrency-aware approach, the guards, depending on the results retrieved, only one of the two metalanguages have to be considered. First, the MoC drinking nodes is executed before the MergeNode is executed. itself, to which the MoCApplication is conforming, with its On the second sub-thread, there is just one node corresponding editor, runtime, and possibly verification tools. Then, the to talking.

MoCMapping formalism, to which the MoCMapping of an xDSML is conforming, with its editor and generator (used to 1https://github.com/ModelDriven/fUML-Reference-Implementation produce the MoCApplication). Besides the merely technical difficulties of integrating these languages and their associated tools (provided the language workbench used is even opensource, or implemented with the abstractions permitting its extension in the first place), the main issue is that the notion of MoCMapping is the main novel artefact of the concurrencyaware xDSML approach. This means that for existing MoCs, the MoCMapping formalism has to be fully defined and tooled before being integrated into the approach, which is complex and time-consuming.

To overcome these issues, we propose a generic approach to define and seamlessly integrate new MoCs into the approach, enabling the tailoring of the MoCs used in the definition of concurrency-aware xDSMLs.

C. A Thread-based MoC for fUML

We illustrate our proposal with the definition and use of

another MoC for fUML, whose semantics is given in the stan

D. Introducing a Recursive Definition of the Concurrency-aware xDSML Approach

More generally, we propose to leverage a previously-defined

concurrency-aware xDSML as the MoC for another xDSML. This means that the concurrency concerns of a model are expressed as a model conforming to another xDSML, more appropriate to capture these concerns.

At the language level, this means that the MoCMapping as described in Section II is now a model transformation from the xDSML to the xDSML used as MoC. The model resulting from the transformation (the MoCApplication) is however not semantically equivalent to the original model: it only represents its concurrent aspects. The MoCTriggers are the

Mappings (from the Communication Protocol) of the xDSML

used as MoC. And since an element in a model may result in several elements in the resulting model (e.g., ForkNodes are transformed into several ”StartThread” instructions), we also need to be able to exploit the trace of the transformation to disambiguate the Communication Protocol of our xDSML.

More formally, we consider two concurrency-aware xDSMLs, LDomain and LMoC . We will detail our approach to specify LDomain using LMoC as MoC, illustrated in the case where the former is fUML and the latter is an xDSML capturing the notions of threads and their instructions. LMoC is considered as already defined, which means that it has been specified either as shown in Section II, or as is being shown in this section. Figure 4 gives an overview of our approach as a class diagram. It relies on two additional models presented in the “Concurrency-aware xDSML Recursive Definition” package of the class diagram. The rest of this section focuses on these two models.

1) Abstract Syntax Transformation: The first one is

a Transformation from LDomain to LMoC , denoted as TDomain!MoC . It specifies how the concurrency concerns of LDomain are represented using LMoC . It effectively corresponds to the MoCMapping of LDomain, as it maps the AS of LDomain to the structure of the formalism used as MoC. For an input model MDomain, its output is MMoC .

For our example, we must first consider the definition of a concurrency-aware xDSML with the notion of Threads. In such an xDSML, we define a ThreadSystem as composed of Threads, with one of them considered as the main one. Each Thread has a number of Tasks which can be of different nature (execution, disjunction, conditional, etc.), in particular they can consist in starting or joining other threads. Inside a Thread, Tasks are executed sequentially. Threads are concurrent by nature, so if several are running at the same time, they can execute their instructions in parallel or in any form of interleaving. Joining on another thread waits for the selected thread to have all its instructions executed. Disjunctions are tasks for which only one of the two operands (Tasks) is executed, while Conditionals are executed if all their conditions (other Tasks) have been executed previously. The main Mapping of interest in the Communication Protocol of this Threading xDSML pertains to the execution of a

Task, denoted as ExecuteTask. Once we have designed

this language using the concurrency-aware approach, we can specify TfUML!T hreading. For our example model of Figure 1, the resulting model corresponds to the right half of Figure 3.

2) Trace of the AS Transformation: When a concept of

LDomain is transformed into several concepts of LMoC , we need a means to specify which one should be used as the trigger for the execution of the LDomain concept. For instance, a ForkNode is transformed into as many Tasks as it has branches. We need to be able to define which Task’s execution will be mapped to the ForkNode’s execution. In order to do so, we thus need an additional language-level model, which consists in an excerpt from the metamodel of the trace of the AS Transformation. This excerpt is called the Projections of LDomain, designated as PDomain!MoC . It specifies, for a concept of LDomain, into which concept(s) of LMoC they are transformed (through TDomain!MoC ) and with which purpose(s), through labels. This allows us to specify, for instance, that the execution of the ForkNode is mapped to the execution of the first Task representing the execution of one of its branches.

This model is then used by the Communication Protocol of LDomain, in order to generate its model-level counterpart without ambiguities. This model, denoted as PfUML!T hreading for fUML, as well as its use by the fUML Communication Protocol model, are both illustrated in Section V using our metalanguage implementations in the GEMOC Studio.

IV. DISCUSSION

We discuss our contribution’s benefits and drawbacks.

A. Modularity

The modularity of the initial approach is not disrupted. The

data and concurrency concerns are still separated. In fact, it even favors the reuse of the AS and Semantic Rules of an xDSML by using different MoCs, for instance to compare two MoCs for the same language. Reversely, concurrency-aware xDSMLs can be reused as MoCs for other xDSMLs.

B. Concurrency-aware Analyses

Depending on the available tools, or expected behavioral properties, the language designer may choose to use one or another MoC for their xDSML. For instance, Petri nets [ 4 ] are commonly used to assess liveness or safety properties. Our approach also remains ultimately rooted in Event Structures. By transitivity, analyzing the concurrency concerns of a model can also be done through its underlying Event Structure [ 12 ]. Our contribution has thus provided an additional hook for potential analyses of executable models.

C. Facilitated MoCMapping Modeling

By facilitating the integration of new MoCs defined as

xDSMLs, we allow concurrency experts to provide a large MoC library. Thus, we allow language designers to use the most appropriate MoC for the xDSML being modeled. This is similar to how DSLs are used for the dedicated abstractions they propose: some formalisms are more adapted for the modeling of some concurrency paradigms. This adequacy translates into an overall simpler or more maintainable MoCMapping model. Moreover, instead of having to learn and master an additional metalanguage for the MoCMapping (e.g., EventType Structures), the language designer now only has to learn the Projections metalanguage (which is small and straightforward), while using a model transformation for the MoCMapping. Execution and debugging facilities of the concurrency-aware approach are also made available for free for the xDSML used as MoC.

D. Unified Interface for MoCs

MoCs are usually defined in many different manners, but usually quite “informally”, either as “formalisms”, or through language or framework constructs (e.g., Erlang actors [ 13 ], Scala’s Akka actors [ 14 ]). Our contribution allows any concurrency-aware xDSML to be used as MoC. MoCs are thus all modelled as xDSMLs in a uniform way and it is the use made of an xDSML that determines whether it corresponds to a MoC or not. In other words, “MoC” is a role played by an xDSML, not its nature.

E. Comparison with translational semantics

Our contribution bears resemblance with translational se

mantics (i.e., a language’s execution semantics is defined by a translation to another well-defined language). Indeed, we propose to define a transformation from LDomain to LMoC : TDomain!MoC . However, the purpose of this transformation is very different from that of translational semantics. In our approach, the source model (MDomain, conforming to LDomain) and the target model (MMoC , conforming to LMoC ) are not semantically equivalent. MMoC is only a representation of the concurrency concerns of MDomain, using LMoC as a formalism; whereas in translational semantics, the intention of the transformation is to produce a semantically equivalent model. The data treatments done in the Semantic Rules of LDomain are never translated in terms of concepts of LMoC , and only the concurrency concerns of LDomain are transformed into LMoC . This is quite similar to the abstraction phase done during the modeling of a system in order to assess the correctness of the concurrency concerns using model checkers.

V. IMPLEMENTATION

Our proposal has been implemented in the GEMOC Studio2, an EMF-based language workbench. The full execution of the example fUML Activity is available as a video at http: //gemoc.org/exe16/. An archive file also provides the GEMOC Studio with our implementation of fUML based on Threads, as well as the source files for these xDSMLs.

A. Existing Elements

The Abstract Syntax is captured as an Ecore metamodel.

The Semantic Rules can be weaved into the AS by defining aspects using the Kermeta 3 Action Language (K3AL) [ 15 ]. EventType Structures are specified using a combination of MoCCML [ 16 ] and ECL [ 8 ]. The Communication Protocol is specified using a dedicated metalanguage called the GEMOC Events Language (GEL) [ 3 ].

2http://www.gemoc.org/studio

B. Projections, Transformation and concurrency concerns are either embedded in the metalanguages Communication Protocol provided by the approach used, or implicitly inherited from the

To specify PDomain!MoC , we have designed a small dedi- underlying execution platform. In concurrency-aware xDSMLs, cated metalanguage. Listing 1 shows the projections for fUML, they are made explicit at the language level through the use PfUML!T hreading. fUML ActivityNodes are transformed into of a MoC.

Tasks. When a Task is executable, then the corresponding Then, concurrency theory has studied Models of Concurrency ActivityNode (if there is one) is also executable. In the same like Petri nets [ 4 ], the Actor model [ 5 ] or Event Structures [ 6 ] manner, some Tasks correspond to the evaluation of the guard for a long time [ 25 ]. Different MoCs typically split a task in of an ActivityEdge, while some others represent the fact that a different manners and the communication and collaboration branch may or may not be executed, based on the result of its between the computing entities (e.g., threads, actors, etc.) is guard (more details in [ 3 ]). done in different ways (i.e., in terms of data-sharing, scheduling, cooperation, etc.). A unification of MoCs has been proposed through the use of the “tagged signal” model [ 26 ] or of category

Listing 1. fUML projections on Threading. theory [ 27 ], but they do not focus on defining and integrating 21 PLroajnegcutiaognesP:rojection Proj_Execution: new MoCs. For instance in Ptolemy3 [ 28 ], adding new MoCs 3 fuml.ActivityNode projected onto threaded.Task end (called “directors”) requires complex changes in its source 45 Lfanugmula.gAecPtroijvecittioynEdPgreojp_rEovjeaclteudatoinoton:threaded.Task end code. 6 LanguageProjection Proj_MayExecute: In “Why Do Scala Developers Mix the Actor Model with 87 Lfanugmula.gAecPtroijvecittioynEdPgreojp_rMoajeycNteodtEoxnetocutther:eaded.Task end Other Concurrency Models?” [ 11 ], the authors analyze Scala 9 fuml.ActivityEdge projected onto threaded.Task end code-bases and interview their developers to determine how 10 end often and why the Actor MoC was not the only one used.

Reasons were categorized into three groups: a) inadequacies of

TDomain!MoC can be specified using any Model to the actor library, b) inadequacies of the MoC, and c) inadequate Model (M2M) transformation language [ 17 ] and any general- developer experience. Concurrency-aware xDSMLs move a) purpose programming languages relying on an appropri- and c) away from the system designer to the language designer, ate model management library such as Java with EMF’s greatly reducing the number of users concerned with the correct APIs. However, the transformation must not only transform uses of MoCs. Thanks to our contribution, b) is removed as MDomain into a corresponding MMoC , but it must also any MoC can be used for any xDSMLs, as long as they have generate the model-level models of PDomain!MoC , denoted been modeled as concurrency-aware xDSMLs. This enables as PDomainModel!MoCModel, and used during the generation the language designer to use the best MoC for the concurrency of the model-level Communication Protocol. Indeed, we have concerns of each xDSML. extended GEL with the capacity to reference projections from Finally, we would like to point out the similarity of our PDomain!MoC when defining the Mappings. Listing 2 shows proposal with the design pattern identified by Bran Selic as the the Communication Protocol for our implementation of fUML. “Recursive Control” pattern in [ 29 ]. In our case, LMoC and its runtime is the internal control for LDomain and its runtime.

Since LMoC itself can be specified using another xDSML, this effectively corresponds to an application of the “Recursive Control” pattern.

VII. CONCLUSION AND PERSPECTIVES

Listing 2. Communication Protocol for fUML. 1 DSE ExecuteActivityNode: 2 upon event ExecuteTask with Proj_Execution 3 triggers ActivityNode.execute blocking end 4 5 DSE EvaluateGuard: 6 upon event ExecuteTask with Proj_Evaluation 7 triggers ActivityEdge.evaluateGuard returning bool 8 feedback: // Presented in more details in [ 3 ]. 9 [bool] => allow event ExecuteTask with Proj_MayExecute 10 default => allow event ExecuteTask with Proj_MayNotExecute 11 end 12 end

The concurrency-aware xDSML approach advocates making explicit, in the operational semantics, the concurrency concerns using a Model of Concurrency. However, only one MoC, namely Event Structures, was provided initially. We have shown that not all MoCs are good fits for all concurrency paradigms, and that manually integrating new MoCs into the approach is complex and time-consuming as it requires two metalanguages VI. RELATED WORK (for the language and model levels) as well as their tools.

To ease this activity, we have proposed an extension

The concurrency-aware xDSML approach we have extended which enables the use of previously-defined concurrencybrings together two fields of research. aware xDSMLs as MoCs. This contribution relies on two

First, language Workbenches [ 18 ], such as MPS [ 19 ] or main changes in the language models: an abstract syntax MetaEdit+ [ 20 ] traditionally focus on the language syntaxes. transformation to specify how the xDSML’s concurrency Executability of DSMLs is provided using “meta-programming” concerns are encoded in the xDSML used as MoC; and like Rascal [ 21 ] or Spoofax [ 22 ]; or “executable metamodeling” like xMOF [ 23 ] or the K Framework [ 24 ]. In both cases, the 3http://ptolemy.eecs.berkeley.edu/ the Projections, a part of the transformation metamodel, that accounts for cases where an xDSML concept is transformed into several MoC concepts. The former can be specified using any classical model transformation languages, while we have devised a dedicated metalanguage for the latter. Our contribution has been implemented in the GEMOC Studio, an Eclipse-based language workbench, and illustrated on fUML defined using an xDSML capturing the notions of threads with instructions. We have made available a video showing the execution of an example fUML Activity, and the sources for the two xDSMLs. By defining MoCs as concurrency-aware xDSMLs, we give them a systematic structure, enabling their use at the language-level for the modeling of other concurrencyaware xDSMLs. In particular, it enables the use of the MoC that is the best fit for the concurrency paradigm of the language being developed. It also eases the development of an xDSML, since the model-level application of the MoC is simply a model conforming to an xDSML, that can be executed, debugged and animated like a regular model. Moreover, different formal behavioral properties can be assessed on executable models depending on the MoC used by the language.

Although any concurrency-aware xDSML can be used as a MoC, the concurrency theory community has studied in details a large number of MoCs such as the Actor Model [ 5 ] or Petri nets [ 4 ]. We plan to provide reference implementations for these ones in a MoC standard library, including Event Structures to bootstrap our approach. Afterwards, existing tools around these formalisms, such as model-checking tools, can be integrated seamlessly in our approach. Still, the approach remains rooted in the seminal MoC (i.e., Event Structures), so higher-order transformations could be used to verify domain properties using the underlying MoC, while translating their results back to the domain [ 30 ]. Finally, even though we have considered concurrency-aware xDSMLs as language models for the implementation of more efficient tools, we plan to study how code generation or scheduler synthesis could be used to generate more efficient implementations of concurrency-aware xDSMLs.

ACKNOWLEDGMENTS

This work is partially supported by the ANR INS Project GEMOC (ANR-12-INSE-0011). REFERENCES

References

[1] B. Combemale , J. Deantoni , M. Vara Larsen , F. Mallet , O. Barais , B. Baudry , and R. France, “Reifying Concurrency for Executable Metamodeling,” in SLE' 13 .

[2] F. Latombe , X. Cre´gut , J. Deantoni, M. Pantel , and B. Combemale , “ Coping with Semantic Variation Points in Domain-Specific Modeling Languages,” in EXE 2015 . Ottawa, Canada: CEUR, 2015 .

[3] F. Latombe , X. Cre´gut, B. Combemale , J. Deantoni , and M. Pantel , “ Weaving Concurrency in eXecutable Domain-Specific Modeling Languages,” in 8th ACM SIGPLAN International Conference on Software Language Engineering (SLE 2015 ), 2015 . [Online]. Available: https://hal.inria.fr/hal-01185911

[4] T. Murata , “ Petri nets: Properties, analysis and applications , ” Proceedings of the IEEE , vol. 77 , no. 4 , 1989 .

[5] G. A. Agha , “ Actors: A model of concurrent computation in distributed systems . ” DTIC Document, Tech. Rep. , 1985 .

[6] G. Winskel , “ Event structures,” in Petri Nets: Applications and Relationships to Other Models of Concurrency, ser . LNCS , 1987 .

[7] V. Pratt , “ Modeling concurrency with partial orders ,” International Journal of Parallel Programming , vol. 15 , no. 1 , pp. 33 - 71 , 1986 .

[8] J. Deantoni and F. Mallet , “ ECL: the event constraint language, an extension of OCL with events , ” Inria, Tech. Rep. , 2012 .

[9] OMG, “ fUML specification v1.1 ,” 2013 . [Online]. Available: http: //www.omg.org/spec/FUML/

[10] G. D. Plotkin , “ The origins of Structural Operational Semantics,” The Journal of Logic and Algebraic Programming , 2004 .

[11] S. Tasharofi , P. Dinges , and R. E. Johnson, “ Why do scala developers mix the actor model with other concurrency models? ” in ECOOP 2013 . Springer, 2013 .

[12] F. Mallet and R. De Simone , “Correctness Issues on MARTE/CCSL constraints,” Science of Computer Programming , vol. 106 , pp. 78 - 92 , Aug. 2015 . [Online]. Available: https://hal.inria.fr/hal-01257978

[13] J. Armstrong , R. Virding , C. Wikstro¨m, and M. Williams , Concurrent programming in ERLANG . Citeseer, 1993 .

[14] M. Gupta , Akka essentials . Packt Publishing Ltd , 2012 .

[15] DIVERSE-team, “Github for k3al ,” 2016 . [Online]. Available: http://github.com/diverse-project/k3/

[16] J. Deantoni , P. Issa Diallo , C. Teodorov , J. Champeau , and B. Combemale , “ Towards a Meta-Language for the Concurrency Concern in DSLs ,” in DATE, 2015 .

[17] OMG, “QVT specification v1 .2,” 2015 . [Online]. Available: http: //www.omg.org/spec/QVT/

[18] Language Workbenches Challenge , “ Comparing tools of the trade ,” 2014 . [Online]. Available: http://www.languageworkbenches.net/

[19] M. Voelter and V. Pech , “ Language modularity with the MPS language workbench,” in ICSE . IEEE, 2012 .

[20] S. Kelly , K. Lyytinen , M. Rossi , and J. P. Tolvanen , “MetaEdit+ at the age of 20,” in CAiSE. Springer, 2013 .

[21] T. Van Der Storm , “The Rascal Language Workbench,” 2011 .

[22] L. C. Kats and E. Visser, “ The spoofax language workbench: rules for declarative specification of languages and ides ,” in ACM Sigplan Notices, 2010 .

[23] T. Mayerhofer , P. Langer , M. Wimmer , and G. Kappel, “xMOF: Executable DSMLs based on fUML,” in SLE, 2013 .

[24] G. Rosu and T. F. Serbanuta , “K overview and simple case study,” in Proceedings of International K Workshop (K'11) , 2014 .

[25] M. Nielsen , “ Models for concurrency,” in Mathematical Foundations of Computer Science , 1991 .

[26] E. Lee , A. Sangiovanni-Vincentelli et al., “A framework for comparing models of computation,” Computer-Aided Design of Integrated Circuits and Systems , IEEE Transactions on, 1998 .

[27] M. Nielsen , V. Sassone , and G. Winskel, Relationships between models of concurrency . Springer, 1994 .

[28] C. Ptolemaeus , System Design, Modeling, and Simulation: Using Ptolemy II. Ptolemy . org Berkeley, CA, USA, 2014 .

[29] B. Selic , “ An architectural pattern for real-time control software ,” in Workshop on Frameworks and Architectures, PLoP Conference, 1996 .

[30] F. Zalila , X. Cre´gut, and M. Pantel, “A transformation-driven approach to automate feedback verification results,” in Model and Data Engineering . Springer, 2013 , pp. 266 - 277 .