Report - URN Focus Group Meeting
List of attendees
- D. Amyot (
- U. Glässer (U. of
- N. Mansurov (KLOCwork
Solutions,
- A. Metzger (U. of
- R.L. Probert (U. of
- R. Reed (
Invited:
- J. Poncela (U. of Málaga, Spain)
The agenda proposed by D. Amyot was approved. U. Glässer and A. Metzger officially joined the URN Focus Group.
An overview of URN, GRL, and UCM was given by D. Amyot in the morning during the 3rd SDL and MSC Workshop (presentation). Some of the main concepts and goals were recalled at the meeting.
The core of the discussion was on the notation itself, on modularity,
and on patterns. D. Amyot gave a demonstration of the OME tool (developed at
the
- We seem to need a cookbook on when to use softgoals versus regular goals. This is a question that comes over and over again. Do we really need this distinction in the standard?
- GRL seems to be used mainly in a top-down way (by decomposing higher-level goals into sub-goals). However, partly because of a lack of modularity and hierarchical decomposition of GRL models, we can get models that contain elements of a fairly low level of abstraction (e.g. almost operational). It is good to show that the language supports this transition from abstract to concrete, but both levels may not always belong to the same model.
- N. Mansurov presented a GRL model from one of his papers. This model presents goals and criteria for the selection of a requirements capturing notation. He shares his experience with the OME tool and its usability. He argued that having everything in one flat diagram has also its benefits, and that if the model does not fit in one diagram, then perhaps this is a symptom of a model that is too complex. However, sub-diagrams would enable a better scalability of the models.
- Instead of hierarchical decomposition of goals into sub-diagrams, it was suggested to hide parts of the diagram that are not relevant to a given context. For instance, we could hide the nodes that are not connected to a specific node in the graph (actors do that to some extent, but this could be generalized to other constructs). This however seems to be a tool issue rather than a language issue (yet this will be proposed as a suggestion to the OME developers). A related language issue would be the need for more formal views of a model that could be user-defined.
-
In GRL, goals/softgoals/tasks can be refined by
decomposition. Do we need the ability to refine
dependencies as well?
- A. Metzger suggested the use of typed patterns as a way to support modularity in GRL models. Right now, we have flat models where we can import GRL patterns from a catalogue (a library) through a tool (OME), with no specific language support. For instance, the knowledge related to “security” can be captures as a GRL graph that can be tailored to a particular context and inserted in a larger model. However, a change to whatever is imported does not modify the catalogue or the other instantiations used in the model, causing a lack of consistency in large models. An alternative to this would be to keep a reference to the sub-model imported (the pattern) and make modifications at that level. N. Mansurov mentioned that since most GRL models are rather small, there might not be enough benefits to justify this additional complexity.
-
It was also observed that many NFRs are not
always localizable to one actor (e.g. performance) but may span multiple ones.
Is the dependency/relationship mechanism that GRL has sufficient to capture
this?
-
Is there a way to compute/analyze indirect
contributions based on transitivity?
D. Amyot gave a demonstration of the UCMNav tool (developed mainly
at
-
In the elevator example that was presented,
there was some discussion about the use of start points as preconditions when
they are triggered by other paths. Many felt this was confusing and that variables
in the data model (associated to the first start point in the scenario) are
probably more representative.
-
Bindings
of plug-in start/end points to stub input/output segments could benefit from
being shown visually.
-
In another meeting, it was also mentioned that requirements
that are functions (e.g. the threshold is f(a,
b, c)) cannot really be expressed with UCM (or GRL). UCMs really abstract
from the nature of these functions and really focus on the scenarios where such
requirements are met or not. In a sense, the functions are reduced to logical
expressions (could be as simple as one Boolean variable) used in conditions and
other locations in UCMs. The nature of these requirements needs to be described
elsewhere.
D. Amyot gave an overview of the content of the Z.150
document, to be sent for consent at the next SG 17 meeting in November.
-
A. Metzger raised the point that any notation
satisfying this set of requirements would be Z.150 compliant, and hence could
be called URN. Indeed, the process is still open to suggestions for concrete
notations other than GRL and UCM. However, to be Z.150 compliant, a notation
should be an ITU-T standard itself.
-
R. Reed mentioned that most of the referenced standards
and recommendations should not be in section 2 and should be moved to a
bibliography in some appendix. What is cited in section 2 really becomes part
of the URN standard. Z.151-153 should probably reference Z.150 and each other.
-
There was some discussion on the use of the
terminology in section 3 (Definitions). These should be in line with other SG
17 definitions or reuse existing ones as much as possible.
- We should also identify what requirements should be addressed in this study period and what can be left to the next study period. We do not have to deliver everything all at once, especially as more advanced concepts require more thinking and experiments.
-
There was a discussion about the definition of
syntax and semantics for the languages. XML is mainly seen as an interchange
format. There is still no decision on how to go about GRL and UCMs: abstract
grammars or metamodels.
-
U. Glässer, who worked on the ASM semantics for
SDL, and D. Amyot will collaborate to work on an ASM semantics for UCM/URN,
with hopefully some results to be presented at the next SDL Forum conference.
The meeting ended at
-
Eckhardt Holz (Institut für Informatik, Humboldt-Universität
zu
-
Peter
Graubmann presented a paper on MSC connectors
which could perhaps be combined to the MSC generation from UCMs in order to
produce more meaningful and complete MSCs. Paul Bristow (Motorola) also
presented a paper on test generation from MSCs where macros were used to expand
messages to more detailed ones. This might be of interest for Z.153.
Daniel Amyot,
Rapporteur Q18/17