First presented at CALECO
Previous: Analysis of existing spreadsheets

Progress and future work

We have a prototype compiler, written in Java, which generates LotusScript code for Lotus 1-2-3, and VisualBasic code for Excel. A demonstration, as an applet, can be found at http://users.ox.ac.uk/~popx/mm_demo.html . This demonstration uses a Java applet. For those whose browsers cannot run Java, or for whom it would be slow to download, there are screen shots in a slide-style presentation at http://users.ox.ac.uk/~popx/mm_presentation/title.html .

The research reported here has not been funded at all, so is spare-time only. Further work depends on finding funds. Our research goals include:

• Looking at how people use spreadsheets. The current version of MM is suited to models where time is irrelevant (e.g. in calculating the cost of paint needed to paint a number of rooms, given their dimensions and material costs), and to ``closed-form'' models, where we simulate the progress of attributes through time, each attribute depending on its own past values, or on present and past values of other attributes.

  Going in the direction of added modelling complexity, some users need to build ``non-closed-form'' models, where to get results, it is necessary to solve a system of equations. This normally requires explicit programming of an equation solver. We believe MM could be extended so that its programmer could just write the equations, in a similar style to above, and then name an algorithm to be used for solving them. The MM compiler would then look this up in a library and ``fold'' it into the generated code. This would save programming effort and, again, make programs more modular and easier to read and maintain.

• Building a drag-and-drop interface for specifying and connecting objects graphically. There are already some programs that do this for spreadsheets, for example CleanSheet from the UK company Working Title [Pountain 1994,Working Title: Home Page]. However, they do so in an ad-hoc fashion. The use of Goguen's sheaf semantics as a foundation would guarantee the correctness and appropriateness of any visual representation of a model. It is possible that Goguen's work on algebraic semiotics could help here [Goguen 1993,Goguen 1996,Goguen 1997].

• Building a reverse-engineering tool that will ``decompile'' existing worksheets into MM source code. It should be fairly easy to do this by using Inductive Logic Programming, for example in Stephen Muggleton's Progol.

• Semantics: how best to describe plural objects, e.g. a company serving clients who vary in number during a run; parameterisation and modularisation.

First presented at CALECO
Previous: Analysis of existing spreadsheets