John Duimovich
October 21, 2005
Position
I am interested in making it simpler to create new programmming language IDEs for the Eclipse platform. I am motivated by my work on a Smalltalk IDE for Eclipse in which I found I was writing too much code for the functionality I was getting in return. I found that was that much of the code ends up being very similiar to code found in JDT and CDT. In fact, it has become a common joke that the way to get a new IDE for eclipse, is to take code from JDT and replace the "J" with your language letter in the class name (I used ST). I have seen many initial ports of new languages done this way and as a quick and dirty way to get started it's a pretty good way to go, especially if you borrow ideas (and code) from a well architected piece of code like JDT. Of course, the drawback to this technique is that you "bake in" the bugs you copied from the original code and you don't get the fixes as they are made in the original. You also miss new innovations as they appear in JDT and the underlying platform. Given that many of these components end up being quite similiar I feel it should be possible to create generic components for creating language IDEs, possibly starting from existing components from the existing language IDEs and generalizing them to allow re-use in other contexts. Clearly over time as you re-use components from one language in another, you will create common components as the problem accross many languages becomes better understood.So, this suggests a path forward. As IDEs for different languages are built, there will be opportunities to harvest components which will be applicable to different languages.
A non-exhaustive list of tools (and problems to solve) may look something like this (I make no claims on how hard or easy the specific items are).
Once we've identified these common components, the next steps would be the "IDE Wizard" tools which help you generate "your first" IDE. What I really would like to see is the Eclipse IDE equivalent of "glasses in about an hour" in which much of the the first implementation of an IDE is generated for you. You can then tailor the generated components for the specific needs or wholesale replace them with custom versions which provide tighter integrations. The reaction I get when talking to most people about this type of tool for creating IDEs is typically divided into two distinct groups. There's the when can I have it? people and the that's impossible! people. I understand the motivation of the first group, they want a simpler way to build IDEs. The latter group, is clearly thinking that we intend to auto generate JDT level functionality which is a pretty high bar or if you don't do a JDT then it won't be usefull. I don't think you're going to generate a JDT class IDE for "free" but I think you may be able to generate a usefull first cut. The tricky part is defining what "usefull" means in this context or even describing what you could expect from a generated IDE. One way, is to look at the level of functionality that an IDE has and decide what kind of functionality you could expect to find in a generated IDE.
Levels of integration in an Eclipse based IDE
There are many possible levels of integration with Eclipse. The simplest integration is to use Eclipse as your edit, compile, run (and possibly debug)system. The next level is to create a program model (parse the source code) so that you can implement language aware functionality. As the integration becomes more advanced, the opportunities become richer and features like program manipulation (refactoring, code generation, and full round trip editing for code generation systems that use programming languages as their intermediate language) become possible
For simplicity, I decided to describe IDEs in terms of levels of integration with eclipse.