I just read an article calling for a way to get an AST for Ruby code from within Ruby. I think it’s a great idea, and I think static analysis, code manipulation/refactoring are great reasons to support such a thing. Another consequence would be macros, which are a more controversial feature. The only downside I can think of is that it could be misused, and to leave out a feature for that reason would go against the Ruby philosophy, in my opinion. One of the regulars at the Phoenix Ruby Users’ Group argued against using a dumbed-down “teaching language”, saying that when you suggest using a language you don’t want to use yourself, you’re talking down to the learner. I agree. I think the best way to keep people from misusing features is not to remove the features, but to educate them. People in the Ruby community have always done a good job of making sure there’s lots of good example code out there, and giving constructive criticism when someone posts a bad example.
While the reasons in the article are enough, I can think of another good reason for having AST support — having the ability to constrain code to a certain set of features. There would be two different uses for this:
- Running code from an untrusted source – This could include web template designers, or even users. Ning is an example of a site that lets users run their own code — but there is a huge overhead to facilitate this and to sandbox everything. If it could be verified that code doesn’t do anything dangerous, I think a new type of Web app plug-ins would emerge. Instead of having to set up an app on a separate site and use REST or SOAP to communicate, people could just throw together a little script in a domain-specific language. One idea I have for this sort of technology would be a Flickr for html and image generators (like ajaxload.info).
- Enforcing decoupling – When managing a large software project, I think it would nice to specify which classes do what, and have it enforced. Maybe there is some class that’s supposed to be all about math, but that goes in an application that gives output to users. To keep math programmers from putting presentation logic in mathematical code, you could constrain them to a DSL that doesn’t have strings. Or, you could keep template programmers from doing networking code. The check could be done at runtime, or commit time, with plug-ins to the version control system.
There is a Perl library that parses Perl, but it would be nice to have one for Ruby that’s written in C and optimized for speed, and that with certainty matches up to what the interpreter understands.
I think that once an AST implementation was built, something like RSpec could be created which does compile-time or commit-time analysis. That would be cool.