After porting the code to 2.x, which proved to be a somewhat tricky process, the code was committed in the 2to3 sandbox repository. This concludes the project on my part, although I will still be around to provide support for the new code, at least until it reaches a sufficient point of maturity.
The project was overall a great experience for me; the proposal was challenging on both the algorithmic and coding part. What I've really come to realize was how much the theoretical description of a problem can differ from it's practical applications. Matching tree regular expressions in an efficient way in 2to3's context went well beyond what the research articles I had in mind at first offered, and implementing a new algorithm that can retain it's soundness while optimizing it was a quite stressful, but rewarding, experience. Most of the effort was spent on preserving compatibility with the previous matcher; the fixers were made to match and transform code in assumption of a pre/post-order traversal of the AST and various issues had to be handled on a case by case basis.
Although the benefits proved substantial, the effort taken and the complexity of the process often made me think if the whole concept is worth a few machine cycles; admittedly, as it is also the case with 2to3 itself, optimizations suffer from a low impresiveness/complexity ratio that tends to drive people away from involvement and involved people to insanity. But in conclusion, I'm proud(and, to be honest, still find it hard to believe) that this thing actually worked.
No comments:
Post a Comment