2 comments

[ 32.1 ms ] story [ 766 ms ] thread
It's an interesting idea, but seems unfeasible. I assume the editor will be shipped in a binary, which contains the trained models if machine learning is used in it, but it would still take a fair bit of computation power and memory to run. Furthermore it seems like overkill. I can express a concept much clearer in code than I can in comments. Comments are there to help me get back into the code if I've stepped away from it.

I think that writing function skeleton and automatically generating test, and vice versa, however could be feasible and it should be possible to do. The decision of which skeleton/test to generate could be done with a very inexpensive statistical computation or a simple machine learning model. The outline of the project seems far too large to be feasible and I think some of it would hinder development rather than ease it, but the test/function generation seems useful.

> I can express a concept much clearer in code than I can in comments.

I doubt it. Just from your one post, I can see that you have a perfectly adequate ability to use human language, and human language is much more powerful and expressive than any programming language.

If you disagree, you should make your counter-argument in a programming language.