I feel the approach could have some merit, but the article unfortunately stops short of demonstrating it. Also the problem of analytically finding similar code

I feel the approach could have some merit, but the article unfortunately stops short of demonstrating it. Also the problem of analytically finding similar code sub-trees seems highly non-trivial. As baseline you want to parse the code, but really you'd also want to somehow "normalize" identifiers, as usually it is the common structure rather than specific identifiers that matter.