It seems to me that refactoring/rewriting only gets you closer to a local optimum, which could still be far away from the solution. As in gradient descent in machine learning, one circulates around a small sinkhole, just building up frustration rather than gaining insight.
I recently failed hard to extend some Ruby code that performed a lot of in-place modifications on a huge data structure. I re-approached the task after a couple of months by reading the code, explaining the code in English in a separate file, and then built up new code by reading both my English comments for the domain knowledge and the original code for the technical details. I got farther in a week than previously in one month. However, this was just a problem of expressing the same solution in a more understandable and better extensible way. (The resulting code was way less object-oriented, by the way.)
Even though my problem is only on the expression side, and the Sudoku problem clearly was on the domain side, stepping outside the sinkhole was absolutely necessary in both cases, for it only lead to a local optimum.
It seems to me that refactoring/rewriting only gets you closer to a local optimum, which could still be far away from the solution. As in gradient descent in machine learning, one circulates around a small sinkhole, just building up frustration rather than gaining insight.
I recently failed hard to extend some Ruby code that performed a lot of in-place modifications on a huge data structure. I re-approached the task after a couple of months by reading the code, explaining the code in English in a separate file, and then built up new code by reading both my English comments for the domain knowledge and the original code for the technical details. I got farther in a week than previously in one month. However, this was just a problem of expressing the same solution in a more understandable and better extensible way. (The resulting code was way less object-oriented, by the way.)
Even though my problem is only on the expression side, and the Sudoku problem clearly was on the domain side, stepping outside the sinkhole was absolutely necessary in both cases, for it only lead to a local optimum.