There is a place on the long line between max-performance and max-convenience that a lot of projects can build their house on but simply do not because they do

There is a place on the long line between max-performance and max-convenience that a lot of projects can build their house on but simply do not because they do not care. Some of my projects are heavily dependant on parsing strings when the data is initially consumed. I could speed them up by shifting to more efficient parsing/replacement algorithms. But that would make the code too cryptic. So I am ready to give up a lot of performance for ease of understanding because this is a rare enough activity to not matter in the larger scheme of things. But this would be a huge headache for the users if it happened on the critical path. How many projects are mindful of these things, you think? As for the transitive dependencies problem that Pike refers to, this is why I try to avoid anything outside the standard library for interpreted languages like python. cpython is already terrible performance-wise without forcing it to deal with a million lines of additional code from random people.