3 ms·
What. A lovely article and some of the comments show that its context isn’t understood (despite the author’s efforts to imply it!) I’m going to start by ignori
by gumby 6y ago
What. A lovely article and some of the comments show that its context isn’t understood (despite the author’s efforts to imply it!)
I’m going to start by ignoring this in my code. Just make it work correctly, right? But as we get closer to deployment performance probably matters (it does to us). Then we start to notice lots of important coupling. For example input validation typically* isn’t required inside the system: datastructures and generated and used by our code. Validation applies at the “surface” or external inputs of the system. Log records which are generated by one part of the system may need to be dredged rapidly — forcing the representation to be fixed means the code that parses it can make assumptions the library code cannot. Metering will tell you where optimization opportunities lie.
* super long-lived applications sometimes need to do consistency checks even on internal data to avoid subtle corruption issues that could have hardware or software origin and could be very expensive to encounter. Things like phone switches and spacecraft that need to go years or even decades without a reboot.